Power Electronics Essentials and Applications

Chapter 07.

indd 421 3/13/2009 7:08:49 PM

Chapter 07.indd 421 3/13/2009 7:08:49 PM
Power Electronics
Essentials and Applications
Power Electronics
L. Umanand
Centre for Electronic Design & Technology
Indian Institute of Science
Bangalore 560 012
Power Electronics
Copyright © 2009 by Wiley India Pvt. Ltd., 4435-36/7, Ansari Road, Daryaganj, New Delhi-110002.

All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any
form or by any means, electronic, mechanical, photocopying, recording or scanning without the written
permission of the publisher.
Limits of Liability: While the publisher and the author have used their best efforts in preparing this book, Wiley
and the author make no representation or warranties with respect to the accuracy or completeness of the
contents of this book, and specifically disclaim any implied warranties of merchantability or fitness for any
particular purpose. There are no warranties which extend beyond the descriptions contained in this paragraph.
No warranty may be created or extended by sales representatives or written sales materials. The accuracy and
completeness of the information provided herein and the opinions stated herein are not guaranteed or warranted
to produce any particular results, and the advice and strategies contained herein may not be suitable for every
individual. Neither Wiley India nor the author shall be liable for any loss of profit or any other commercial
damages, including but not limited to special, incidental, consequential, or other damages.
Disclaimer: The contents of this book have been checked for accuracy. Since deviations cannot be precluded
entirely, Wiley or its author cannot guarantee full agreement. As the book is intended for educational purpose,
Wiley or its author shall not be responsible for any errors, omissions or damages arising out of the use of the
information contained in the book. This publication is designed to provide accurate and authoritative
information with regard to the subject matter covered. It is sold on the understanding that the Publisher is not
engaged in rendering professional services.
Trademarks: All brand names and product names used in this book are trademarks, registered trademarks, or
trade names of their respective holders. Wiley is not associated with any product or vendor mentioned in this
book.
Other Wiley Editorial Offices:

John Wiley & Sons, Inc. 111 River Street, Hoboken, NJ 07030, USA
Wiley-VCH Verlag GmbH, Pappellaee 3, D-69469 Weinheim, Germany
John Wiley & Sons Australia Ltd, 42 McDougall Street, Milton, Queensland 4064, Australia
John Wiley & Sons (Asia) Pte Ltd, 2 Clementi Loop #02-01, Jin Xing Distripark, Singapore 129809
John Wiley & Sons Canada Ltd, 22 Worcester Road, Etobicoke, Ontario, Canada, M9W 1L1
First Edition: 2009

ISBN: 978-81-265-1945-3
ISBN: 978-81-265-8016-3 (ebk)
www.wileyindia.com
Printed at:
Foreword
T he book that you have in your hands is the outcome of years of patient but enthusiastic learning and
exploration of the field of power electronics by its author. You may feel a bit apprehensive in coming to
terms with more than eight hundred pages on this topic; however, a closer look at the table of contents will
reveal the comprehensive coverage of the subject, explored from many angles and with various tools. This
book, indeed, reflects well the professional trajectory of its author. Having known Dr L. Umanand from his
student days and later as a colleague at the Centre for Electronics Design and Technology, I had the pleasure
to interact with him during his entire professional career and share his passion for power electronics. I can
clearly recognize in this book that same passion that he progressively developed for various aspects of the
subject.
He has taken up the challenge to present here a comprehensive synthesis of the field with the required
theoretical approach, combined with the pragmatism that comes from experience accumulated through
years of direct involvement in design, development and testing of the power electronics systems. This,
together with the originality of certain topics like Bond graphs, Design of Magnetism or Design for
Reliability, will surely make this book stand out.
The book reflects equally well the deep interest the author has taken in the last few years of the “teaching/
learning” process. The learning objectives of each chapter are clearly stated and the material is illustrated
with numerous examples. The reader is also challenged throughout the book with numerous questions,
problems, laboratory assignments, that, if carried out sincerely, should ensure proper anchoring of the newly
acquired knowledge.
In spite of the limitations of any written material, I wish that through this book Dr Umanand will be
able to inspire you in sharing his passion and enthusiasm for the subject as he does so well with his students,
even if you miss the immediate proximity of his broad smile and great sense of humor.
Andre Pittet
Chief Project Advisor
CEDT, Indian Institute of Science
Bangalore
PE_Preface.indd v 3/24/2009 7:20:16 PM

PE_Preface.indd vi 3/24/2009 7:20:16 PM
Preface
T he subject of power electronics has been treated in numerous literatures with various viewpoints and
styles. At the outset, it may appear as though one more book is being introduced into the market,
which may at most be a marginal update on the existing literature with the same topics that are addressed in
other books. Likewise, I would like to reiterate that the topics and concepts covered in this book are not new
as one would expect, considering the field of power electronics is rather mature. Notwithstanding the
maturity of the field, I must add that the topics in this book are neither old wine in new bottle nor a
marginal update. Power electronics that initially started its career in processing power revolved primarily
around devices like the vacuum tubes, mercury arc rectifiers and, later on, the thyristors. Power electronics
has come a long way since then. Technology has improved by leaps and bounds making the power devices
more closely to an ideal switch, and the control of the switches are performed in the discrete domain with
complex control structures. Thus, power electronics now spans a very wide knowledge base such as power
devices, drives, circuit topologies, magnetics, system modeling, control configurations, digital processing,
thermal and reliability aspects. If all these aspects are to be treated in a single book, then the topics will, in
general, have a superficial bearing. If the topics are to be treated in depth, then it becomes difficult to
accommodate all these topics into a single book. Here in lies the challenge and hence this book.
All the topics discussed in this book have been handed down from my teachers and teachers before them
by way of exhortation; and to them, I offer my sincere salutations. In the run up to this book, I must say
that I have been indeed very lucky to have been tutored by some of the best minds in the field. Every topic
has a flavor of the teaching styles of my mentors. From the confluence of the teaching styles of my teachers,
I hope that a style will emerge that I may call my own. If the topics are well treated and addressed, then the
credit, I must say, should go to my mentors; and if there are mistakes which I presume there will be some, I
am solely responsible and will make efforts to correct them in future editions.
The book has been broadly divided into two types of topics: (a) circuit-oriented aspects and (b) system-
oriented aspects. The first seven chapters deal with circuit-oriented aspects of power electronic systems, and
Chapters 8–15 deal with system-oriented aspects like controls and reliability. Chapter 1 discusses the power
semiconductor switches which are the main building blocks of the power electronic systems. The treatment
of the semiconductor switches addresses static characteristics, dynamic characteristics, losses and modeling
issues. Chapter 2 discusses the drive circuit requirements of BJTs and MOSFETs. Typical drive circuits and
their designs are dealt in this chapter. This chapter also introduces the concepts of the series and shunt snubber
circuits for power semiconductor devices that behave as power switches.
Chapters 3–6 deal with the power electronic circuit applications. Chapter 3 is on the topic of AC–DC
converters or rectifiers. This chapter discusses single- and three-phase rectification topologies along with a
detailed discussion on the popular rectifier–capacitor filter circuit addressing its design aspects. The
rectifier–LC filter circuit is also discussed. This is followed by a discussion on controlled rectification. Chapter 4
does not fall into the switched-mode category. This chapter addresses the linear regulators wherein the
power semiconductor devices are operated in the linear region. Apart from design, the analysis of the linear
regulators by progressively including non-idealities is dealt with in a systematic manner.
PE_Preface.indd vii 3/24/2009 7:20:16 PM

Preface viii
Chapter 5 discusses the DC–DC converter application. The primary non-isolated topologies of the
DC–DC converters are discussed, followed by isolated converters and other special converters. The discus-
sions in this chapter are based on the steady-state analysis that primarily addresses the component design
issues. Chapter 6 handles the DC–AC converters or inverters. The various generic topologies are explained
followed by a detailed Fourier series analysis of pulse-width modulation strategies for both single- and three-
phase inverters. Chapter 7 is a generic topic on magnetic devices that is needed for all applications in AC–DC,
DC–DC, DC–AC conversions and isolated drive circuits. This chapter discusses the general principles of
magnetism applied to design of inductors, potential transformers and current transformers.
Chapters 8–15 discuss system-oriented aspects of power electronics. Chapter 8 discusses the different
modeling methods and their applications to power electronic circuits and systems in order to obtain a
dynamic model. Special emphasis is placed on the circuit averaging method, bond graph method and the
space vector methods, including detailed discussions with application examples. Chapter 9 is not strictly
related to power electronics, but the topics discussed here are essential for designing controllers. It is a cock-
tail of many topics such as z-transform basics, digital filters, sampling, analog-to-digital conversion methods
and performance specification issues.
Chapter 10 develops a formal and systematic approach towards design of controllers for the power elec-
tronic systems, wherein the dynamic models are obtained from the concepts of Chapter 8. Both the classical
methods of controller design and the state space methods of controller and estimator design are addressed
with more emphasis on digital controller and estimator design. Chapter 11 is the extension of the state space
controller and estimator design with focus on optimality and robustness issues.
Chapter 12 discusses an important implementation aspect which is discrete computation methods. As
most of the control implementations of the power electronic systems are in the digital domain, a detailed
discussion on the numeric formats and practical arithmetic algorithms are addressed. This chapter also dis-
cusses the implementation of important components like the PI controller and pulse-width modulators
within a digital processor. Chapter 13 discusses yet another important and practical aspect of power elec-
tronic system which is the thermal aspect. This discusses the heat transfer mechanisms for conducting the
heat away from the junction of the power semiconductor devices to the ambient. The selection of heat sinks
for the power electronic applications is addressed in this chapter.
Chapters 14 and 15 are devoted to reliability aspects. Chapter 14 discusses in detail the concepts of
modeling systems from the reliability point of view. Chapter 15 presents the methods for predicting the reli-
ability of a system. A formal and systematic method to predict the reliability of circuits by part stress co-variate
approach is presented by integrating the functional and life aspects of the circuit specifications at the design
stage itself. A MATLAB based toolbox called reliability for electronic circuits (REC) is included in the
accompanying CD. This toolbox is developed based on the concepts presented in Chapter 15.
As is evident from the topics and the chapters, the vast area of power electronics cannot be handled in a
single semester course of the engineering curriculum. In order to handle the topics for semester duration,
the chapters may be re-organized in the following manner that is only suggestive and also not exhaustive.
• Parts of Chapters 1, 3, 4 and 7 are a possible combination for a course on linear power supplies.
• Parts of Chapters 1–3, 5 and 7 are a possible combination for a course on switched-mode DC–DC
power supplies.
• Chapters 5 and 7 can be suitable for a course on DC–DC converters.
• Parts of Chapters 1–3, 6 and 7 are a possible combination for a course on inverters.
• Chapters 6 and 7 can be suitable for a course on inverters along with pulse-width modulation
strategies.
PE_Preface.indd viii 3/24/2009 7:20:16 PM

Preface ix
• Chapters 5, 8–10 can be useful for an advanced-level course on dynamics of DC–DC converters.
• Chapters 8–11 by themselves can be suitable for an advanced-level course on control of power elec-
tronic systems.
• Chapters 5, 7, 12 and 13 can be a possible combination for a course on implementation aspects of
DC–DC converters.
• Chapters 6, 7, 12 and 13 can be a possible combination for a course on implementation aspects of
inverters.
• Chapters 14 and 15 by themselves can be useful for a course on reliability of power electronic systems.
• Parts of Chapters 5, 7, 13–15 can be suitable for a course on reliability design for DC–DC converters.
• Parts of Chapters 6, 7, 13–15 can be suitable for a course on reliability design for inverters.
I must acknowledge that though it appears that I am the sole author of this book, the material in the book
has evolved over many years with constant interaction with my teachers, colleagues and students who have
directly and indirectly contributed to the knowledge base of the book; my sincere salutations to all of them.
For the past year and a half, I have not given sufficient time to my family and the matters of the home
due to the long and late hours spent in writing this book. My wife and son have been very patient and sup-
portive in this aspect, awaiting the time for the completion of the book. I sincerely acknowledge my gratitude
for their continuous support.
L. Umanand
PE_Preface.indd ix 3/24/2009 7:20:16 PM

PE_Preface.indd x 3/24/2009 7:20:16 PM
Contents
Foreword v
Preface vii
1 Power Semiconductor Switches 1

1.1 Ideal Switch 2
1.2 Diodes 2
Static Characteristics 6
Dynamic Characteristics 7
Diode Classifications 8
Diode Parameters 8
1.3 Surge Current 9
1.4 Thermal Viewpoint 9
Diode Model 10
1.5 Bipolar Power Transistors 11
Calculation of Power Dissipation in a Transistor 17
1.6 Safe Operating Area (SOAR) 19
1.7 Paralleling of Transistors 20
1.8 Darlington Connection 21
1.9 Power MOSFETs 22
Paralleling of MOSFETS 29
Handling Precautions for MOSFET 29
Remarks on BJT and MOSFET 29
Power Dissipated in a Power MOSFET 30
Model for Power MOSFET 30
1.10 Insulated Gate Bipolar Transistor (IGBT) 30
Remarks on IGBT 37
1.11 Thyristor Family 37
Silicon-Controlled Rectifier (SCR) 38
DIACs and TRIACs 42
Gate Turn-OFF Thyristor (GTO) 44
MOS-Controlled Thyristor (MCT) 46
Concluding Remarks 46
Laboratory Exercises 47
Fill in the Blanks 51
PE_TOC.indd xi 3/24/2009 7:51:18 PM

xii Contents
Descriptive Questions 52
Problems 53
Answers 53
2 Drive Circuits 55
2.1 Transistor Drive Circuits 55
Turn-ON Behavior 56
Turn-OFF Behavior 56
Characteristics and Classification of the Drive Circuits 58
BJT Drive Circuit-1 59
2.2 MOSFET Drive Circuits 70
MOSFET Drive Circuit-1 71
2.3 Snubber Circuits 80
Turn-OFF Snubber or Shunt Snubber 80
Turn-ON Snubber or Series Snubber 83
Problems 93
Answers 94
PE_TOC.indd xii 3/24/2009 7:51:18 PM

Contents xiii
3 Rectifiers 95
3.1 Uncontrolled Rectifiers 96
3.2 Rectifier Circuits 96
Single-Phase Circuits 96
Three-Phase Circuits 103
3.3 Capacitor Input Filter 109
Design of Capacitor Input Filter Rectifier 110
Turn-ON Currents and Surge Limiting 117
3.4 Power Factor 121
3.5 Rectifier–LC Filter 124
Output Ripple 127
Turn-ON Current 128
Design Summary 128
3.6 Controlled Rectifiers 128
Single-Phase Power Circuits 129
Three-Phase-Controlled Rectifier Circuits 134
Problems 147
Answers 148
4 DC–DC Linear Regulators 149

4.1 Generic Linear Regulator 149
Operating Principle 150
4.2 Linear Regulator Topologies 151
Shunt Regulator 151
Modified Shunt Regulator 154
Series Regulator 154
Negative Voltage Regulator 156
IC Linear Regulators 158
Practical Tips 160
Four-Wire Connection 162
Protection 163
Current Regulation 165
4.3 Parameters of the Linear Regulator 166
4.4 Analysis of a Series Regulator 166
First-Level Modeling 166
Effect of Current through Rb 170
Effect of Current through R1 171
Regulator Parameters 172
Range of Load Resistance RL 173
PE_TOC.indd xiii 3/24/2009 7:51:18 PM

xiv Contents
Second-Level Modeling 174

Influence of Temperature 179
4.5 Analysis of a Current Regulator 180
Line Regulation 182
Range of RL 182
Second-Level Modeling 183
Problems 189
Answers 192
5 DC–DC Switched-Mode Converters 193

5.1 Single Pole Double Throw Switch 193
5.2 The Chopper 194
5.3 DC Steady-State Principles 200
5.4 The Three Basic Converters 201
Step-Down or Buck Converter 201
Step-Up or Boost Converter 206
Step-Up/Down or Buck–Boost Converter 213
5.5 Selection of Power Switches 219
Device Selection 219
Electrical Stress Rating 221
Thermal Stress Rating 221
5.6 Continuous and Discontinuous Conduction Modes 221
5.7 Isolated Converters 225
Forward Converter 226
Semiconductor Switches 228
Waveforms 229
Governing Equations 230
Forward Converter with Demagnetizing Winding 233
Dual Switch Forward Converter 236
Push–Pull Converter 238
Half-Bridge Converter 243
Full-Bridge Converter 248
Flyback Converter 253
5.8 Practical Issues 258
5.9 Other Topologies 258
Cuk Converter 259
High-Boost Converter 263
Soft Switching in Converters 265
PE_TOC.indd xiv 3/24/2009 7:51:18 PM

Contents xv

Problems 286
Answers 288
6 DC–AC Switched-Mode Converters 289

6.1 Inverter Topologies 290
Single-Phase Topologies 290
Three-Phase Topologies 296
6.2 Self-Driven Inverters 297
Saturable-Core Inverter 298
Saturable Base Drive Inverter 302
6.3 Driven Inverter 304
Push–Pull Configuration 304
Half-Bridge Configuration 306
Full-Bridge Configuration 308
6.4 Quasi-Square Wave Inverter 309
6.5 Three-Phase Inverter Topologies 312
Six-Step Inverter 314
Current-Controlled Inverter 315
6.6 Pulse-Width Modulation 318
Quasi-Square Modulation 321
End Pulse Modulation 324
Center Pulse Modulation 327
Sinusoidal Pulse-Width Modulation 329
PWM through Phase Modulation 337
Space-Vector Modulation 339
6.7 Pulse–Width Modulation with Harmonic Elimination 350
Harmonic Elimination with Multiple Notches 351
Determination of Notch Angles 352
Staircase PWM 355
Summary of Staircase PWM Design 357
Problems 367
Answers 368
7 Design of Magnetics 369

7.1 Magnetic Concepts 369
Dissipative Component 371
Flux and Flux Density 372
Volt-Second Balance 374
PE_TOC.indd xv 3/24/2009 7:51:18 PM

xvi Contents
7.2 Inductor 375

Inductor Value 375
Energy Storage 375
Area Product 377
Window Area 378
Cross-Section Area 378
Design of Inductors 379
Multiple-Winding Inductors 382
7.3 Potential Transformer 383
Operating Principle 384
Turns Ratio 385
Leakage Flux Linkage 386
Equivalent Circuit 388
Area Product 391
Transformer Design 394
7.4 Current Transformer 398
Design of Current Transformers 401
7.5 General Notes on Magnetics 403
Window Utilization Factor 403
Core Shapes 404
Polarity and Dot Convention 406
Magnetic Losses 407
Skin Effect 412
Proximity Effect 413
Problems 420
Answers 421
8 Modeling of Systems 423

8.1 Input–Output Relations 425
8.2 Differential Equations and Linearization 426
8.3 State Space Representation 428
8.4 Transfer Function Representation 431
Modeling an Armature-Controlled DC Motor 434
Poles and Zeros 436
8.5 Block Diagrams 436
Mason’s Rule 440
8.6 Lagrange Method 442
8.7 Circuit Averaging 447
8.8 Bond Graphs 453
Standard Elements 454
One-Ports 454
PE_TOC.indd xvi 3/24/2009 7:51:18 PM

Contents xvii
Two-Ports 455
Multi-Ports (Junctions) 457
Rules for the Selection of Causality 457
Steps in Obtaining the System Model 457
Bond Graph Construction 458
Causality Assignment 462
State Equation Extraction 463
Modeling Switched Power Systems 467
8.9 Space-Vector Modeling 473
Space Vectors 475
Representation of Space Vectors in Orthogonal Co-ordinates 476
Space-Vector Transformations 476
Modeling of Induction Motor 479
State Space Representation of the d–q Model of the Induction Motor 483
Tutorial Exercises 492
Problems 497
Answers 498
9 Control System Essentials 499

9.1 Representation of System in Digital Domain 501
9.2 The z-Transform 502
Physical Significance of z–1 506
9.3 Digital Filter 508
Few Common Digital Filters 512
9.4 Mapping between s-Plane and z-Plane 515
s = 0 + j0 → z = 1 515
s = 0 + jπ/ T → z = –1 516
s = 0 − jπ/ T → z = –1 516
s = –σ + j0 → z = Real Axis between 0 and 1 517
s = σ + j0 → z = Real Axis between 1 and Infinity 517
9.5 Effect of Sampling 518
9.6 Continuous to Discrete Domain Conversion 521
Pole–Zero Map 522
Numerical Integration Methods 523
Hold Equivalence 530
Discretizing the State Equation 533
9.7 Control System Basics 535
Performance Specifications 535
Stability 535
Steady-State Accuracy 536
Speed of Response 538
Measures for Performance Parameters 538
PE_TOC.indd xvii 3/24/2009 7:51:18 PM

xviii Contents
9.8 Control Principles 540

9.9 State Space Method 542
Problems 547
Answers 548
10 Digital Controller Design 549

10.1 Controller Design Techniques 550
10.2 Bode Diagram Method 551
Bode Diagrams 552
Controller Design with Bode Diagram 556
10.3 PID Controller 562
PID Controller Design 567
Integrator Windup in PID Controller 568
Practical PID Controllers 570
10.4 Root Locus Method 573
Root Locus Plot 574
Controller Design 575
10.5 State Space Method 582
Controllability 582
Observability 583
10.6 Full-State Feedback 584
Regulator 585
Tracker 585
10.7 Regulator Design by Pole Placement 587
10.8 Estimator Design 591
Estimator Error Dynamics 592
Separation Principle 593
Current Estimator 594
Reduced Order Estimator 596
10.9 Tracker: Controller Design 599
10.10 Controlling Voltage 600
Operation 605
10.11 Controlling Current 607
Current-Controlled Converter 607
Unity Power Factor Converter 610
Front-End Converter 613
10.12 Control of Induction Motor 619
Scalar Control 619
Vector Control 622
10.13 Output Feedback 631
PE_TOC.indd xviii 3/24/2009 7:51:18 PM

Contents xix
10.14 Induction Motor Control with Output Feedback 633

Plant Dynamics 636
Anti-Alias Filter Dynamics and Augmentation 636
Controller Dynamics and Augmentation 638
Problems 646
Answers 647
11 Optimal and Robust Controller Design 649

11.1 Least Squares Principle 650
11.2 Quadratic Forms 651
11.3 Minimum Energy Principle 653
11.4 Least Squares Solution 655
11.5 Weighted Least Squares 658
11.6 Recursive Least Squares 659
LMS Algorithm 660
Kalman Filter 664
11.7 Optimal Control: Linear Quadratic 668
Linear Quadratic Regulator 670
Linear Quadratic Gaussian 672
Linear Quadratic Tracker: Output Feedback 672
11.8 Induction Motor Example 674
11.9 Robust Controller Design 679
Frequency Domain Performance Specifications 681
Modeling Uncertainties 681
Controller Design Steps 682
Speed Controller Synthesis 685
Problems 693
Answers 694
12 Discrete Computation Essentials 695

12.1 Numeric Formats 695
Fixed-Point Numeric Format 696
Floating-Point Numeric Format 699
General Notes on Number System 702
PE_TOC.indd xix 3/24/2009 7:51:18 PM

xx Contents
12.2 Tracking the Base Point in the Fixed-Point System 703

Addition of Numbers 703
Subtraction of Numbers 703
Multiplication of Numbers 703
12.3 Normalization and Scaling 704
Multiplication Algorithm 707
12.4 Arithmetic Algorithms 709
Reciprocal 709
Square Root 711
Reciprocal of Square Root 713
Hypotenuse 713
Polynomial 714
Sine and Cosine 714
Exponential 715
Logarithm 716
Implementation Examples 717
PI Controller 717
Sine and Cosine 719
Pulse-Width Modulation (PWM) 720
Space Vector PWM 724
Over-Modulation 727
Problems 733
Answers 734
13 Thermal Aspects 735

13.1 Modeling Heat Transfer 736
13.2 Heat Transfer by Conduction 738
13.3 Heat Transfer by Convection 742
Determination of Nusselt Number 743
Free Convection 744
Forced Convection 747
13.4 Heat Transfer by Radiation 753
Reflectance, Absorptance and Transmittance 755
13.5 Heat Transfer by Mass Transport 756
13.6 General Notes on Heat Transfer 758
Mixed Heat Transfer Mechanisms 758
Thermal Resistance of Varying Cross-Sections 759
Mounting of Power Devices on Heat Sink 760
Transient Model for Heat Transfer 761
Best Method of Heat Transfer 762
PE_TOC.indd xx 3/24/2009 7:51:18 PM

Contents xxi

Problems 767
Answers 768
14 Reliability Modeling 769

14.1 Reliability Concepts 771
Failure Rate 771
Failure-Time Probability Density Function 773
Mean Time Between Failure (MTBF), Mean Time to Failure (MTTF) 776
14.2 Modeling for Reliability 777
Series System 777
Parallel System 778
Mesh System 786
14.3 Redundant Systems 788
Full-On Redundancy: Multiple-Survivor Sub-System 789
Switching in Redundant Systems 791
Mean Time to Failure 795
Standby Redundancy 797
Standby Redundancy with Imperfect Switching 799
14.4 Redundant Systems with Non-Identical Components 799
14.5 Design for Reliability 804
14.6 Reliability Apportioning/Allocation 813
Method 1: Equal Apportionment Technique 814
Method 2: AGREE Apportionment Technique 814
Method 3: ARINC Apportionment Technique 815
14.7 Inter-Connection and Wiring 816
14.8 General Notes on Designing for Reliability 822
Connectors 822
14.9 Part Derating 828
Problems 832
Answers 833
15 Reliability Prediction 835

15.1 Similar Equipment Technique 836
15.2 Similar Complexity Technique 836
15.3 Prediction by Function Technique 836
PE_TOC.indd xxi 3/24/2009 7:51:18 PM

xxii Contents
15.4 Parts Count Technique 837

Part Count with Non-Series Model 839
15.5 Part Stress Technique 839
15.6 Environments 842
15.7 Reliability Prediction Process 842
15.8 Component Naming Convention 843
15.9 Component Object Notation 844
15.10 Example Reliability Prediction Process 849
Steps to Design a Circuit 849
15.11 Physics of Failure (PoF) 852
Problems 856
Answers 857
Appendix I 859
Appendix II 861
Appendix III 863
Appendix IV 865
Appendix V 866
Appendix VI 867
Bibliography 909
Index 911
PE_TOC.indd xxii 3/24/2009 7:51:18 PM

Power Semiconductor Switches
Learning Objectives
CHAPTER
1
After reading this chapter, you will be able to:
understand the requirements of an ideal switch and the characteristics of important
power semiconductor switches.
estimate the conduction and switching power losses in various power semiconductor
switches.
model and simulate the power semiconductor switches.
P ower electronic applications deal with the flow of power. However, majority of the applications in power
electronics are based on switching the power flow in order to improve efficiency. As a consequence, one
has to use power semiconductor devices that act as power switches. This chapter discusses various power
semiconductor devices with an emphasis on using them as power electronic switches. There are many semi-
conductor devices that can be used as power switches. Few of the common power semiconductor switches are
diodes, bipolar junction transistors (BJTs), metal oxide semiconductor field effect transistor (MOSFETs),
insulated gate bipolar transistors (IGBTs), thyristors, gate turn-OFF thyristors (GTOs) and metal oxide
semiconductor controlled thyristor switches (MCT). Power MOSFETs and IGBTs have a large role to play
in modern power control equipments and hence their study is emphasized.
The power semiconductor switches may be studied from various viewpoints:
1. physics viewpoint;
2. circuit viewpoint;
3. protection viewpoint;
4. drive viewpoint;
5. modeling viewpoint;
6. packaging viewpoint.
The physics viewpoint explains the operation and the functional features of the device. The circuit view-
point deals with the static and dynamic characteristics. For reliable operation of the power switches, one
must ensure that the electrical and thermal stresses within the device are well below the stated ratings of
the device. These switches handle large currents and dissipate a considerable amount of heat; consequently,
the thermal aspects need detailed attention to ensure that the switch operates within the permissible junc-
tion temperatures. The protection viewpoint focuses on the electrical and thermal stresses within the
Chapter 01.indd 1 3/12/2009 3:47:46 PM

2 Power Electronics
device. In this respect, the safe operating limits for the operation of the power semiconductor device are
addressed. The drive viewpoint emphasizes and clarifies the switching behavior of the power devices that
enables one to synthesize reliable and meaningful drive circuits. The packaging viewpoint focuses on the
device mounting strategies, removal of heat through heat sinks, forced cooling devices and connection
issues.
Modeling switches for purposes of simulation are also important aspects that need to be addressed. The
mathematical representation of the switches will aid in simulating the power topologies so that the circuit
behavior and the waveforms at various parts in the circuit may be determined before the actual hardware
implementation. In fact, it is essential to complement the hardware bread-boarding sessions with electronic
bread-boarding sessions not only to enhance the understanding of the circuit and its operation, but also to
reduce the design cycle time.
1.1 Ideal Switch
T his chapter focuses on semiconductor devices that are operated in such a manner that they behave as
switches. These semiconductor switches are supposed to emulate the operation of an ideal single pole
single throw (SPST) switch. However, it will become clear from the discussions in the sections to follow that
none of the semiconductor switches have all the characteristics of an ideal SPST switch. Before discussing
the features of the semiconductor switches, it is necessary to have a reference list of features that an ideal
SPST switch has in order to aid in a better understanding of the practical semiconductor switch. An ideal
SPST switch will have the following features:
1. ON resistance = 0 (or zero forward voltage drop).
2. OFF resistance = infinity (or zero reverse current).
3. When ON-conducts infinite current in both the forward and reverse directions.
4. When OFF-withstands infinite forward and reverse voltages.
5. It can switch instantaneously from OFF to ON and from ON- to OFF-states.
6. Power dissipated in the switch is zero, that is, both the conduction and the switch transition losses
are zero.
7. ON-to-OFF and OFF-to-ON transitions of the switch are fully controllable.
8. It requires no power to drive or control the switch.
While none of the power semiconductor switches have these ideal characteristics, efforts are continuously
made to improve the performances of the semiconductor switches such that they may tend towards ideal
behavior. There are many different types of power semiconductor devices available commercially. However,
in this chapter, only a few popular generic power semiconductor switch types will be discussed. The discus-
sion will initially try to provide some basic insights into the physics of operation of the semiconductor
junction taking diode as an example. Subsequently the circuit viewpoint is discussed for the diode as well
as few other device types like BJTs, MOSFETs, IGBTs and thyristors to aid in the selection of the devices as
power switches for power electronic applications.
1.2 Diodes
D iodes are devices that have two terminals, namely, (a) anode and (b) cathode. When the anode is more
positive than the cathode, the diode is said to be forward-biased and allows a flow of current from the
anode to the cathode. The forward-biased state of the diode is called the ON-state or the conducting state
Chapter 01.indd 2 3/12/2009 3:47:47 PM

Power Semiconductor Switches 3
of the diode. On the other hand, when the cathode is more positive than the anode, the diode is said to be
reverse-biased and does not permit any flow of current. The reverse-biased state of the diode is called
the OFF-state or the blocking state of the diode. The switching action of the diode is solely dependent on
the anode-to-cathode potential and this is determined by the external circuit. As there is no control on the
switching state, a diode is termed as an uncontrolled switch.
In a metal (e.g. copper), the valence electrons are completely free and roam from atom to atom. In a
semiconductor (e.g. silicon), a covalent bond is imposed between the valence electrons of adjacent atoms.
As a result the motion of valence electrons is coordinated with the motion of valence electrons of an
adjacent atom. Thus, in a pure semiconductor there are very few free electrons. Evidently, the conduction
property of a metal and a semiconductor is very different. If a fraction of a volt is applied across a semi-
conductor, a small amount of current will flow whereas if applied across a metal, a large current will result.
The conductivity of pure semiconductors can be changed by adding impurities. This process of adding
impurities into a pure semiconductor is called doping. By adding some impurities in a pure semiconductor,
its conductivity can be increased and electron (or hole) flow can be easily controlled, for example, pure
silicon plus a pentavalent impurity (e.g. arsenic) increases the mobile free electron charges and is called
n-material, whereas pure silicon plus a trivalent impurity (e.g. boron) increases the mobile hole charges and
is called as p-material.
A p–n junction diode can be formed by growing a single crystal of semiconductor material and doping
with the above impurities in a controlled manner. Figure 1.1 shows a schematic diagram of a p–n junction.
The p-side has a high density of holes (i.e. electron vacancies) and the n-side has a high density of electrons.
These electrons will diffuse from the n- to the p-side and the holes will diffuse from the p- to the n-side. This
will create a space charge layer on either side of the junction called the depletion region wherein the holes
and the electrons combine to form immobile charges as shown in Figure 1.1. This gives rise to an electric
field and a potential barrier. The electric field creates a drift current that acts in such a way as to oppose the
diffusion current. Equilibrium is reached when the diffusion and the drift currents balance each other.
The depletion region at the junction has only immobile charges. The value of the potential barrier in
volts depends on the charge carriers and is governed by the Boltzmann’s relation given in Eq. (1.1). The
p n
Mobile hole Mobile electron
− +
− − − − + + + +
Immobile
− − − − + + + + charge
fo
Depletion region
Figure 1.1 Schematic diagram of p–n junction.
Chapter 01.indd 3 3/12/2009 3:47:47 PM

4 Power Electronics
p n
fο
pp
No. of charge
pn
carriers
0 x
Figure 1.2 Charge carrier levels in the p- and n-regions in the absence of external bias.
charge carrier levels at equilibrium for the p-region and the n-region in the absence of any external bias
voltage are depicted in Figure 1.2. The number of holes in the p-region is denoted by pp and the number of
holes in the n-region is denoted by pn:
−φo /( KT /q ) −φo /V T
pn = pp × e = pp e (1.1)
where VT = KT/q and K is the Boltzmann constant in Joules per degree Kelvin; T is the junction tempera-
ture in degree Kelvin; q is the electron charge in coulomb; pn is the number of holes in the n materials; pp is
the number of holes in the p material; fo denotes the barrier potential.
Note that Eq. (1.1) is an exponential relationship. VT is of the order of few tenths of volt. With this basic
relationship one can try to understand the operation of a p–n junction that is the building block of almost
all semiconductor devices.
On applying a forward bias (positive to p and negative to the n) to the p–n junction as indicated in
Figure 1.3(a), the barrier potential across the p–n junction reduces from the equilibrium value of fo to
(fo − f) and the resulting equilibrium hole density in the n-material is changed, which is given by
Boltzmann law as
−(φ0 −φ )/( KT/q ) φ /VT
pn* = pp × e = pn e (1.2)
Equation (1.2) is a fundamental relationship in junction theory. This shows that a small forward bias
increases the minority carrier (holes) concentration in the n-region exponentially. These excess holes come
from the p-side and are in turn replenished from the external source. As the holes cross over from the
p-region to the n-region, swift recombination results. The charge density decreases with distance, x from the
junction as depicted in Figure 1.3(a). For every electron in the n-material that combines with a hole, there
is another electron entering the n-region from the negative side of the external voltage source. It is important
to understand that this change of charge density with distance in the semiconductor involves a transport of
charge and thereby constitutes a current flow. It takes place predominantly by diffusion.
The hole-current through diffusion is proportional to dpn*/dx. Similarly diffusion current due to the
electron density difference also exists. The diffusion current in the p–n junction is given as
I = I o (e qV /KT − 1) (1.3)
where I is the diode current; Io the saturation current; q the electronic charge in coulombs; K the Boltzmann’s
constant; T the junction temperature in degree Kelvin.
Chapter 01.indd 4 3/12/2009 3:47:48 PM

p n
fο −f
No. of charge pp
carriers
pn
0 x
(a)
p n
fο +f
pp
No. of charge pn
carriers
0 x
(b)
Figure 1.3 Charge densities near a biased p–n junction.
In the conventional conductor, the current and the voltage are related by the well-known Ohm’s law.
However, it is evident from the expression given in Eq. 1.3, that in the case of a p–n junction, the current is
exponentially related to the voltage as a consequence of the Boltzmann relation. On applying a reverse bias
(positive to n and negative to p) to the p–n junction as indicated in Figure 1.3(b), the barrier potential across
the p–n junction increases to (fo + f) and the resulting equilibrium hole density in the n-material is given
by the Boltzmann law as
−(φo + φ )/( KT /q ) −φ /V T
pn′ = pp e = pn e (1.4)
Chapter 01.indd 5 3/12/2009 3:47:48 PM

6 Power Electronics
Under reverse-biased condition, p n′ is less than the equilibrium density of holes in the n-region. This
implies that the holes diffuse from the n-region into p-region. Thus, an extremely small current called the
reverse saturation current results and is denoted as I0. A very large negative voltage disrupts all covalent
bonds and results in a large current. Now it behaves like a metallic conductor and the diode is said to have
attained reverse breakdown.
Static Characteristics
The symbol of a diode is shown in Figure 1.4(a). Lead A that is connected to the p-region of the p–n junc-
tion is called the anode and lead K that is connected to the n-region of the p–n junction is called the
cathode. The v–i characteristics of an ideal diode are depicted in Figure 1.4(b). From Figure 1.4(b), it can
be observed that the voltage across the diode when forward-biased is zero. In forward-biased condition it
allows flow of currents from the anode to the cathode. When the diode is reverse-biased, then the diode
blocks any flow of current. However, the diode can withstand the applied reverse voltage as shown in
Figure 1.4(b).
i
A K v
(a) (b)
i i
1/rd
v Vd v
(c) (d)
Figure 1.4 (a) Symbol of diode; (b) ideal v–i characteristics; (c) actual characteristic;
(d) piece-wise linear characteristic.
Chapter 01.indd 6 3/12/2009 3:47:48 PM

The practical v–i characteristic of the diode is shown in Figure 1.4(c). It shows that during the forward-
bias condition, the diode has a finite resistance called the forward dynamic resistance (rd). For quick
engineering calculations, it is normal to simplify the characteristics by the piece-wise linear characteristic,
as shown in Figure 1.4(d).
Dynamic Characteristics
Turn-OFF of Diode
A forward-biased diode has a charge distribution as shown in Figure 1.3. If a reverse bias is now applied to
the diode, the charges will have to re-distribute from pn* to pn′. This takes a certain amount of time. The
waveform for the current and voltage are shown in Figure 1.5. The test circuit used to obtain the dynamic
characteristics is as shown in Figure 1.5(a).
The excess charge stored in the diffusion region has to be removed before the junction can be reverse-
biased. As long as there are excess charge carriers in the diffusion region (also called the space-charge
region), the junction will be in forward-biased state. The diode voltage will not change from its ON-state
value except for the small decrease due to Ohmic drop caused by the reverse current. After the current goes
negative and the excess charges are removed at time t2, the junction becomes reverse-biased and quickly
acquires the applied negative voltage value. At time t2, the junction charge distribution would have reached
the equilibrium charge distribution of a non-biased junction like that shown in Figure 1.2. From t2 to t3
V1
i
R
−VR
(a)
i i
V1/R V1/R
di/dt
trr
trr
0 t3 t 0 t1 t3 t
di /dt
−VR /R Qrr −VR /R Qrr
Irr Irr
t1 t2 t2
VF − Irr rd VF − Irr rd
VF VF
0 0
−VR −VR
(b) (c)
Figure 1.5 Diode turn-OFF: (a) Test circuit for turn-OFF; (b) idealized dynamic characteristic
during turn-OFF; (c) practical dynamic characteristic during turn-OFF.
Chapter 01.indd 7 3/12/2009 3:47:48 PM

8 Power Electronics
the charge distribution tends towards that of the reverse-biased junction like that shown in Figure 1.3(b).
The diode now starts to withstand the applied reverse voltage and the diode current falls quickly to zero.
In the case of a practical diode, owing to parasitic inductance, the current in the diode does not immedi-
ately reverse like that of the ideal recovery shown in Figure 1.5(b). It reduces at a certain rate decided by
the lead inductance as indicated in Figure 1.5(c). The time interval trr is called the reverse recovery time
and is an important parameter in switching applications. The interval between t1 and t2 is sometimes called
the storage time.
Turn-ON of Diode
If the diode is under reverse bias and it has to be forward-biased, it requires certain time, known as turn-
ON time or forward recovery time before all carriers in the whole junction can contribute to the current
flow. Here, the charges will have to re-distribute from pn′ to p*.
n In the case of turn-OFF, the charges have
to be removed or re-combined whereas in the case of turn-ON the junction has to acquire charges. This
is generally a faster process than removal of the charges. Therefore, turn-ON times will be much faster
than turn-OFF times. In practical situations, it turns out that forward recovery time does not constitute
a serious problem.
Diode Classifications
The diodes are generally classified based on the turn-OFF times. The reverse recovery time (trr) is a measure
of the speed at which the diode can switch (turn-OFF) and therefore gives an indication on the external
switching frequencies that may be used. Based on this, the diodes are classified as follows:
1. Rectifier diodes (trr in microsecond range).

2. Fast recovery diodes (trr in 200–500 ns).
3. Ultra fast recovery diodes (trr ∼ 30–200 ns).
4. Schottky diode (metal semiconductor junction –trr < 30 ns).
In high-frequency switching circuits, fast recovery diodes (types 2–4) should be used depending on the
frequency range of interest and voltage rating. Type 1 is used for low-frequency applications like mains
rectification, as in front-end rectifier for the capacitor input filter.
Diode Parameters
There are various parameters in the datasheet of a diode that need to be understood before selecting a diode
for a particular type of application. These parameters must be calculated a priori before selecting a diode for
the specific application. The important parameters in the datasheet of a diode are
1. Average current, I Fav

2. Root mean square (RMS) current, I Frms
3. Peak current, I F
4. Surge current, I FSM
5. Breakdown or reverse voltage, V RRM
6. Forward drop, V F
7. Dynamic resistance, rd
8. Reverse recovery time, trr
9. I 2t rating
Chapter 01.indd 8 3/12/2009 3:47:49 PM

For sinusoidal currents, IFav , IFrms and IF are related and the relationships are well known. It is not so when
current in the circuit is non-sinusoidal. In such cases, depending on the current and voltage waveforms of
the diodes for the specific circuitry, the corresponding average, rms and peak values must be calculated from
the fundamentals.
In high-power circuits, it is important to make sure that the chosen diode ratings are higher than the
respective values of currents that will flow in the circuit to ensure that the diode can reliably operate. In
detailed datasheets, some of the parameters may be given in the form of nomo-graphs. One may choose the
diode based on the parameter nomo-graphs also.
1.3 Surge Current

T he rectifier has the ability to sustain overload currents for a brief period of time without damaging the
device. Two parameters are specified in datasheets: IFSM and I 2t rating. IFSM is the maximum peak half-
cycle non-repetitive current rating. In applications that use the capacitor input rectifier filter, the problem of
surge current is severe. When the capacitor is fully discharged, the surge current will be enormous at the
time when the rectifier filter is switched ON and the input voltage waveform is at its peak. This problem is
most severe in the case of line-operated switched-mode power supplies when the input or the mains voltage
is at its worst case maximum value of around 270 V rms. If the expected surge current is higher than what
the device can sustain, one has to limit the current using some limiting device such as a series resistor. The
amp-squared second (I 2t) rating is normally associated with protective fuse co-ordination and the rating is
usually given for 10 ms. For proper functioning, the fuse I 2t rating has to be less than the diode I 2t rating.
1.4 Thermal Viewpoint
A n ideal diode does not have any power dissipation. However, a practical diode will have a power loss.
It is essential to quantify the amount of power loss in the diodes so that one can select a suitable heat
sink in addition to budget the power loss in the circuit so that the efficiency can be estimated. There are
different components of power loss in a diode, viz. (a) the ON-state loss (Pon); (b) the OFF-state loss (Poff)
and (c) the switching loss (Pswitching). The total power dissipation (Pd) is given as
Pd = Pon + Poff + Pswitching (1.5)
Pon is the average power loss when the diode is in the ON-state. One can use the piece-wise linear model [see
Figure 1.4(d)] to estimate the power loss. The power loss is estimated as follows:
T
1
T ∫0
Pon = v × i × dt (1.6)
From the piece-wise linear model of the diode depicted in Figure 1.4(d), it can be observed that
v = Vd + (i × rd ) (1.7)
where Vd is the cut-in voltage or the knee voltage of the diode as shown in Figure 1.4(d). From Eqs. (1.6)
and (1.7), the ON-state power loss in the diode is given as
Pon = (Vd × I Fav ) + ( I Frms
2
× rd ) (1.8)
The knowledge of the average and rms values of the current in the circuit can be used to evaluate
the ON-state conduction losses. OFF-state losses are due to the flow of the reverse saturation current
Chapter 01.indd 9 3/12/2009 3:47:49 PM

10 Power Electronics
in the reverse-biased p–n junction. The reverse saturation current is negligibly small and therefore the i2R
loss due to it is generally not significant. The reverse recovery losses are negligible at low frequencies (e.g., at
50 Hz operation). However, at higher switching frequencies the reverse recovery losses are significant and
affect the efficiency of the circuit considerably. In Figure 1.5(c), the shaded area represents the reverse recovery
charge Q rr. The loss during the reverse recovery is given as follows:
Pswitching = Eswitching × fs (1.9)
where Eswitching is the energy spent during the reverse recovery process and fs is the switching frequency.
⎛1 ⎞
Pswitching = ⎜ Q rrVR ⎟ × f s (1.10)
⎝2 ⎠
Q rr can be obtained from the datasheets and can be used to evaluate Pswitching. If the shaded portion of
Figure 1.5(c) is approximated as a triangle, then Q rr can be estimated from the reverse current during
turn-OFF Irr as
1
Q rr = t rr I rr (1.11)
2
Substituting Eq. (1.11) in Eq. (1.10), the switching loss can also be expressed as
1
Pswitching =I V t f (1.12)
4 rr R rr s
Irr, VR and fs are circuit parameters which are known for the specific circuit. trr is obtained from the datasheet
for the diode used. It is important to note that the switching losses are a function of the switching frequency.
The junction temperature of the device should not exceed a certain limit (150°C as specified in the data-
sheet). The heat generated in the junction should be removed in such a manner that at thermal equilibrium
the junction temperature is well below the rated junction temperature. This is made possible by mounting
the device on a heat sink. The thermal issues and the selection of heat sink for a specific calculated device
power dissipation are discussed in Chapter 13.
Diode Model
A true model of a diode should take care of the steady-state and transient behaviors. One of the commonly
used models for simulation is shown in Figure 1.6. Referring to Figure 1.6, id–Vd variables are characterized
by an exponential relationship in accordance with Eq. (1.3). Rs is the Ohmic resistance of bulk material plus
CD
id
A + − K
Rs Vd
CT
Figure 1.6 Diode model.
Chapter 01.indd 10 3/12/2009 3:47:49 PM

the contact resistance. It is ideally zero; however, in practice it has a finite but small value. When a diode is
abruptly switched from the forward to the reverse bias, the excess mobile charge has to be removed. This can
be modeled as a diffusion capacitance CD. For a transition from reverse to forward bias, the mobile charge
must build up. This is characterized by a transition capacitance CT . It should be noted that both CD and CT
are capacitances whose values are not constant but dependent on the charge in it. Circuit simulator pro-
grams also use a similar model.
1.5 Bipolar Power Transistors
T he bipolar transistor (bipolar junction transistor, BJT) is a three-terminal device. It has an emitter lead
E, a collector lead C and a base lead B. The flow of current from the collector to the emitter is con-
trolled by the current through the base. This base current is a fraction of the current through the collector.
Therefore the BJT is called a current-controlled device. In the ON-state, the BJT can allow the flow of
current in only one direction (from collector to emitter for NPN transistors; from emitter to collector for
PNP transistors). Further, they can support only unidirectional voltages during the OFF-state.
BJT is a dual junction device with two possible configurations, namely, NPN or PNP. Figure 1.7(a)
shows a schematic of the vertical cross-sectional structure of the NPN transistor. Figure 1.7(b) shows the
circuit symbol of an NPN transistor. Referring to Figure 1.7(a), the transistor consists of a highly doped
n-type emitter. The base region is a p-type doped semiconductor. The bottom layer is the n-type collector.
The doping of the n-type collector is light towards the collector–base junction and higher towards the
collector terminal. Generally, the collector is the largest region. Larger the collector region, greater is the
amount of voltage that the transistor can withstand during the OFF-state. The arrow in the NPN transistor
symbol points out of the emitter. In the case of the PNP transistor, the arrow in the emitter would point
inwards as depicted in Figure 1.7(c). In general, the arrow would indicate the conventional direction of the
current flow in the device. In the discussions to follow, NPN transistors only will be considered as the
concepts of the NPN transistors can be extended to the PNP transistors without loss of generality.
Figure 1.8 depicts the basic operation of an NPN transistor. The base–emitter junction is forward-biased
by connecting an external voltage VE as shown. Electrons are injected from the emitter into the base. They
B E B
N
P
E
N
C
B B
C E C
(a) (b) (c)
Figure 1.7 NPN transistor: (a) Structure; (b) symbol (c) Symbol of PNP transistor.
Chapter 01.indd 11 3/12/2009 3:47:49 PM

E−B depletion layer C−B depletion layer
IE IC
N P N
E C
− −
B
IB
VE VC
Figure 1.8 Operation of NPN transistor.
appear at the emitter–base junction. Diffusion takes place in a manner similar to that of the p–n junction
diode. However, the base is made intentionally small to prevent re-combination of electrons with holes in
the p-region. This results in a very small IB. As a result, the vast majority of electrons injected across the
emitter–base junction move on to the base–collector junction and further into the n-material of the collector.
Once the injected electrons appear in this region, the effect of VC accelerates them towards collector, through
the load resistor, through VC and then back to VE, thus completing the circuit.
The ratio of the collector current to the emitter current is called the “alpha parameter” of the transistor,
that is, IC/IE = a. This parameter varies between 0.9 and 0.98 depending on the base width. If the base were
wide, then all electrons that have crossed the emitter–base junction will re-combine with holes in the base
and there will be lesser or no electrons reaching the collector. In this case the emitter–base junction and
collector–base junction will act as though they were diodes connected back-to-back. There is no transistor
action taking place in such a case. From the discussion above, it should be noted that the transistor is not
simply a back-to-back connection of two diodes.
The transistor connected as shown in Figure 1.8 is called the normal mode of operation and a is referred
to as aF or aN. One can also have an inverted mode of connection wherein the collector acts as the emitter
and vice versa. The a under this configuration is denoted as aR or aI. One can write the equivalent circuit
model of the NPN transistor as shown in Figure 1.9. It is called the Ebers–Moll model. The equations for
IE and IC for the normal and inverted mode of operations can be written as
VC V T
I C = α F I F − I CO (e − 1) (1.15)
I E = α R I R − I EO (e
VE VT
− 1) (1.16)
Referring to the equivalent circuit model of the NPN transistor depicted in Figure 1.9,
VE V T VC VT
I F = I EO (e − 1); I R = I CO (e − 1); I C + I E + I B = 0
Here IEO is the emitter–base junction saturation current when collector–base junction is zero-biased (Vc = 0);
ICO is the collector–base junction saturation current when emitter–base junction is zero-biased (VE = 0).
Chapter 01.indd 12 3/12/2009 3:47:49 PM

aRIR aFIF
IE IC
E C
IB
IF IR
B
Figure 1.9 Equivalent circuit of an NPN transistor.
Including the bulk resistance and capacitance (similar to the diode case) results in the model shown in
Figure 1.10. CDE and CTE are the emitter–base junction diffusion and transition capacitances, respectively.
Similarly, CDC and CTC are the collector–base junction diffusion and transition capacitances, respectively.
Circuit simulation programs use a further refined version of the model called as the Gummel–Poon model.
BJT is a current-controlled device. Its output characteristics, that is, ic versus Vce is dependent on the base
current ib. Thus,
ic = f (Vce , ib ) (1.17)
where Vce is the collector-to-emitter voltage; ib is the base current and ic is the collector current.
Ideally ic = aie = bib and should be independent of Vce. However, due to “Early effect” (base width
changes as Vce changes) the parameter b changes with Vce. Therefore, the iC versus Vce characteristics change
with different values of the base current as depicted in Figure 1.11.
aRIR aFIF
IE IC
E RE RC C
IF IR
C TE C TC
B′
CDE CDC
RB
Figure 1.10 Equivalent circuit of a transistor with bulk resistance and capacitance.
Chapter 01.indd 13 3/12/2009 3:47:50 PM

ic
Saturation (ON)
region ib
Vcc /RL
Vcc
RL
C
B
Load
line OFF region
0 Vcc Vce
Figure 1.11 Output characteristics of a transistor.
At very low Vce values, the transistor is said to be in the ON-state. This region is called the saturation
region. In this region, the collector current is independent of the base current and depends on the value of
Vce and load resistor RL. A transistor parameter of interest is the ratio of the collector current to the base
current, denoted by b or hFE. The parameter hFE is a useful number essential for design and is supplied by
the manufacturers. Knowledge of iC and hFE gives the base current that will be needed to saturate the transis-
tor or maintain the transistor in the ON-state.
At very low collector-current values, the transistor is said to be in the OFF-region as depicted in
Figure 1.11. Here the transistor is capable of supporting the applied external voltage. The other region of
the output characteristics that is not shaded is called the active region wherein the transistor behaves as a
dynamic resistor. For a transistor to be operated as a switch, it is primarily switched between the satura-
tion region or the ON-region and the OFF-region of the characteristics. However, during the transition
from the saturation region to the OFF-region and back, the operating point may transit through the
active region resulting in a loss called the switching loss.
This sub-section describes the switching processes within a transistor under various base drive conditions.
For a transistor that is in the ON-state, the following three charges may be said to exist.
1. A charge Q b located in the base area.

2. A charge Q ce located in the collector region underneath the emitter.
3. A charge Q cb located in the collector region underneath the base contact.
Chapter 01.indd 14 3/12/2009 3:47:50 PM

Q Q
Qcb
Qce
Qce
Qb
Qb
Vce Ic
Figure 1.12 Charge dependency on Vce and Ic.
However, there are no charges present in a transistor that is in the OFF-state. Figure 1.12 illustrates the
dependency of these charges with respect to the collector–emitter voltage and the collector current.
Referring to Figure 1.12, it may be observed that the charge Q b is independent of the collector–emitter
voltage drop. However, Q b increases linearly as collector current increases. The charge Q ce also increases
linearly with the collector current but decreases linearly with increases in collector–emitter drop. To achieve
lower collector–emitter drop, Q ce must increase. The charge Q cb that is located in the collector underneath
the base contact has a significant effect on the collector–emitter voltage as depicted in Figure 1.12. The
charge Q cb increases rapidly as Vce drop decreases and the transistor goes towards saturation.
Turn-OFF of Transistor
During turn-OFF, the charges in the transistor must be removed. For a transistor that is in the ON-state, the
charges Q b, Q ce and Q cb are as depicted in Figure 1.13(a). A collector current will flow only when there is a
base charge Q b. The charge Q ce will be located in the collector region underneath the emitter. This results in
a low Ohmic collector consequently making the collector–emitter voltage low. An increase in the charge,
Q ce results in a decrease in Vce. The charge Q cb will be located in the collector region underneath the base
contact. When the base–collector region becomes forward-biased, Q ce rapidly increases and Vce decreases.
The process of turn-OFF of the transistor is illustrated in Figure 1.13. Referring to Figure 1.13(a), when
the base voltage is made zero or negative, a negative base current flows as the collector potential is much
greater than the base potential. This removes the charge Q cb. Then the charge Q c will be removed, starting
from the area underneath the edges of the emitter as shown in Figure 1.13(b). Q b also will now start to
decrease.
As Q b starts to decrease, the collector current will gradually be forced towards the center as shown in
Figure 1.13(c). As long as Q b is sufficiently high, the collector circuit will force the collector current to
flow. However, Vce will increase now as the current density becomes higher. With the emitter current
reducing, the negative base-current flows through the base resistance underneath the emitter (Rb – base-
spreading resistance). The charges Q ce and Q b now become increasingly located beneath the center of the
emitter and thus have to be extracted through an increasing resistance. Consequently, the base–emitter
terminal voltage becomes more negative. Subsequently, the emitter current is concentrated in the middle
of the emitter. This marks the end of a period called the storage period wherein most of the stored charges
in the base are removed. The storage period is denoted as storage time ts and is in general given in the tran-
sistor datasheet.
Chapter 01.indd 15 3/12/2009 3:47:50 PM

B E B B E B
n n
p p
Qb Qb
Qcb Qce Qce

n n
C C
(a) (b)
B E B B E B
n n
p p
Rb Qb Rb
n n
Qr
C C
(c) (d)
Figure 1.13 Turn-OFF of the transistor.
Fall time tf begins as soon as Q b is so low that the emitter injection of electrons starts to reduce. The
emitter current decreases with a speed depending on the rate of decrease in Q b. Ie = 0 when Q b = 0. A
trapped rest charge Q r in the collector must still be removed by way of a collector–base current which
appears as a tail current in the turn-OFF waveforms. This rest charge is indicated in Figure 1.13(d).
Turn-ON of Transistor
A transistor that is in the OFF-state has no charges within it. To obtain a low ON-state voltage, a base
current is applied such that the collector charge Q ce is built up, which will accordingly reduce the col-
lector resistance and therefore the collector–emitter voltage (see Figure 1.12). To quickly build up the
collector charge, the base-current waveform should have a peak at the beginning.
The dynamic characteristics of the transistor are shown in Figure 1.14. Referring to Figure 1.14(a),
during the fall time tf , the voltage across the collector–emitter [shown shaded in Figure 1.14(a)] is decided
by the external circuit. If the load is resistive, then the voltage will rise linearly as the collector current falls
linearly. If the load were an inductive one, then the voltage during the tf period would be governed by
LdiL/dt, where L is the inductance of the load and iL is current through the load inductance.
Referring to Figure 1.14(b), during the rise time tr , the current during this time [shown shaded in Figure
1.14(b)] is decided by the external circuit. If the load is resistive, then the current during this time will rise
linearly. If the load across the collector–emitter of the transistor is capacitive, the current during the period
tr is governed by CdVce/dt, where C is the capacitance across the collector–emitter.
Chapter 01.indd 16 3/12/2009 3:47:51 PM

Vce OFF Turn-ON ON Turn-OFF OFF
Determined by
external circuit
(a)
Vce(sat)
0 t
ic
(b)
Determined by
external circuit
Tail current due
to Qr
0 t
ib
(c)
0 t
td
tr ts
tf
Figure 1.14 Switching characteristics of a transistor: (a) Collector–emitter voltage waveform;

(b) collector current; (c) base current.
Calculation of Power Dissipation in a Transistor

The different components of power loss in a transistor are: (a) the ON-state loss (Pon); (b) the OFF-state loss
(Poff) and (c) the switching loss (Pswitching ). The total power dissipation (Pd) is given by
Pd = Pon + Poff + Pswitching (1.18)
Pon is the average power loss when the transistor is in the ON-state. The power loss is estimated as follows:
Chapter 01.indd 17 3/12/2009 3:47:51 PM

T
1
T ∫0
Pon = v × i × dt (1.19)
where v is the voltage across the collector–emitter of the transistor and i is the collector current through the
transistor. During the ON-state, the voltage across the transistor, v = Vce(sat). Therefore Eq. (1.19) becomes
T
1
T ∫0
Pon = V ce(sat) ×
i × dt = Vce(sat) I cavg (1.20)
For applications wherein the collector current is a pulsed current with a flat top Ic during the ON-state, the
ON-state loss is given by
Pon = Vce(sat) × Ic × D (1.21)
where D is the duty cycle given by the ratio of the ON-time to the total switching period.
Like in the case of the diodes, the OFF-state losses are generally negligible. In the case of the switching losses,
Pswitching depends on the nature of the load as is evident from the dynamic characteristics shown in Figure 1.14.
A representative example case is discussed wherein the load is resistive. For resistive loads, assuming a linear rise
and fall of voltages and currents, the switching losses can be calculated as discussed in the following sub-sections.
Switching Loss from OFF-State to ON-State

Referring to Figure 1.14, when the base drive signal is initiated, there is a time duration (td) called the delay
time during which time the voltage and the current waveforms do not change much as the charges Q b, Q ce
and Q cb are being set up. This is followed by the time duration tr. During the time tr, the voltage across the
device falls from Vcc to Vce(sat). Without much error, this fall may be assumed to be linear for a resistive load.
The collector–emitter voltage during the rise time tr is
⎛ t⎞
v ce = Vcc ⎜ 1 − ⎟ (1.22)
⎝ t r⎠
The collector current for a resistive load will linearly rise from zero to Ic. This is given as
⎛t⎞
ic = I c ⋅ ⎜ ⎟ (1.23)
⎝ tr ⎠
The power dissipation during the OFF-state to ON-state transition is given by
t
r
⎛ t⎞ ⎛t⎞ V It f
POFF-ON = ∫ Vcc ⎜1 − ⎟ I c ⎜ ⎟ dt = cc c r s (1.24)
0
⎝ t r ⎠ ⎝ t r ⎠ 6
where fs is the switching frequency.
Switching Loss from ON-State to OFF-State

Referring to Figure 1.14, when the base drive signal is stopped, there is a time duration ts called the storage
time during which the voltage and the current waveforms do not change much as the charges Q b, Q ce and
Q cb are being removed. This is followed by the time duration t f . During the time t f , the current through the
device falls from Ic to zero. Without much error, this fall may be assumed to be linear for a resistive load. The
collector current during the fall time tf is given as
⎛ t⎞
ic = I c ⎜ 1 − ⎟ (1.25)
⎝ tf ⎠
Chapter 01.indd 18 3/12/2009 3:47:51 PM

The collector–emitter voltage for a resistive load will linearly rise from Vce(sat) to Vcc. This is given as
⎛t⎞
v ce = Vcc ⎜ ⎟ (1.26)
⎝ tf ⎠
The power dissipation during the ON-state to OFF-state transition is given by
t
f
⎛ t ⎞ ⎛t ⎞ Vcc I c t f f s
PON-OFF = ∫ I c ⎜1 − ⎟⎠ Vcc ⎜⎝ t ⎟⎠ dt = (1.27)
0
⎝ tf f 6
where fs is the switching frequency. The total switching loss is given as
Pswitching = POFF-ON + PON-OFF (1.28)
Substituting Eqs. (1.24) and (1.27) in Eq. (1.28), one obtains
Vcc I c f s (t r + t f )
Pswitching = (1.29)
6
It should be observed that the switching losses are proportional to the frequency of switching. The load is generally
never resistive. If the load is inductive, the voltage across the transistor during turn-OFF is determined by Ldic/dt,
where L is the external load inductance. If the load is capacitive, the current through the transistor during turn-
ON is determined by CdVce/dt, where C is the load capacitance as seen at the collector. The switching losses must
be estimated for non-resistive loads also in a manner similar to that discussed for the resistive load. The total power
dissipated is the sum of the ON-state loss and the switching loss. The power dissipated as above must be trans-
ported away from the junction such that the junction temperature remains at a safe value in equilibrium condi-
tions. A proper heat sink has to be selected. The thermal calculations are discussed in Chapter 13.
1.6 Safe Operating Area (SOAR)
T he maximum electric field across the collector–emitter of the transistor must remain below a critical value
at all instants for proper functioning of the transistor. The electric field across the transistor is dependent
on the collector-current density and the applied collector–emitter voltage. The electric field increases with
increasing collector voltage. It also increases with increasing collector-current density. If the collector voltage is
lowered then a higher collector-current density is permitted and vice versa. During switching transitions, there
are some destructive combinations of the collector-current density and the collector voltage that are likely to
occur. The SOARs give information about a given device on the current and voltage handling capabilities.
The collector-current density is dependent on the collector current and the amount of current crowding
in the regions of the collector. The amount of current crowding is different for turn-ON (positive base voltage)
and turn-OFF (negative base voltage) conditions. Therefore, the allowed combinations of the collector cur-
rent and collector voltage will differ for turn-ON transition and turn-OFF transition. This information is
available in the forward safe operating area (FSOAR) and the reverse safe operating area (RSOAR). A transis-
tor in the ON-state or the OFF-state has the operating point along the Y-axis or the X-axis when either Vce or
ic is zero. However, during switching transitions, both Vce and ic are non-zero and finite. This means that the
operating point will be in the region of the I-quadrant of the Vce–ic characteristic. In selecting a transistor, the
one which operates within both the FSOAR and the RSOAR, for the specific circuit, should be chosen.
With positive voltage applied to the base, the shape of a typical SOAR characteristic is as shown in Figure
1.15(a). The solid line shows the SOAR for DC operation and the dashed lines shows the SOAR for pulsed
operation. Operation outside the safe operating area is not allowed. For pulsed operation, the FSOAR increases
Chapter 01.indd 19 3/12/2009 3:47:51 PM

ic ic
ICM Pdmax ICM

fs
Secondary
breakdown
0 Vceo Vce 0 Vceo Vcesm Vce
(a) (b)
Figure 1.15 Safe operating area (SOAR): (a) Forward SOAR (FSOAR); (b) reverse SOAR (RSOAR).
and for very small duty cycles, the FSOAR becomes square. The SOAR is designed to indicate the current, power
dissipation, voltage and second breakdown limits of the transistor as depicted in Figure 1.15(a). The power dissi-
pation limit is the hyperbolic portion of the limit curve that is given by the maximum possible dissipation allowed
for the device. The second breakdown limit is the straight sloped line connecting the power dissipation limit and
the voltage limit lines. The second breakdown is generally triggered by combinations of high collector-voltage and
high collector-current density. With a positive voltage applied to the base, the region of highest current density is
at the edge of the emitter which conducts a substantial proportion of the collector current. During sudden change
of currents (especially during switching), a thermal gradient gets generated across the cross-section of the collec-
tor–emitter current flow. This thermal gradient will result in the non-uniform spreading of the current in the
device. This uneven spreading produces localized hot spots and it can in turn reduce the local resistance, further
increasing the non-uniformity. This will result in hot spots in the device that will finally destroy the device.
During turn-ON of the transistor, the high resistance of the collector region is reduced by the introduc-
tion of holes from the base and electrons from the emitter. This process is known as conductivity modulation.
However, during turn-OFF of the transistor, these extra holes and electrons constitute a stored charge that
must be removed from the collector before the voltage across the depletion region can develop. To turn OFF
the transistor, a negative voltage is applied to the base and reverse base-current flows. During turn-OFF, it is
essential that the device stays within its reverse safe operating area (RSOAR) that is shown in Figure 1.15(b).
1.7 Paralleling of Transistors
I n high-power switching applications, transistors can be paralleled to share the load current. The transistor
has a negative temperature co-efficient of resistance. As a result, one has to ensure that the current sharing
between various parallel transistors is uniform, such that no transistor is over burdened and goes into a
thermal runaway. It is very common to add an emitter resistance as shown in Figure 1.16 to equalize the
currents. The emitter resistors give a negative feedback in the base–emitter circuit. If Ic increases then Vbe
will decrease, which would result in a decrease of Ib. This consequently will counteract the increase of Ic. The
emitter resistor values are governed by the following equations:
Chapter 01.indd 20 3/12/2009 3:47:51 PM

Ic
Ic1 Ic2
+ + +
Vbe1 − Vbe2 −
Vbe
R1 R2
Figure 1.16 Paralleling of transistors.
Vbe = Vbe1 + RI c1 = Vbe2 + RI c2 (1.30)

where R1 = R2 = R. From Eq. (1.30), one obtains
ΔVbe = Vbe1 − Vbe2 = R ΔI c (1.31)
Limit DVbe to around 0.2 V. Let DIc (difference in collector currents between the two transistors) of about
0.5 A be tolerated. This means R = 0.2/0.5 = 0.4 Ω. It should be noted that R dissipates power. Higher
R means more dissipation and higher Vbe (and Vcc) requirements. Paralleling transistors simplify heat trans-
fer problem by spreading the heat produced in a group of transistors over a large surface area heat sink.
1.8 Darlington Connection
H igh-power, high-voltage transistors generally have a low hFE (b ). The hFE of saturated high-voltage
transistors can be as low as 2. This means a very high base drive current is required. Therefore, the base
drive circuit is no longer a simple low-power circuit. To circumvent this problem, Darlington connection as
shown in Figure 1.17 can be used. In this case, the Ib requirement can be cut down to Ic/(hFE*hFE). It should
Ic
Q1
Ib
Q2
Figure 1.17 Darlington connections of transistors.
Chapter 01.indd 21 3/12/2009 3:47:52 PM

be noted that Vce(sat) of the Darlington device is higher (∼1 V to 1.2 V) as opposed to 0.4–0.6 V Vce(sat) of a
single transistor. This implies that the power dissipation is higher in Q2 for a given load.
1.9 Power MOSFETs
T he metal oxide semiconductor field effect transistor (MOSFET) is a three-terminal device. It has a
source lead S (analogous to the emitter of the transistor), a drain lead D (analogous to the collector of
the transistor) and a gate lead G (analogous to the base of the transistor). The flow of current from the drain
to the source is controlled by the voltage applied between the gate–source terminals. Thus the MOSFET is
a voltage-controlled device. In the ON-state, the MOSFET can allow current in only one direction (from
drain to source for n-channel MOSFETs and from source to drain for p-channel MOSFETs). In the OFF-
state, they can support only unidirectional voltages in a manner similar to the BJT.
A structural schematic of an n-channel power MOSFET is shown in Figure 1.18(a). It consists of three
layers, viz., (a) an n-type semiconductor that is connected to the drain D; (b) an n-type semiconductor that
is connected to the source S and (c) in between the drain and the source is a p-type semiconductor that is on
the substrate. Generally the substrate is connected to the source within the MOSFET. The gate is connected
to a metallic conductor. This is insulated from the bulk of the MOSFET by an insulator that is generally an
oxide of some metal such as silicon dioxide. The symbol for the n-channel MOSFET is shown in Figure
1.18(b) and the symbol for the p-channel MOSFET is shown in Figure 1.18(c). Note that for the n-channel
MOSFET, the arrow mark shown in the symbol points inwards towards the gate and in the case of a p-chan-
nel MOSFET, the arrow mark points outwards, that is, away from the gate.
When a positive voltage is applied at the gate as shown in Figure 1.19, an electric field will be generated
across the insulating oxide layer. This brings about a polarization of charges within the oxide layer. Conse-
quently, these charges attract electrons from the p-material thereby creating an induced n-channel that
D
D
G
Polysilicon n
metal
S
Substrate (b)
p
G
n S
G
Insulating layer of
metal oxide S D
(e.g. silicon dioxide)
(a) (c)
Figure 1.18 (a) Structural schematic of an n-channel MOSFET; (b) symbol of an

n-channel MOSFET; (c) symbol of a p-channel MOSFET.
Chapter 01.indd 22 3/12/2009 3:47:52 PM

+ −
+ −
+ − p
+ −
G + −
+ −
n
Vgs
Figure 1.19 Operation of the MOSFET.
bridges the n-type drain and source regions of the MOSFET. If a positive drain voltage is now applied
between D and S terminals, the n-channel bridge provides a conducting path between the two n-regions.
This conducting path that bridges the two n-type semiconductors in the MOSFET is called the inversion
layer. As the voltage Vgs is increased, the width of the inversion layer increases. The MOSFET is said to have
reached full enhancement when the inversion layer width is maximum.
It can be noted that there is a parasitic NPN BJT between the drain and source contacts with p substrate
serving as the base of the parasitic BJT. To minimize the possibility that this transistor is ever turned ON,
the p-type substrate is shorted to the source region as indicated in Figure 1.18. As a result of this short, a
parasitic diode called the body diode exists between the drain and source of the MOSFET. This integral
body diode can be used in many circuit configurations to advantage, thereby avoiding the use of an external
diode that is required in such topologies.
Figure 1.18(a) gives the structural schematic of the MOSFET from a functional viewpoint. However, a typi-
cal cross-section of the practical MOSFET is as shown in Figure 1.20. The MOSFET has associated capacitances
with respect to its terminals. This parasitic capacitance model of the MOSFET is shown in Figure 1.21. One should
be aware that while manufacturing the MOSFET, a reverse diode from source to drain (i.e., anode at source and
cathode at drain) gets inherently built by shorting the substrate and the source. This internal diode or the body
diode is advantageous as it can be used for freewheeling purposes in inverters that are driving inductance loads.
Referring to Figure 1.20, the capacitance between the drain and the source Cds varies in accordance with
the width of the depletion layer that in turn depends on the voltage being supported by the device. The gate
source capacitance consists of three components Cgsn, Cgsp and Cgsm (the metal to gate lead capacitance that
is not indicated in Figure 1.20). Of these capacitances, Cgsp depends on the width of the inversion layer that
in turn depends on the applied gate–source voltage. Of particular interest is the feedback capacitance Cgd.
This capacitance plays a dominant role during switching. This is also the most voltage dependent. Cgd is
essentially two capacitors in series such that
1 1 1
= + (1.32)
C gd C gdox C gdbulk
Chapter 01.indd 23 3/12/2009 3:47:53 PM

Cgsn Cgsp Cgdox

S S
n n
p p
Cds Cgdbulk n
Figure 1.20 Structure of the MOSFET.
Cgd
G
Cds
Cgs
S
Figure 1.21 Parasitic capacitance model of the MOSFET.
Chapter 01.indd 24 3/12/2009 3:47:53 PM

All the capacitances vary according to the thickness of the depletion region. Figure 1.21 depicts the capaci-
tance model of the MOSFET. Most MOSFET datasheets do not refer to these capacitances. However, they
provide information on the input capacitance Ciss, the output capacitance Coss and the feedback capacitance
Crss. The datasheet capacitances relate to the parasitic capacitances shown in Figure 1.21 as follows:
Ciss: Parallel combination of Cgs and Cgd
Coss: Parallel combination of Cds and Cgd
Crss: Equivalent to Cgd
The magnitude of the gate–source voltage Vgs essentially determines the drain current. Figure 1.22(a) shows
the drain current id versus gate–source voltage Vgs characteristics. It can be observed from Figure 1.22(a) that
there is a threshold voltage VgsT below which the device is OFF. This VgsT is of the order of 3–4 V in most
power MOSFETs. Figure 1.22(b) shows the drain current id versus drain–source voltage Vds. In the active
region, the drain current is independent of the drain–source voltage and depends only on Vgs. This relation
is approximately given by
id ∝ (Vgs − VgsT )2 (1.33)
The MOSFET is said to be in the Ohmic region if Vds is less than (Vgs – VgsT). The boundary between the
Ohmic and active region is depicted in Figure 1.22(b). The Ohmic region corresponds to the saturation
region of BJT output characteristics. The MOSFET in the ON-state will operate in the Ohmic region. The
drain to source resistance of the MOSFET in the Ohmic region is denoted by RDS(ON) and is an important
selection parameter for the device.
In order to reduce the power dissipation in the ON-state, the device should have a low RDS(ON). A Vgs
of ∼10 V will take the MOSFET to the ON-state. However, in practical circuits, a Vgs of 15 V is applied in
order to take the MOSFET to full enhancement such that the RDS(ON) is low. The maximum Vgs that can be
applied is limited to +20 V, as only this much Vgs can be supported by the thin oxide layer.
Power MOSFETs are intrinsically faster than BJTs as they have no excess minority carriers that must
be moved into or out of the device during turn-ON and turn-OFF conditions. The only charges that must be
Vgs − VgsT
id id Vgs
Ohmic
region
Active
region
VgsT Vgs Vds

(a) (b)
Figure 1.22 Static characteristics of MOSFET.
Chapter 01.indd 25 3/12/2009 3:47:53 PM

Gate
drive (a)
signal
Vgs
VgsT
(b)
t0 t1 t2 t3 t4 t5 t6 t7 t8
Q0 Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8
id
(c)
Vds
(d)
Figure 1.23 Switching characteristic of MOSFET.
moved in or out are those on the stray parasitic capacitances due to oxide and depletion layers. The parasitic
capacitances have been shown in Figure 1.21. The dynamic or the switching characteristics of the MOSFET
are shown in Figure 1.23. The switching characteristic of the MOSFET is divided into various intervals.
These intervals are explained based on the capacitance model of the MOSFET shown in Figure 1.21. For
the purpose of the discussion regarding the switching behavior of the MOSFET, the current fed to the gate
can be assumed to be a constant flowing into the gate terminal during the turn-ON period, that is, t0–t4.
During the period t4–t5 when the MOSFET is ON, the gate current is negligible. During the period t5–t8,
when the MOSFET is turning OFF, the gate current is again assumed as constant but flowing out of the
gate terminal. The equivalent circuit of the MOSFET during the turn-ON process at the various intervals is
depicted in Figure 1.24. Referring to Figures 1.23 and 1.24, the switching action of the MOSFET is dis-
cussed referring to the various time intervals.
Chapter 01.indd 26 3/12/2009 3:47:54 PM

Vdd Vdd
Load Load
igd
D D
ig Cgd ig Cgd
G G
id = f(Vgs)
igs
Cgs Cgs
gnd S gnd S
(a) (b)
Vdd Vdd
Load Load
D D
− −
Cgd Cgd
ig + ig +
G G
RDS(ON)
+ +
Cgs Cgs
− −
gnd S gnd S
(c) (d)
Figure 1.24 Equivalent circuits during turning ON of a MOSFET: (a) During interval t0 – t1;
(b) during interval t1 – t2; (c) during interval t2 – t3; (d) during interval t3 – t4.
During Interval t0–t1

At t0, let there be an initial charge of Q 0 at the gate to source capacitance. An amount of charge (Q1 – Q 0) is
required to charge the gate to the threshold voltage VgsT . The gate current flows through Cgs and Cgd as shown
in Figure 1.24(a).
At turn-ON, Cgs is charged by a current
dVgs
igs = C gs
dt
Chapter 01.indd 27 3/12/2009 3:47:54 PM

and Cgd is charged by a current

dVgd dVds
igd = C gd ≈ −C gd .
dt dt
An amount of charge (Q2 – Q1) is required such that the full drain current id flows. The equivalent circuit is
as shown in Figure 1.24(b).

The equivalent circuit is as shown in Figure 1.24(c). During this interval, almost the entire gate current
flows through the capacitance Cgd. As a consequence, the charge on Cgs is almost constant. This leads to an
almost flat Vgs waveform during this interval as indicated in Figure 1.23(b). The drain-to-source voltage
waveform starts falling as the MOSFET moves further into the Ohmic region. The rate at which Vds falls as
Cgd gets charged is
dVds dVgd ig
=− = (1.34)
dt dt C gd

Once the drain–source voltage has dropped to its ON-state value of idRDS(ON), Cgs gets charged such that the
MOSFET is driven to full enhancement. An amount of charge (Q4 – Q3) is required for the MOSFET to be
taken to full enhancement. The equivalent circuit is as shown in Figure 1.24(d).

During this interval, the MOSFET is fully ON and no charge is required to maintain it in its ON-state.
Thus the gate current is zero during the time when the MOSFET is fully ON.

At instant t5, the gate signal to the MOSFET is made zero. Now the MOSFET begins the turn-OFF pro-
cess. The charge (Q6–Q5) is the excess gate charge from Cgs that is required to be removed to bring the
MOSFET out of full enhancement.

(Q7–Q6) amount of charge has to be discharged to bring the MOSFET out of the Ohmic region to the
OFF-state. The drain–source voltage Vds increases during this time as indicated in Figure 1.23. The feedback
capacitance Cgd is discharged during this period. The rate at which the voltage across the feedback capaci-
tance Cgd, reduces will be same as the rate at which the drain–source voltage rises.

(Q8–Q7) is the charge required to be discharged from Cgs to fully turn-OFF the MOSFET as indicated in
Figure 1.23. One should note that the gate drive source should be able to sink during the discharge dura-
tion. The interval t1–t2 where the drain current rises up to its full value corresponds to the rise time tr and
the interval t7–t8 where the drain current falls from its full value to zero corresponds to the fall time tf . [A
typical MOSFET has the following specifications: for example, for BVZ 45 B, 500 V, 10 A MOSFET has
Cgs ∼ 4000 pF; Cgd ∼ 100 pF; Cds ∼ 300 pF; tr ∼ 100 ns; tf ∼ 120 ns.]
Chapter 01.indd 28 3/12/2009 3:47:54 PM

D D
Rg1 Rg2
G
G
S
S
Figure 1.25 Paralleling of MOSFETs.
Paralleling of MOSFETS
The Ohmic region resistance RDS(ON) is due to majority carriers. It has a positive temperature coefficient.
This is advantageous when paralleling the devices. Current sharing resistors in series with the source are
normally not required (as in the case of BJTs). Incidentally it can be noted that due to positive temperature
coefficient of RDS(ON), the second breakdown will not take place in MOSFETS. In case there is a non-uni-
form distribution of the drain current, as the current increases in any section, the RDS(ON) will increase for
that particular local section and as a consequence the current will reduce in that section and spread from the
hot spot region to other areas.
If MOSFETS have to be connected in parallel, the schematic shown in Figure 1.25 is generally used.
The drains and sources can be connected in parallel directly. While this can take care of the static condition,
it is necessary to put small gate resistances of 10–100 Ω to damp high frequency oscillations that might
result between gate leads and the parasitic device capacitances.
One should note that for a given rated current, the voltage drop will be much higher for a high-voltage
MOSFET (RDS(ON) ∼1 to 2 Ω) than for a low-voltage device (RDS ∼ 0.1 Ω). This is due to the fact that the
bulk n-region, that is, the drain region has to be larger in order to be capable of withstanding the electric
field due to the higher applied voltage.
Handling Precautions for MOSFET

The gate–source oxide layer is very thin. Therefore, when touched with our body (that can have a minimum
of 1–2 μC charge) the fields created can be extremely high so as to breakdown the oxide layer.
Remarks on BJT and MOSFET

1. BJTs have lower conduction losses in ON-state but higher switching losses whereas power MOSFETs
have higher conduction losses and lower switching losses. Thus at higher frequencies, power MOSFETS
are the obvious choice.
2. BJTs are current-controlled devices whereas MOSFETs are voltage-controlled devices.
3. It is easier to parallel power MOSFETs when compared to BJTs.
4. BJTs suffer from secondary breakdown, but the power MOSFETs do not suffer from secondary
breakdown.
Chapter 01.indd 29 3/12/2009 3:47:54 PM

EXAMPLE 1.1 Consider a MOSFET with a Cgs = 4000 pf. From the basic charge–potential
relationship,
Q = CV
The voltage across the gate–source capacitance that would get developed for a charge
of 1 μC is
Q 1 × 10−6
Vgs = = = 250 V
C gs 4 × 10−9
The maximum gate–source voltage that is allowable is only 20 V. Therefore, just

touching a MOSFET may damage it permanently and hence the MOSFETs must
be always carried in a conducting pad, so that the charge leaks if it tries to build up.
While not in use, it is recommended to short the gate and source of the MOSFET
with a conducting jumper or wire so that the gate–source voltage does not build
due to spurious electrostatic fields. In order to limit the Vgs to 20 V, it is always a
good practice to clamp the voltage across the gate-source terminals of a MOSFET
by connecting a 15 V Zener diode close to gate and source leads.
Power Dissipated in a Power MOSFET

The ON-state loss for the MOSFET is given by
PON = I drms
2
RDS( ON) (1.35)
RDS(ON) has a temperature coefficient of approximately 0.6% per oC and should be considered to cal-
culate RDS(ON) (and PON) as the device temperature rises. In a manner similar to the discussion with
regard to the BJT, the switching loss for resistive loads, assuming linear rise and decay of Vds and id ,
is given as
1
Pswitching = Vds I d (t r + t f ) f s (1.36)
6
Model for Power MOSFET

There are different levels of MOSFET models that have been used depending on the application. For
integrated circuit fabrication using simulation, more complex models are necessary. Figure 1.26 shows a
simple model that can be used for macro circuit simulation. In Figure 1.26, the capacitance subscript
“gs” means gate to source, “gd” means gate to drain, “gb” means gate to bulk and “sb” means substrate
to bulk.
1.10 Insulated Gate Bipolar Transistor (IGBT)
T he IGBT is a combination of the advantageous features of MOSFET (like high drive input imped-
ance, voltage control, fast switching) and BJTs (like low ON-state losses and high OFF-state voltage
capability). The IGBT is also a three-terminal device. It has an emitter lead E (analogous to the emitter of
Chapter 01.indd 30 3/12/2009 3:47:55 PM

D
Cgb
RD
Cgd
Substrate
RDS Id
G Rgs Rss
Cgs Csb
RS
Figure 1.26 Simulation model of a MOSFET.
the BJT), a collector lead C (analogous to the collector of the BJT) and a gate lead G (analogous to the gate
of the MOSFET). The flow of current from the collector to the emitter is controlled by the voltage applied
between the gate–emitter terminals. Therefore, like the MOSFET, the IGBT is also a voltage-controlled
device. In the ON-state, the IGBT can allow current in only one direction (from collector to the emitter
for n-channel IGBTs and from emitter to the collector for p-channel IGBTs). Further, they can support
bipolar voltages during the OFF-state.
A structural schematic of an n-channel IGBT is depicted in Figure 1.27(a). It consists of four layers: a
p-type semiconductor that is connected to the collector C, an n-type semiconductor that is connected to the
emitter E and in between is a p–n semiconductor junction. The gate is connected to a metallic conductor.
This is insulated from the bulk of the IGBT by an insulator that is generally an oxide of metal, like, silicon
dioxide. The symbol for the n-channel IGBT is shown in Figure 1.27(b) and that for a p-channel IGBT is
shown in Figure 1.27(c).
When a positive voltage is applied at the gate as indicated in Figure 1.28, an electric field will be directed
through the insulating oxide layer. This brings about polarization of charges within the oxide layer. As a
consequence, electrons from the p-material get attracted towards the insulating oxide layer thereby creating
an induced n-channel that bridges the two n-type sections of the IGBT. The induced n-channel along with
the two n-sections forms an equivalent n-type semiconductor. The p-type semiconductor connected to the
collector C along with the equivalent n-type semiconductor forms an equivalent p–n junction and behaves
like a diode. Because of the formation of the equivalent p–n junction, the IGBT is capable of withstanding
reverse voltages also during the OFF-state.
From the structural schematic of the IGBT, it is evident that an IGBT can be viewed equivalently as a
MOSFET with a diode connected to its drain as indicated in Figure 1.29. Alternatively, it can also be
viewed as a MOSFET and a PNP transistor as shown in Figure 1.30.
The detailed structure of the IGBT is shown in Figure 1.31. One should note that the IGBT has a para-
sitic thyristor between its collector and emitter as indicated in Figures 1.31 and 1.32. This parasitic thyristor
should not latch. If it latches then the IGBT will loose the gate control. Therefore, when the IGBT is
Chapter 01.indd 31 3/12/2009 3:47:55 PM

C C
p
G
Polysilicon E
metal n
(b)
p E
G
n
G
Insulating layer of
metal oxide
(e.g., silicon dioxide) E C
(a) (c)
Figure 1.27 (a) Structural schematic of an n-channel IGBT; (b) symbol of an n-channel IGBT;
(c) symbol of a p-channel IGBT.
+ −
+ −
+ − p
+ −
G + −
+ −
n
Vge
Figure 1.28 Operation of the IGBT.
Chapter 01.indd 32 3/12/2009 3:47:55 PM

p
G
n G
Figure 1.29 Equivalent circuit of IGBT: Diode and MOSFET.
C
C
p
n n
p p
G
n
G
E
Figure 1.30 Equivalent circuit of IGBT: Transistor and MOSFET.
manufactured, Rs as indicated in Figure 1.32 is made very low so that the drop across it is less than cut-in
voltage of the base–emitter of the NPN transistor in the parasitic thyristor. This way the NPN transistor is
always off, thereby avoiding the latching problem. One should be aware that while manufacturing the
IGBT, a reverse diode from emitter to collector (i.e., anode at emitter and cathode at the collector) is built
separately as it is not inherent like in the case of the MOSFETs. This internal diode or the body diode
is required for freewheeling purposes in topologies that are driving inductance loads.
Chapter 01.indd 33 3/12/2009 3:47:55 PM

E E
n n
p p
n n
Figure 1.31 Structure of an IGBT.
Parasitic
thyristor
G Rs
Figure 1.32 Equivalent circuit of the IGBT showing the parasitic thyristor.
The ic–Vce characteristics of an n-channel IGBT are shown in Figure 1.33 and the ic–Vge characteristics are
shown in Figure 1.34. From Figure 1.33 it is observed that the IGBT can support both forward and reverse
voltages during the OFF-state. From Figure 1.34, it is seen that if Vge is less than the threshold voltage, the
IGBT is in the OFF-state.
Chapter 01.indd 34 3/12/2009 3:47:56 PM

ic
Active region
Vge
Vce
Reverse characteristics Forward characteristics
Figure 1.33 Static ic–Vce characteristic of an IGBT.
ic
VgeT Vge
(Threshold voltage)
Figure 1.34 Static ic–Vge characteristic of an IGBT.
In comparison, for a given ic, the Vce(sat) of the IGBT is less than the Vds of a comparable MOSFET
when ON but greater than Vce(sat) of a comparable BJT.
The dynamic behavior of an IGBT is similar to that of a MOSFET when viewed from the gate side and is
similar to that of a BJT when viewed from the collector side. However, the maximum switching frequency
is limited as compared to that of a MOSFET.
The switching times of the IGBT are related to the gate–emitter voltage and collector-current wave-
forms. Figure 1.35 shows the typical switching waveforms. The switching times of an IGBT are mainly
determined by its internal capacitances and parasitic inductances together with the internal resistance
Chapter 01.indd 35 3/12/2009 3:47:56 PM

Vge
90%
(a)
10%
t
Vce
(b)
t
ic
90%
(c)
10%
t
td(on) td(off) tf
tr
ton toff
Figure 1.35 Dynamic switching characteristics of an IGBT.
of the gate-side voltage source. In order to charge and discharge the capacitances rapidly and to reduce the
transients caused by the gate circuit inductance, a low internal impedance of the gate-side voltage source is
desirable. It shortens the switching times and reduces the switching losses. On the other hand, a very fast
turn-ON causes a high peak reverse recovery current through the body diode that appears as an additional
peak collector current. During very fast turn-OFF, a high transient voltage is caused by the parasitic collec-
tor–emitter inductance as shown in Figure 1.35(b).
It is very important to keep the parasitic inductance in the gate circuit at a minimum by using very short
leads. This inductance might otherwise generate parasitic oscillations in conjunction with the IGBT
capacitances. The maximum rated gate–emitter voltage as specified in the datasheets in most cases is ±20 V.
It is recommended to connect a 15 V Zener diode between the gate and the emitter as a protective clamper.
Turn-ON
During turn-ON, the IGBT behavior is more like that of a MOSFET. The collector–emitter voltage, the
collector-current and the gate–emitter voltage waveforms are similar to the turn-ON waveforms of the
Chapter 01.indd 36 3/12/2009 3:47:57 PM

MOSFET. When the gate–emitter voltage reaches the threshold value of VgeT , the collector-current starts to
rise. The time interval between the instant when Vge reaches 10% of its final value and the instant when ic
reaches 10% of its final value is called the turn-ON delay time td(on). The subsequent time interval up to the
instant when the collector current reaches 90% of its final value is called the rise time tr. During this period
of time, most of the turn-ON power dissipation takes place. The sum of the turn-ON delay time td(on) and
the rise time tr is called the turn-ON time ton. At the end of ton, the collector–emitter voltage Vce often has
not yet fallen to its final value of Vce(sat). This has to be considered when calculating the turn-ON dissipation.
The collector-current peak shown in Figure 1.35 indicates the peak reverse recovery current of the body diode
or the freewheeling diode. This peak current has to be taken into account in any turn-ON power dissipation
calculations as well.
Turn-OFF
During turn-OFF, the IGBT behaves more like a BJT. To turn the IGBT OFF, the voltage in the gate con-
trol circuit is switched to zero. Referring to Figure 1.35, there is first a turn-OFF delay time td(off ) that is the
interval between the instant when the gate–emitter voltage has fallen to 90% of its initial value and the
instant when the collector current has fallen to 90% of its initial value. The subsequent period of time up to
the instant when the collector current has fallen to 10% of its initial value is called the fall time tf . The sum
of td(off) and tf is called the turn-OFF time toff .
Tail Current
The tail current and tail time are properties that are specific to the IGBT. However, they also depend on the
operating conditions. The tail current is higher for lower saturation voltage Vce(sat). The trade-off for the favor-
able reduction of the conducting state power loss is an increase in the tail current. Since at pulse frequencies
of 10–20 kHz the switching losses are more prevalent, most IGBTs are designed for a low tail current at the
expense of the saturation voltage which would be a little higher than the minimum possible value.
Remarks on IGBT
1. The power loss calculations for the IGBT are similar to the loss calculations indicated for BJTs and
MOSFETs.
2. The gate drive power is low such as in the case of MOSFETs.
3. IGBT has considerably greater overload capability as compared to a MOSFET. For a MOSFET, the
peak overload is about 5–6 times the continuous drain current whereas for the IGBT, the peak overload
is about 20 times the continuous collector current. An IGBT is capable of taking a short-circuit current
(across a 600 V bus) for a period of 10 μs.
4. Over voltage robustness is less than that of MOSFETs.
5. IGBTs are more like the BJTs during turn-OFF and like the MOSFETs during turn-ON.
6. Unlike the BJT, where the collector current is limited by the current gain, that is (hFE(sat)ib), the IGBTs
do not have such limitations. The collector current is limited only by the external source.
1.11 Thyristor Family
T he term thyristor is a generic name for a semiconductor switch having four or more layers and is in
essence a p–n-p–n structure. Thyristors form a large family of semiconductor switches. If an Ohmic
connection is made to the first p-region and the last n-region and no other connection is made to any other
intermediate region, then the device is a diode thyristor. If an additional Ohmic connection is made to the
Chapter 01.indd 37 3/12/2009 3:47:57 PM

intermediate n-region or the intermediate p-region, the device is called a triode thyristor. If an Ohmic con-
nection is made to both intermediate regions, then the device is a tetrode thyristor. All such devices have a
forward characteristic of the form shown in Figure 1.36.
There are three categories of thyristor reverse characteristic: blocking (as in normal diodes like in sym-
metric SCRs), conducting (large reverse current at low reverse voltages like in asymmetric SCRs and GTOs)
and approximate mirror image of the forward characteristic (bi-directional thyristors like in DIACs and
TRIACs). The simplest thyristor structure and the most common is the reverse blocking triode thyristor
usually referred to as the silicon-controlled rectifier (SCR). The more complex thyristor structure is the
bi-directional triode thyristor or TRIAC. In this section, the various devices in the thyristor family will be
briefly reviewed.
Silicon–Controlled Rectifier (SCR)

SCR is a three-terminal device with an anode terminal A, a cathode terminal K and a control or gate termi-
nal G. Structurally it is similar to a BJT with an extra p-layer at the anode. The structure of the SCR is
depicted in Figure 1.37(a) and the circuit symbol is shown in Figure 1.37(b).
iA
ig = 0
VBR
VBF V
AK
Figure 1.36 Forward characteristics of the thyristor family.
K G K
n n
A
n
p
G
A K
(a) (b)
Figure 1.37 (a) Structure of an SCR; (b) symbol of an SCR.
Chapter 01.indd 38 3/12/2009 3:47:57 PM

Application of a negative voltage to the anode–cathode of an SCR will reverse bias the SCR. In this condition,
the junctions J1 and J3 are reverse-biased and J2 is forward-biased. The device is said to be in the OFF-state and
blocks the reverse voltage. On the other hand, applying a positive voltage to the anode–cathode, as shown in
Figure 1.38, with the gate open, will forward bias the SCR. With the anode–cathode voltage positive, it can be
observed that junctions J1 and J3 are forward-biased and J2 is reverse-biased. The SCR is still in the OFF-state
and blocks the applied forward voltage. This is a property unique to the thyristor family. If forward-bias voltage
is further increased, the junction J2 will breakdown and the device turns ON behaving in a manner similar to
a diode. The static characteristic is shown in Figure 1.39. Figure 1.39(a) shows the static characteristics when
the gate current is zero. The voltage at which the junction J2 breaks down and the SCR conducts is called the
forward breakover voltage VBF . If the gate of the SCR is connected to a voltage source and a gate current is
allowed to flow, then the breakover voltage reduces as indicated in Figure 1.39(b).
The SCR action is best understood by modeling it as two BJTs connected as shown in Figure 1.40. With
the gate open or shorted to the cathode, the device is OFF and no current flows from anode to cathode,
except for a very low leakage current. If an external positive current pulse is applied to the gate, it becomes
the base current of the NPN transistor Q2. Consequently, the collector current of Q2 supplies the base cur-
rent to the PNP transistor Q1. The collector current of Q1 then further increases the base current of Q2 and
so on. This iterative action maintains the device in the conducting state even if the gate signal is now
removed. The device continues to conduct till the anode voltage is less positive than the cathode voltage.
Figure 1.39(b) shows the static characteristics with gate current as a parameter. It should be noted that once
the device is ON, the gate looses complete control of the device and the device cannot be turned OFF
through the gate. The SCR can be switched OFF only if the anode current is brought below a threshold
called the holding current value for that SCR.
di/dt Effect On supplying the gate trigger to a forward-biased SCR, the SCR will turn ON. The anode
current starts to rise after a small delay. The rate of change of the anode current will depend on the nature of
J1
n VAK
J2
G
p
J3
Figure 1.38 Biasing the SCR.
Chapter 01.indd 39 3/12/2009 3:47:57 PM

iA
ig = 0
VBR
VBF VAK
(a)
iA
ig
VBR
VBF VAK
(b)
Figure 1.39 V–I characteristic of an SCR (a) with gate open; (b) with gate current applied.
A
p
Q1
n
p
p
G G
Q2
n
(a) (b)
Figure 1.40 Two transistor equivalent circuit of an SCR.
Chapter 01.indd 40 3/12/2009 3:47:58 PM

the load. Initially the ON-state current is concentrated in a small area around the gate region. If the rate of
rise of this ON-state current is high, then the area around the gate region will become overheated that may
permanently damage the device. Therefore, it should be ensured that the external circuit di/dt encountered
by the SCR should be less than the rated di/dt of the device.
Turn-OFF Characteristics The turn-OFF characteristic waveforms of the anode current and the anode–
cathode voltage are similar to that of the diode with the added restriction that all junctions including the
gate junction should fully recover before the SCR gets into a forward-blocking state. The turn-OFF charac-
teristics are depicted in Figure 1.41.
A high rate of rise of the applied forward voltage across the anode and cathode (dv/dt) can trigger the
SCR. To keep the SCR turned OFF, the re-application of the OFF-state forward voltage must be delayed to
avoid re-triggering due to high dv/dt. The circuit commutated recovery time tq is measured from the zero
crossover of the current as shown in Figure 1.41. It should be noted that tq is an important parameter. SCRs
are often classified as line-commutated or converter grade if tq > 50 μs and as inverter grade if tq < 20 μs. For
the line commutated converter, tq is less important since the half-period is 10 ms. But for inverter circuits
switching at higher frequencies, inverter grade SCRs with appropriately lower tq will have to be used.
dv/dt Effect Figure 1.41 shows a possible problem with large rate of forward voltage, dv/dt. If dv/dt is
large, the SCR may self-trigger even in the absence of gate current. This is due to the fact that there exists
a leakage current through the reverse-biased junction. This leakage current flows through the capacitance
of junction J2 and is given by i = c dv/dt. This leakage current is dependent on dv/dt. If this leakage current
iA
trr
VAK
dVF /dt Forward

blocking
tq
Figure 1.41 Anode current and anode–cathode voltage during turn-OFF.
Chapter 01.indd 41 3/12/2009 3:47:58 PM

is larger than the latching current of the device, then the device will self-trigger and go into conduction
even in the absence of the gate current. Therefore a high dv/dt has the possibility of undesirable latch up
of the device. Practical SCRs have dv/dt limits of about 200–500 V/μs. In circuits wherein the dv/dt
exceeds the rated dv/dt limits of the device, additional circuits called snubber circuits will have to be used
to lower the dv/dt.
Other Parameters
The various current ratings like ITav , ITrms , ITsurge are similar to those discussed for the diode. The ratings are
a bit complicated compared to a diode. The value given in the datasheet for the mean ON-state current ITav
is valid for a certain waveform and case temperature. ITrms signifies the heating effect due to i2R dissipation
and is limited due to the thermal stress on the device. ITSM is the maximum permissible peak current of half
sine wave with a duration of 10 ms at a specified temperature.
Power dissipated in a thyristor is calculated in the same way as that of the diode. Gate-current losses add
to the power losses in the case of thyristors as compared with the diode losses. Power loss calculation depends
on the type of waveform too. Thyristors are very rugged devices as they can carry considerable overload cur-
rents without exceeding the junction temperature.
Circuit Model
There are many circuit models of SCRs available in the literature and circuit simulator libraries like SPICE.
One such simple model is given here. It can be constructed using the two-transistor analog of an SCR as
shown in Figure 1.42. Diode DFOR is added to model the forward breakdown. A resistance is added across
the base–emitter of Q2 to provide a discharge path for the stored charges in Q2. The various parameters for
the PNP and NPN transistors are included in the model to give a satisfactory simulation performance.
DIACs and TRIACs

DIAC is a two-terminal device without any gate. It is a bi-directional switch. The breakdown voltage is gener-
ally kept low (∼30 V) because it is generally used as a threshold device in many applications. Figure 1.43(a)
shows the DIAC characteristics and Figure 1.43(b) depicts its symbol.
Q1
DFOR
125 Ω Q2
Figure 1.42 Circuit model of an SCR.
Chapter 01.indd 42 3/12/2009 3:47:58 PM

iA
MT2
VAK
MT1
(a) (b)
Figure 1.43 DIAC: (a) v–i characteristic; (b) symbol.
iA
ig
MT2
VAK
G MT1
(b)
(a)
Figure 1.44 TRIAC: (a) v–i characteristic; (b) symbol.
TRIAC is also a bi-directional thyristor switch but it has a gate terminal that is used for controlling the
turn-ON of the TRIAC. It is mainly used for AC control applications. As the switch is bi-directional, its
terminals are called main terminal 1 and 2 (MT1 and MT2), instead of anode and cathode. Figure 1.44(a)
shows the characteristics of the TRIAC and Figure 1.44(b) depicts its symbol. A simple TRIAC control
circuit is shown in Figure 1.45. This circuit uses a DIAC as a threshold device for triggering the TRIAC at a
particular source phase angle.
There are different ways in which the gate can be fired. The most common is with reference to MT1. The
polarity of voltage across MT2 and MT1 decides the direction of current flow, and gate pulse is always positive
irrespective of direction of current flow. TRIACs are generally used for 50 Hz applications. With inductive
load, the TRIAC will be subjected to large dv/dt and therefore protective snubber circuits must be used.
Chapter 01.indd 43 3/12/2009 3:47:58 PM

Load
MT2
DIAC
TRIAC
MT1
230 V AC
MT2 G MT1
Figure 1.45 TRIAC control circuit using DIAC.
Gate Turn-OFF Thyristor (GTO)

It can be recalled that an SCR cannot be turned OFF by gate control. It needs an extra turn-OFF/commutation
circuit. GTO has been devised to provide turn-OFF through the gate terminal. The GTO is a four-layer
device that can be turned ON with a positive gate current pulse and turned OFF with a negative gate current
pulse. Figure 1.46(a) shows some of the commonly used GTO symbols.
The physical structure of a GTO is shown in Figure 1.46(b). Referring to figure, it may be observed that
the gate and cathode are interspersed. Such a sectionalized geometry helps in having better gate control.
The anode layer has n- and p-regions interspersed as shown in Figure 1.46(b). This gives a structure termed
as the “anode short” structure that is found to improve the turn-OFF time.
GTO action can also be described by the two-transistor model as discussed for the case of an SCR. By
suitably controlling the doping profiles of the n- and p-regions, the a and b of the transistors, it is possible
to achieve gate turn-OFF with a negative gate current. Removal of excess charge is an important part of
the turn-OFF process. The negative gate current and the anode current during the turn-OFF process are
depicted in Figure 1.47.
The SCR and GTO differences are highlighted below:
1. GTO has higher ON-state drop as compared to the SCR. The GTO has an ON-state drop of about
3–4 V whereas the SCR has an ON-state drop of around 1.5–2 V.
2. The latching and holding currents are high for a GTO. As an example, a typical 110 A SCR has IH of
150 mA and IL of 300 mA, whereas even a low current GTO, a typical 18 A GTO has IH = 600 mA and
IL = 900 mA. For 30 A GTO, they are more than 1 A. Therefore the triggering pulse must be long
enough or continuous for the GTO unlike the case of the SCR.
3. The reverse-blocking capability of a GTO is very low (∼20 V) and hence cannot be used in rectifier
applications.
4. When the device is to be turned OFF, the turn-OFF amplification factor (anode current with respect to
the turn-OFF gate current) can be anywhere between 1 and 5 and the turn-OFF time between 1 and
several milliseconds. There is a maximum current rating beyond which gate cannot turn OFF the GTO.
This is called repetitive controllable ON-state current (or extinguishable current).
5. GTO has a large dv/dt capability (1000 V/μs) compared to a SCR (200–500 V/μs).
6. GTOs can be used up to 1–4 kHz of switching frequency range.
Chapter 01.indd 44 3/12/2009 3:47:59 PM

A A A
G G G
K K K
(a)
K K K
G G
n+ n+ n+
J3
p
J2
n−
J1
p n+ p n+
(b)
Figure 1.46 (a) Symbols of GTO; (b) structure of a GTO.
ig
0
t
iA
0
t
ts tf
Figure 1.47 Turn-OFF characteristics of a GTO.
Chapter 01.indd 45 3/12/2009 3:47:59 PM

A
A
ON
(n-channel)
OFF
(p-channel)
G
G
K
K
(a) (b)
Figure 1.48 MCT configurations.
MOS-Controlled Thyristor (MCT)

MCT is an interesting device wherein both the turn-ON and turn-OFF can be controlled. However, com-
mercially it is not as popular as the MOSFETs and the IGBTs. Figure 1.48 shows some possible MCT
functional configurations. Figure 1.48(a) shows an MCT where turn-ON is done in the usual way but turn-OFF
is done using a MOSFET. Figure 1.49(b) shows another configuration wherein an n-channel MOSFET is
used for turning ON the MCT and a p-channel MOSFET is used for turning OFF the MCT. The major
features of an MCT are:
1. It has low conduction voltage drop: 1/3 to 1/2 of a comparable IGBT.
2. It combines high-speed, low-power gate turn-ON–turn-OFF advantages.
3. It has high dv/dt and di/dt capabilities.
| CONCLUDING REMARKS
This chapter discussed few important viewpoints of There are many semiconductor devices avail-
the semiconductor devices giving a flavor for the able commercially that have been fabricated with
operation of the various devices as a power switch. newer and improved technologies. However, this
It should, however, be observed that no single switch chapter discusses only a few generic types, high-
meets all the features of an ideal switch as described lighting the important issues that need to be stud-
in Section 1.1. One should be aware of the non- ied and understood when encountered with a new
idealities of each type of the power switch so that device. With special regard to the semiconductor
the power electronic circuits can be modeled with devices being used as power switches in power elec-
greater closeness to the physical system. tronic circuits, one should study with emphasis on
Chapter 01.indd 46 3/12/2009 3:47:59 PM

three aspects in general, viz. (a) static characteristics device. The next section provides some laboratory
that give insight on the steady-state operating exercises that should be implemented both on a sim-
points, (b) dynamic characteristics that give the ulation platform like spice and by hardware bread-
behavior of the switch during turn-ON and turn- boarding to appreciate the functional features of a
OFF and (c) power loss within the device due to particular power semiconductor switch. The labora-
conduction and switching. tory exercises are focused on popular devices like
The theoretical insights should be cemented diodes, BJTs, MOSFETs and IGBTs. However, the
with the experience that can be gained from practi- exercises can be extended for any other power semi-
cal work to strengthen the understanding of the conductor switch too.
| LABORATORY EXERCISES
1. Consider the diode test circuit shown in (b) Find out the forward dynamic resistance of
Figure 1.49 where Vi is a voltage source and D the diode from the i–v characteristic.
is a diode. The diode D is appropriately (c) Use a square waveform input voltage source
chosen such that the current and voltages are and observe the current through the diode
within ratings. and the voltage across it during turn-ON
and turn-OFF.
Mode of implementation: The above circuit
(d) Compute the experimental switching loss
can be studied by
and compare with the theoretical estimate.
a. Simulation in Spice
(e) Compute the experimental conduction loss
b. Hardware breadboarding
Tasks for study: (f ) Repeat steps (a)–(c) for different loads by
(a) Use a variable DC voltage source. Measure varying RL.
the current through and the voltage across (g) Repeat steps (a)–(c) for different source fre-
the diode for different input DC voltage quencies. What is the effect on the diode
values. Reverse the polarity of the input power loss?
voltage source and measure the current (h) Measure the trr of the diode from the cur-
through and the voltage across the diode rent waveform through the diode. What is
for different input voltage values. Tabulate the effect of the load on trr?
the diode current and voltage values and (i) Measure the reverse recovery charge Q rr
plot the i–v characteristics of the diode. from the experimental results.
(j) Use a sinusoidal waveform input voltage
D source and observe the current through
and the voltage across it during voltage
id transition from positive to negative.
Vd Compute the power loss in the device
Vi RL and compare the loss with that obtained
when using a square waveform input source.
2. Consider the two BJT test circuits shown in the
Figure 1.50 where Vb is the base drive pulse
Figure 1.49 Diode test circuit. source. Vb should be chosen such that it is capable
Chapter 01.indd 47 3/12/2009 3:48:00 PM

of both sourcing and sinking current. Vc is the a and b parameters. Plot the Ic–Vce charac-
collector DC supply voltage. Rc is the collector teristics for different base current values.
load and Rb is the base drive resistor. The BJT is (b) Set Vb to be a pulse source of frequency 20
chosen with appropriate rating to handle the kHz. Measure the collector current and
collector current and the maximum Vce as the collector–emitter voltage waveforms
decided by Rc and Vc, respectively. The circuit and observe the turn-OFF and turn-ON
of Figure 1.50(a) consists of a resistive collector portions.
load and that of Figure 1.50(b) consists of an (c) Compute the experimental switching loss
R–L load. The diode D is connected as shown and compare with the theoretical estimate.
to provide a freewheeling path for the inductor (d) Compute the experimental conduction
current when the BJT Q is turned OFF. loss and compare with the theoretical
estimate.
Mode of implementation: The above circuit (e) Repeat steps (b)–(d) for different loads by
can be studied by varying RL.
a. Simulation in Spice (f ) Repeat steps (b)–(d) for different base pulse
b. Hardware breadboarding frequencies. What is the effect on the BJT
power loss?
Tasks for study: (g) What is the effect of Rb on the turn-ON
(a) Set Vb to be a DC source. Measure the and turn-OFF times? Why?
collector current and the collector–emitter (h) Perform the steps (b)–(g) for the inductive
voltage across transistor for different load test circuit of Figure 1.50(b).
values of Rb. Tabulate the BJT collector
current, base current, emitter current and 3. Consider the two MOSFET test circuits shown
collector–emitter voltage values. Find the in Figure 1.51 where Vg is the gate drive pulse
Vc
Vc L
D
Rc Rc
ic
Rb ib Rb ib
Q Vce Q Vce
Vb Vb
(a) (b)
Figure 1.50 BJT test circuit.
Chapter 01.indd 48 3/12/2009 3:48:00 PM

source. Vg should be chosen such that it is Id–Vds characteristics for different gate
capable of both sourcing and sinking current. voltage values.
Vd is the drain DC supply voltage. Rd is the (b) Set Vg to be a pulse source of frequency 20
drain load and Rg is the gate drive resistor. kHz. Measure the drain current and the
The MOSFET is chosen with appropriate drain–source voltage waveforms and
rating to handle the drain current and the observe the turn-OFF and turn-ON
maximum Vds as decided by Rd and Vd, respec- portions.
tively. The circuit of Figure 1.51(a) consists of (c) Compute the experimental switching
a resistive drain load and that of Figure 1.51(b) loss and compare with the theoretical
consists of an R–L load. The diode D is con- estimate.
nected as shown to provide a freewheeling path (d) Compute the experimental conduction
for the inductor current when the MOSFET is loss and compare with the theoretical
turned OFF. estimate.
(e) Repeat steps (b)–(d) for different loads by
varying RL.
can be studied by
(f ) Repeat steps (b)–(d) for different gate pulse
frequencies. What is the effect on the
MOSFET power loss?
Tasks for study: (g) What is the effect of Vg on the turn-ON
(a) Set Vg to be a DC source. Measure the and turn-OFF times?
drain current and the drain–source voltage (h) What is the effect of Rg on the turn-ON
across the MOSFET for different values of and turn-OFF times?
Vg. Tabulate the MOSFET drain current, (i) Measure the gate current and explain its
and drain–source voltage values. Plot the pulse shape.
Vd
Vd
L
Rd D
Rd
id
Rg ig Rg ig
Vds Vds
Vg Vg
(a) (b)
Figure 1.51 MOSFET test circuit.
Chapter 01.indd 49 3/12/2009 3:48:00 PM

(j) Perform the steps (b)–(g) for the inductive Tasks for study:
load test circuit of Figure 1.51(b). (a) Set Vg to be a DC source. Measure the col-
(k) Compare the switching speed and power lector current and the collector–emitter volt-
loss of MOSFET with comparable BJT. age across transistor for different values of
Vg. Tabulate the IGBT collector current and
4. Consider the two IGBT test circuits shown in
collector–emitter voltage values. Plot the Ic–
Figure 1.52 where Vg is the gate drive pulse
Vce characteristics for different gate voltage
source. Vg should be chosen such that it is
values.
capable of both sourcing and sinking current.
(b) Set Vg to be a pulse source of frequency 20
Vc is the collector DC supply voltage. Rc is the
kHz. Measure the collector current and
collector load and Rg is the gate drive resistor.
the collector–emitter voltage waveforms
The IGBT is chosen with appropriate rating to
and observe the turn-OFF and turn-ON
handle the collector current and the maximum
portions.
Vce as decided by Rc and Vc, respectively. The
(c) Compute the experimental switching
circuit of Figure 1.52(a) consists of a resistive
loss and compare with the theoretical
collector load and that of Figure 1.52(b) con-
estimate.
sists of an R–L load. The diode D is connected
(d) Compute the experimental conduction loss
as shown to provide a freewheeling path for the
inductor current when the IGBT is turned
(e) Repeat steps (b)–(d) for different loads by
OFF.
varying RL.
Mode of implementation: The above circuit (f ) Repeat steps (b)–(d) for different gate pulse
can be studied by frequencies.
a. Simulation in Spice (g) Perform the steps (b)–(f ) for the inductive
b. Hardware breadboarding load test circuit of Figure 1.52(b).
Vc
Vc L
D
Rc Rc
ic
Rg ig Rg ig
Q Vce Q Vce
Vg Vg
(a) (b)
Figure 1.52 IGBT test circuit.
Chapter 01.indd 50 3/12/2009 3:48:00 PM

| FILL IN THE BLANKS

1. Diodes are devices that have two terminals: 18. The safe operating areas give information about
________ and ________. the capability of a device to handle the speci-
fied ________.
2. When the anode is more positive than the
________, the diode is said to be forward-biased. 19. MOSFET is a ________.
3. When the ________ is more positive than the 20. The Ohmic region resistance RDS(ON) is due to
________, the diode is said to be reverse-biased. ________.
4. During turn-OFF of the diode, the ________ 21. RDS(ON) has a ________ temperature coefficient.
stored in the diffusion region has to be ________
before the junction can be reverse-biased. 22. Compared to MOSFETs, BJTs have ________
conduction losses but ________ switching-losses.
5. As long as there are excess charge carriers in the dif-
fusion region, the junction will be in ________. 23. BJTs are ________-controlled devices whereas
MOSFETs are voltage-controlled devices.
6. The bipolar transistor is a ________ device.
24. IGBT is a ________.
7. BJT is a ________ device.
25. IGBT can support ________ voltages during
8. BJT is a ________ device. the OFF-state.
9. Based on the doping type, there are two types 26. In comparison, for a given ic, the Vce(sat) of the
of BJT: ________ and ________. IGBT is ________ than the Vds of a compara-
10. Larger the collector region, ________ is the ble MOSFET when ON but ________ than
amount of voltage that the transistor can with- Vce(sat) of a comparable BJT.
stand during the OFF-state. 27. IGBT has ________ overload capability as
11. The ratio of the collector current to the emitter compared to a MOSFET.
current is called the ________ of the transistor. 28. IGBT has ________ overvoltage robustness than
12. The alpha parameter varies between ________ MOSFETs.
and ________.
29. IGBTs are more like the ________ during
13. Using the Eber-Molls model, the collector cur- turn-OFF and like the ________ during
rent is given as ________. turn-ON.
14. Using the Eber-Molls model, the emitter cur- 30. Thyristors is a generic name for a semiconduc-
rent is given as ________. tor switch having ________ layers.
15. The ratio of the collector current to the base 31. Application of a ________ voltage to the
current is given by ________. anode–cathode of an SCR will reverse-bias the
SCR.
16. The electric field across the transistor is depen-
dent on the ________ and the ________. 32. DIAC and TRIAC are ________ switches.
17. The electric field increases with ________ and 33. In a GTO, the turn-OFF is ________ whereas
with ________. in an SCR, the turn-OFF is ________.
Chapter 01.indd 51 3/12/2009 3:48:01 PM

34. GTO has ________ ON-state drop as compared 36. The reverse-blocking capability of a GTO is
to the SCR. very low and hence cannot be used in _______
applications.
35. The latching and holding currents are _______
for a GTO as compared to the SCR.
| DESCRIPTIVE QUESTIONS
1. What are the characteristic features of an ideal 14. What is second breakdown?
switch?
15. How are transistors connected in parallel?
2. What is a controlled switch? Discuss.
3. What is an uncontrolled switch? 16. How does the current gain, or the beta parame-
ter, increase with the Darlington configuration?
4. What is the depletion region?
17. What are the semiconductor layers in a
5. “The value of the potential barrier in volts depends MOSFET structure?
on the charge carriers and is governed by the
Boltzmann’s relation.” Discuss. 18. What is the inversion layer? How is it formed?
6. With respect to the diode’s V–I static charac- 19. Discuss the MOSFET structure and the vari-
teristics, explain the difference among the ous associated capacitances?
ideal, piece-wise linear and the actual V–I 20. How are the MOSFET capacitances Ciss, Coss
characteristics. and Crss related to the parasitic capacitances
Cgs, Cgd and Cds?
7. Discuss the turn-OFF process in the diode.
21. What is gate threshold voltage?
8. What is meant by reverse recovery time for a
diode? 22. What is Ohmic region of the MOSFET static
characteristics?
9. What is the ON-state loss in a diode?
23. Explain the MOSFET turn-OFF and turn-ON
10. What is the switching loss in a diode? processes?
11. Describe the switching process within a 24. How is the positive temperature coefficient
transistor. advantageous for paralleling of MOSFETs?
12. Derive relationship for the power dissipation 25. Discuss the functional structure of the IGBT.
within the transistor between the applied
26. Explain the turn-ON and turn-OFF of IGBT
collector–emitter voltage, the collector-
with waveforms.
current and the switching frequency of the
transistor for (a) resistive load in the collec- 27. What is tail current in IGBT?
tor, (b) resistive–inductive load in the col-
28. What are holding and latching currents in SCR?
lector and c) resistive–capacitive load across
the collector–emitter of the transistor. 29. What is forward breakover voltage in SCR?
13. Discuss forward and reverse safe operating 30. Discuss the effect of di/dt and dv/dt on the
areas. SCR switch.
Chapter 01.indd 52 3/12/2009 3:48:01 PM

| PROBLEMS
1. Find the ratio of the diffusion current to the satu- Icm = 15 A, Vcesat = 0.3 V, hFEmin = 100, td = 1 μs,
ration current for a p–n junction at 75oC having ts = 2 μs, tr = 1.5 μs, tf = 1.5 μs. Calculate the
a forward potential of 0.6 V applied across it. power loss in the BJT.
2. A diode and a 10 Ω resistor are connected in 7. For Problem 6 above, plot the device power
series to a square wave voltage source of 50 V dissipation as the switching frequency varies
peak. Find the conduction loss for the diode if from 1 kHz to 100 kHz.
the forward dynamic resistance is (a) 0.1 Ω and 8. Two BJTs are connected in parallel to share the
(b) 0.2 Ω at the operating point. (Assume the load current. In order to ensure sharing, two
forward barrier potential to be 0.7 V.) equal-valued resistors with value R are con-
3. A diode and a 10 Ω resistor are connected in nected in the emitter leads of the BJTs. For a
series to a pulse voltage source of 50 V peak. If mismatch in the base–emitter voltage of 0.2 V
the forward dynamic resistance is 0.1 Ω at the between the two BJTs, a mismatch in the col-
operating point, then find the conduction loss of lector currents of the two BJTs should be less
the diode when (a) pulse frequency is 20 kHz than 1 A. Calculate the value of the resistors
and pulse width 40 μs, (b) pulse frequency is 20 that needs to be connected in the emitter
kHz and pulse width is 10 μs, (c) pulse frequency leads.
is 10 kHz and pulse width is 80 μs and (d) pulse 9. A MOSFET is operated such that the operating
frequency is 10 kHz and pulse width is 20 μs. point is in the active region. The MOSFET has
4. A diode and a 10 Ω resistor are connected in a gate–source threshold voltage value of 2.5 V.
series to a square wave voltage source of 50 V A gate–source voltage of 5 V is applied to the
peak. The reverse recovery time for the diode is gate–source terminals of the MOSFET which
given to be 200 ns. Find the switching loss of results in the flow of drain current. On increas-
the diode when (a) input frequency is 100 kHz ing the gate–source voltage to 7.5 V, what is the
and (b) input frequency is 50 kHz. factor by which the drain current increases?
10. A MOSFET is driving a 10 A resistive load
5. For Problem 4 above, estimate the reverse
from a 100 V DC supply. The base drive
recovery charge.
signal is switching at frequency of 100 kHz
6. A BJT is driving a 10 A resistive load from a and duty cycle of 0.6. The MOSFET has the
100 V DC supply. The base drive signal is following datasheet specifications: RDS(ON) =
switching at frequency of 50 kHz and duty 0.1 Ω, tr = 100 ns, tf = 150 ns. Calculate the
cycle of 0.75. The BJT has the following data- conduction and switching power losses in the
sheet specifications: Vbesat = 0.7 V, Vceo = 30 V, MOSFET.
| ANSWERS
Fill in the Blanks
1. anode and cathode 3. cathode; anode 5. forward-biased state
2. cathode 4. excess charge; removed 6. three-terminal
Chapter 01.indd 53 3/12/2009 3:48:01 PM

7. current-controlled 17. increasing collector voltage; 26. less; greater

8. dual junction increasing collector-current 27. greater
9. NPN transistor and PNP density 28. less
transistor 18. current and voltage 29. BJTs; MOSFETs
10. greater 19. voltage-controlled device 30. four or more
11. alpha parameter 20. majority carriers 31. negative
12. 0.9 and 0.98 21. positive 32. bi-directional
13. I c = αF I F − I co (eVC /VT − 1) 22. lower 33. controlled; uncontrolled
14. I e = α R I R − I eo (eVe /VT − 1) 23. current 34. higher
15. b or hFE 24. voltage-controlled device 35. higher
16. collector-current density; 25. bipolar 36. rectifier
applied collector-emitter
voltage
Chapter 01.indd 54 3/12/2009 3:48:01 PM

Drive Circuits
Learning Objectives
CHAPTER
2
design the base drive circuits for BJTs.
design the gate drive circuits for MOSFETs and IGBTs.
learn the principles of snubber circuits for power switches.
A lmost all power applications are increasingly moving away from linear and dissipative mode of opera-
tion and towards switched-mode operation to achieve improved efficiency. The power devices now are
being used primarily as switches. To achieve proper and efficient operation of the power equipments, the
power devices should be driven in an appropriate manner that makes them behave as switches.
In Chapter 1, the operation, characteristics and models of the various power switches are discussed. Of
the power switches discussed, the bipolar junction transistors (BJTs), the metal oxide semiconductor field
effect transistors (MOSFETs) and the insulated gate bipolar transistors (IGBTs) can be switched ON and
OFF by means of a control signal. This feature of controllability has made the them increasingly popular
in power electronics systems like the DC–DC converters, AC-to-DC rectifiers with power factor ( pf )
correction, DC-to-AC inverters, DC and AC motor drives, etc. The insights gained in the previous chapter
will be used here to design the drive circuits for the controlled switches. The discussion in this chapter will
focus mainly on drive circuits for fully controlled power switches. Two of the generic power switch
drive circuits discussed in this chapter are BJTs and MOSFETs. However it should be noted that the drive
circuits for MOSFETs are directly applicable for IGBTs too. These controlled power switches account for
80–90% of the power electronic applications. In most applications MOSFETs and IGBTs are the more
popular and preferred high-power switches. Their gate drives are in general composed of stages of
BJT-based switches.
2.1 Transistor Drive Circuits
F igure 2.1 shows the typical collector–emitter voltage and collector-current waveforms for a switching
power transistor. The turn-ON and turn-OFF intervals are indicated in Figure 2.1. The switching behavior
of the transistor during these two intervals and the dependency of the turn-ON and turn-OFF behavior on
the transistor base drive are discussed in the following sections.
Chapter 02.indd 55 3/12/2009 4:02:22 PM

Vce
ic
Turn-ON Turn-OFF
interval interval
Figure 2.1 Typical collector-current and collector–emitter voltage waveforms of a transistor.
Turn-ON Behavior
The portion of the waveform corresponding to the turn-ON of the transistor (shown in Figure 2.1) is shown
in Figure 2.2 with an expanded timescale for various base-current waveshapes. A typical set of voltage and
current waveforms at the collector and base of a transistor during the turn-ON interval is depicted in
Figure 2.2(a). One should observe that during transistor turn-ON, a large collector-current spike, as indicated
in Figure 2.2(a), is generated. Such waveforms are found in a power-converter circuit wherein a capacitance
(parasitic or otherwise) is discharged at transistor turn-ON.
Figure 2.2(b) shows the turn-ON situation for a base-current waveshape that has a faster rate of rise. It
can be noted that here the peak and the average values of the turn-ON dissipation are smaller than that
shown in Figure 2.2(a). Figure 2.2(c) shows the effect on the transistor turn-ON for a very fast rising
base-current pulse which initially overshoots the final (steady-ON) value. The turn-ON dissipation is much
lower and narrower than the cases indicated in Figures 2.2(a) and 2.2(b). From Figure 2.2, it is evident that
for the power transistor, the turn-ON conditions are most favorable when the driving base-current pulse has
a fast leading edge and overshoots the final value or steady-ON value of the base current.
Turn-OFF Behavior
The portion of the waveform corresponding to the turn-OFF of the transistor (Figure 2.1) is shown in
Figure 2.3 with an expanded timescale. For the transistor to turn-OFF faster, it is essential that a negative
base-current drive be provided as indicated in Figure 2.3. The turn-OFF dissipation pulse is dependent on
both the transistor turn-OFF time and the collector-current waveshape during turn-OFF. One should note
that the major portion of the turn-OFF dissipation is during the fall time tf and the dissipation during the
storage time ts is negligible.
Chapter 02.indd 56 3/12/2009 4:02:23 PM

Drive Circuits 57
Vce Vce ic Vce

ic ic
ib ib ib
0
ic × Vce
ic × Vce ic × Vce
0
Turn-ON Turn-ON Turn-ON
interval interval interval
(a) (b) (c)
Figure 2.2 Expanded turn-ON interval to show the effect of base current
on the switching characteristics.
ic
0 Vce
0
ib
ic × Vce
ts tf
Turn-OFF interval
Figure 2.3 Expanded turn-OFF interval.
Chapter 02.indd 57 3/12/2009 4:02:23 PM

Characteristics and Classification of the Drive Circuits

Figure 2.4 shows the base drive requirement for a power transistor switch. During the turn-ON period, the
base current should be fast rising and must overshoot the final or steady-ON value. The maximum value of
the base current during turn-ON is denoted by ib +. During the turn-OFF period, a negative base current
(with a maximum value of ib –) should be provided to quickly remove the stored charges in the transistor as
explained in Chapter 1. During the steady-ON period, the base-current value is decided by the expected
collector current and the minimum b (hFE(sat) or hFE(min)) of the transistor. The steady-ON state base current
is denoted by ibon. During the steady-OFF period, the base-current value should be zero. Thus, a transistor
drive circuit should be designed such that the base-current waveshape (as shown in Figure 2.4) is realized.
A rule of thumb that has been found successful in most practical circuits is to take ibon = 2 × ic/hFE(sat); ib + =
1.5(i bon) and ib – = 2(ibon).
A good BJT base drive circuit must be able to:
1. supply adequate positive base current during turn-ON, that is ib +;
2. supply adequate base current to maintain the transistor in the steady-ON state, that is ibon;
3. supply negative base current for proper turn-OFF of the transistor, that is, ib –.
Moreover, depending on the nature of the application, the drive circuit should:
1. provide isolation between the control signals used for the base drive and the power signals that are being
controlled;
2. include protection against high dv/dt or surge voltage and overcurrents.
Thus, a base drive circuit can be classified as
1. drive circuit without isolation;
2. drive circuit with transformer isolation;
3. drive circuits with opto-isolation;
ib+
ib
ibon
ib−
Steady-ON
period
td + tr ts + tf
Turn-ON Turn-OFF
interval interval
Figure 2.4 Base drive requirement.
Chapter 02.indd 58 3/12/2009 4:02:23 PM

Drive Circuits 59
4. drive circuits without negative voltage source;

5. drive circuits with negative voltage source;
6. drive circuits with various combinations of the above.
Few BJT drive circuits examples are discussed in the following sub-sections.
BJT Drive Circuit-1

This base drive circuit is shown in Figure 2.5. This is the simplest of all the base drive circuits. In fact, it is
the basic building block of many other complex base drive circuits. For low-power and low-frequency
applications, this base drive circuit is well suited. This circuit does not provide the starting surge base
current ib + during the turn-ON time. The resistor R2 provides the discharge path for removing the
base charges in the transistor Q. As a rule of thumb, it is found that ibon = 2(ic/hFE(sat)) and iR2 = ibon gives
satisfactory performance. Thus
Vbe(sat) Vbe(sat) hFE(sat)
R2 = = (2.1)
iR2 2ic
Vb − Vbe(sat) (Vb − Vbe(sat) )hFE(sat)

R1 = = (2.2)
2ibon 4ic
BJT Drive Circuit-2

This drive circuit is shown in Figure 2.6. Here, Q p is the power transistor to be controlled. Q 1 and Q 2 are
auxilliary low-power transistors. This circuit is useful when the sourcing capability of Vb is low. The two
transistor stages, Q 2 and Q 1, provide the necessary current gain to suitably drive Q p. One can notice from
Figure 2.6 that the drives for the intermediate transistors Q 1 and Q 2 are based on the basic BJT drive
circuit-1.
When Vb is positive, Q 2 turns ON. As a consequence, collector-to-emitter voltage of Q 2 is almost zero.
As R4 is now connected to a ground at one end, the emitter–base junction of the PNP transistor Q 1 is
forward-biased and therefore Q 1 turns ON. This pumps the necessary base drive current through the resistor
R1 for Q p to turn-ON. When Vb is zero, then Q 2 is OFF and the base of Q 1 is high, thus making Q 1 OFF.
This will cut off the base drive for Q p and thus switch-OFF Q p.
ic
Vb R1 ib
Q
R2
Figure 2.5 BJT drive circuit-1.
Chapter 02.indd 59 3/12/2009 4:02:23 PM

Vcc
R3
Q1
R4
R1 ic
Vb R5 Qp
Q2
R2
R6
BJT Drive Circuit-3

The base drive circuit shown in Figure 2.7 is similar to the BJT drive circuit-2 that is shown in Figure 2.6. Here,
the turn-OFF time of the power transistor Q p is improved by using an inductor L as depicted in Figure 2.7.
When the transistor Q 1 is conducting, a current is flowing through L and some energy gets stored in the
inductor. The current in the inductor is limited to ib – by the resistance R2. When Q 1 is cut off, the inductor
will not allow sudden change in its current, and therefore the voltage polarity across the inductor will
Vcc
R3
ic
Q1 R1
Qp
R4
R2
R5
Vb
Q2
L
R6
Chapter 02.indd 60 3/12/2009 4:02:23 PM

Drive Circuits 61
reverse. This makes the inductor to act as a generator which will now supply the reverse base current ib – for
fast turn-OFF of Q p. The value of L in μH is calculated using the following relationship:
( R1 + R2 ) ⋅ ib − − Vbe(sat)
L= (2.3)
dib / dt
where dib/dt is taken as a value between 0.15ic A/μs for high-voltage transistors (>700 V) to 0.5ic A/μs for
low-voltage transistors (<200 V).
BJT Drive Circuit-4

This base drive circuit is shown in Figure 2.8. This drive circuit is similar to the base drive circuit of
Figure 2.6. But in this case, the resistor R 2 is connected to a negative supply to improve the turn-OFF speed
of the power transistor Q p. The operation is exactly similar to that of the base drive circuit-2, but here during
the turn-OFF of Q p, a negative base current ib – is provided by the negative source through the resistance R 2.
The resistance R2 is decided by
Vbe(sat) + Vcc (Vbe(sat) + Vcc ) ⋅ hFE(sat)
R2 = = (2.4)
ib − 4ic
where ib – is chosen as 2 times ibon.
BJT Drive Circuit-5

This base drive circuit is shown in Figure 2.9. This drive circuit provides a fast turn-ON and turn-OFF of
Q p. This is because the circuit provides a surge base current ib + during turn-ON of Q p and also a negative
base current ib – during turn-OFF of Q p.
When Vb is zero, the PNP transistor Q 3 is ON. This makes the base node of Q 1 and Q 2 positive. This will
forward bias the base–emitter of Q 1 and reverse bias the emitter–base of Q 2. Thus, Q 1 is turned ON and Q 2
Vcc
R3
Q1
R4
ic
R1
Vb R5
Q2 Qp
R6 R2
−Vcc
Chapter 02.indd 61 3/12/2009 4:02:24 PM

Vce
R4
Q3
R2 C ic
Q1
R5
Qp
Vb R1
Q2
R3
is turned OFF. When Q 1 is turned ON, a surge current ib + is delivered to the base of Q p through the R2–C
branch. This base surge current will induce fast turn-ON of Q p. During the time when the power transistor
Q p is fully ON, the capacitor would have charged to Vcc. When Vb is made positive, Q 3 is turned OFF and
this makes the base node of Q 1 and Q 2 zero. The emitter of Q 2 is positive because of the voltage of Vcc on C.
Thus, the emitter–base node of Q 2 is forward-biased and will turn ON Q 2. If emitter–base of Q 2 is forward-
biased, the base–emitter of Q 1 will be reverse-biased automatically. Thus, Q 1 will be turned OFF. The
capacitor C, which was charged to Vcc, will discharge through Q 2 providing the reverse base current ib –,
thus enabling fast turn-OFF of Q p.
The turn-ON base current surge ib + will flow through R2–C branch. The steady-ON base current ibon of
the transistor Q p will flow through R1. Thus,
Vcc − Vce(sat)(Q1) − Vbe(sat)(Qp)
R1 = (2.5)
ibon
where ibon = 2(ic/hFE(sat)); R2 is chosen as 1/5 times R1.

Now
τ ≈ R2C (2.6)
Set t equal to the turn-OFF time of Q p, that is toff = ts + tf . Then,
t off
C= (2.7)
R2
BJT Drive Circuit-6

This base drive circuit is shown in Figure 2.10. This drive circuit is similar in operation to the drive circuit-5
shown in Figure 2.9. Here the R2–C branch is removed and the negative base current during turn-OFF of
Q p is provided by the negative supply as shown in Figure 2.10. In this base drive circuit, there is no turn-ON
base surge current ib +. However, the negative base current for fast turn-OFF is given by (Vcc + Vbe(sat))/R1.
Chapter 02.indd 62 3/12/2009 4:02:24 PM

Drive Circuits 63
Vcc
R3
Q3
ic
Q1
R4
Vb Qp
R1
Q2
R2
−Vcc
One may recall that if a large base drive is injected, then the base charges are more and as a consequence the
turn-OFF time will be longer. If, however, the transistor is allowed to be just hovering at saturation, then the
stored base charge is small and the turn-OFF times will be very fast. But this will lead to higher ON-state Vce
of the transistor which in turn leads to larger ON-state losses. Therefore, if faster switching of the transistor is
required, the transistor should be biased such that it is just out of saturation. This is done by the use of an anti-
saturation circuit or the Baker’s clamp as shown in Figure 2.11. Referring to Figure 2.11, it is evident that
Vce = VD2 + Vbe(sat) − VD1 (2.8)
If the Baker’s clamp was not used, then on providing a base overdrive, the collector–emitter voltage will reduce
(to about 0.1–0.3 V). But by using the Baker’s clamp, when the collector–emitter voltage reduces, diode D1 con-
ducts and clamps the collector–emitter voltage to approximately Vbe(sat) value (assuming approximate cancella-
tion of drops due to D1 and D2). This means that Vce is never lower than about 0.7 V and thus the transistor Q p
is always at the edge of saturation. Therefore, the charges in the transistor will be reduced, thereby improving the
turn-OFF times. One should note that the diode D3 is used to provide a path for the reverse base current. This
anti-saturation circuit, that is, the Baker’s clamp can be used in any of the base drive circuits discussed.
D1 ic
D2
Qp
D3
Figure 2.11 Anti-saturation circuit or Baker’s clamp.
Chapter 02.indd 63 3/12/2009 4:02:24 PM

BJT Drive Circuit-7

This base drive circuit is shown in Figure 2.12. The base drives discussed until now are for a particular value
of collector currents. If the collector current (i.e., load) decreases, the transistor will be overdriven into
saturation for the same base drive. On the other hand, if the collector current increases, the transistor will
come out of saturation and operate in the active region. In applications where the load current can vary
within a wide range, the above problem can be overcome if the base current also varies in proportion to the
collector current. This concept is called the proportional-base drive. The base drive circuit indicated in
Figure 2.12 is a proportional-base drive circuit.
A current transformer (CT) is placed in the collector as shown in Figure 2.12. A short-duration positive
pulse (which is generally obtained by passing the base drive pulse through an R–C differentiator network) is
applied to the base of Q p. The energy in the base drive pulse is just sufficient to turn-ON the transistor Q p.
As a result, the collector current increases. Owing to transformer action, the collector current gets reflected
into the secondary of the CT and a current in2 will flow through the base of Q p. This increased drive will
further turn ON Q p which will, in turn, increase ic. In this manner, the transistor will go into saturation by
re-generative action. The turns ratio for the CT is chosen in the following manner:
ib 1 i n
= = n2 = 1 (2.9)
ic hFE(min) ic n2
One can note from Eq. (2.9) that the base current to the transistor Q p varies in proportion to the load
current, that is, the collector current. Therefore, the turn-ON and turn-OFF speeds will be fast at all loads.
To turn OFF the transistor, a negative pulse is applied as shown in Figure 2.12. This will bring the transistor
out of saturation and will cause the collector current to decrease. This will lead to re-generative turn-OFF of
the transistor. The transistor Q 1 is provided to demagnetize the core of the CT. During the time when Q p is
OFF, Q 1 is turned ON. This causes a current to flow through the winding n3 in a direction which will re-set
the core of the CT. The design of the CT is discussed in Chapter 7.
Vcc
R1
n3
Q1 n1
n2
ic
D1
Vb
Qp
Chapter 02.indd 64 3/12/2009 4:02:24 PM

Drive Circuits 65
BJT Drive Circuit-8

Until now, all the base drive circuits considered had no isolation between the base or control portion of the
circuitry and the load or the power portion of the circuitry. However, in many situations isolation between
the high-power side and the low-power base side of the circuit is essential. In such cases, the base drive
circuits will have to be designed taking isolation into account. Generally, isolation can be provided either by
using opto-couplers or by using transformers. The base drive circuit shown in Figure 2.13 is one such
isolated drive circuit which uses an opto-coupler to provide the isolation between the base side and the
high-power collector side circuits.
One can note that the base drive circuit portion in Figure 2.13 after the opto-isolation is exactly similar
to the base drive circuit-5 shown in Figure 2.9. Therefore, the turn-ON and turn-OFF operations of the
power transistor Q p are similar. A positive base drive pulse Vb causes a current to flow through the photodiode
of the opto-coupler. This causes the transistor Q 4 of the opto-coupler to turn ON. As a consequence, Q 3 is
OFF and Q 1 is ON as discussed for the base drive circuit-5 of Figure 2.9. Likewise, when Vb is zero, no current
flows through the photodiode and Q 4 is OFF. This causes Q 3 to turn ON. As a consequence, the power
transistor Q p turns OFF with reverse base current flowing through Q 2. One should note that the choice of
the opto-coupler depends on the current transfer ratio of the opto-coupler, the propagation delay for turn-
ON and turn-OFF and the breakdown voltage between the opto-coupler diode and transistor terminals.
BJT Drive Circuit-9

This is a transformer-isolated base drive circuit and is shown in Figure 2.14. When a positive voltage Vb is
applied to Q 1, the transistor Q 1 will turn ON and the full voltage of Vcc is applied across the primary of the
transformer. The transformer is a pulse transformer designed for the specific frequency and it passes the
primary pulse to the secondary which turns ON Q p. When the transistor Q 1 is turned OFF, the current in
the primary will be suddenly cut off and will give rise to a large voltage (of Ldi/dt) across the primary. This
will also appear across Q 1 (by Kirchhoff’s voltage law) and will damage the transistor. Therefore, to prevent
the abrupt cut off of the primary current when Q 1 turns OFF, a freewheeling path through Df and R3 is
provided. (The design of the transformer is discussed in Chapter 7.)
The ratio of the time for which the power transistor Q p is ON to the switching time period is called the
duty cycle or the duty ratio, D. The base drive circuits 1–8 discussed until now can be used for duty cycle range
Vcc
R4
R7
ic
R8 Q1 R2 C
Vb R5
Q3
Qp
R1
R6
Q4 Q2
R3
Chapter 02.indd 65 3/12/2009 4:02:25 PM

Vcc ic
D1
Qp
R1
Df
R2
R3
Vb R4
Q1
R5
from 0 to 1. But in this case, the duty cycle is limited because the transistor has to be OFF for the period of time
when the core is being re-set by freewheeling action. If the value of R3 is large, then the core re-setting is faster
and therefore the range of duty cycle is larger as Q p needs to be compulsorily OFF only for a smaller time. How-
ever, in such a case, as the drop across R3 is higher, the Vceo rating of Q 1 is higher which is evident from
Vceo(Q1) > Vcc + I mag R3 + VDf (2.10)
where Imag is the magnetizing current in the primary at the instant when Q 1 is being turned OFF. Therefore,
to have a reasonable Vceo rating for Q 1, the value of R3 cannot be chosen too large. In practice, the duty ratio
is limited to less than 0.5 (50%).
BJT Drive Circuit-10

This base drive circuit is shown in Figure 2.15. This is exactly similar to the base drive circuit of Figure 2.14.
The only difference lies in the freewheeling circuitry. In Figure 2.14, the freewheeling path contained
Vcc
ic
D1
n2
Qp
R1
n1 n3 R2
Vb R3
Df
Q1
R4
Chapter 02.indd 66 3/12/2009 4:02:25 PM

Drive Circuits 67
Vcc R2 C ic
D1 D2
Qp
R1
Df
Q1
R4 R3
Vb R5
Q2
R6
a resistor R3 which leads to dissipation. In Figure 2.15, the dissipation is avoided during freewheeling. When
Q 1 turns OFF, the dot poles of the transformer become negative with respect to the other pole. This causes
Df to be forward baised. Df conducts and freewheels the magnetic energy stored in the core. This causes the
required core re-setting which prevents core saturation. As winding n2 is used for demagnetizing the core,
this winding is also called the demagnetizing winding. In practice, to achieve a very tight coupling (i.e., low
leakage) between n1 and n2, these two windings are wound bifilar. Therefore, the turns ratio n1:n2::1:1 is
maintained. As a consequence, a time equal to the ON time of Q 1 is required for the core to re-set. Therefore,
the duty cycle in this case cannot exceed 0.5 (or 50%).

This is a transformer-isolated base drive circuit as shown in Figure 2.16. This circuit is similar to the transformer-
isolated base drive circuits of Figures 2.14 and 2.15. But here, the turn-OFF time of Q p is improved by using
an active turn-OFF circuit with the help of a PNP transistor Q 1.
When Vb is positive, the transformer secondary is positive at the dot pole. This will forward bias the
diodes D1 and D2. The drop across D2 will ensure that the emitter–base junction of Q 1 is reverse-biased. Thus,
Q 1 is in the turned-OFF condition. The R2–C will provide a low-impedance path during pulse transitions
which will provide the needed surge base current for fast turn-ON, that is ib +. When Q p is to be turned OFF,
Vb is made zero. Then, D1 will be reverse-biased and the base of Q 1 is pulled low. Owing to the presence of
the capacitor C, the emitter of Q 1 will now be positive which ensures that the emitter–base junction of Q 1 is
forward-biased. This causes Q 1 to turn ON, thereby providing the necessary reverse base current for fast
turn-OFF of Q p. One should note that here also the duty cycle is limited to 0.5.

This base drive circuit is shown in Figure 2.17. This is an isolated proportional-base drive circuit. As discussed
in the case of the non-isolated proportional-base drive circuit of Figure 2.12, there is a CT (nc – nb windings)
which provides the necessary base current to maintain the power transistor Q p in the ON-state. A flyback
tranformer with air gap (np – nb windings) is used to supply the required turn-ON and turn-OFF pulses to
start the turn-ON and turn-OFF re-generative processes, respectively, for Q p.
Chapter 02.indd 67 3/12/2009 4:02:25 PM

Vcc
nc
ic
R1
Qp
C
np nb
Vb R2
Q1
R3
The operation of the base drive circuit of Figure 2.17 is similar to the non-isolated proportional-base drive
circuit. Consider the situation where Q p is in the ON condition. During this time, Q 1 is OFF. The capacitor
C charges to Vcc through R1 with a time constant of R1C. Now, when Q 1 is turned ON, the secondary voltage
will be the same as the base–emitter voltage of Q p which is around 0.7 V because as yet Q p is not turned OFF
due to the presence of stored charges in it. Therefore, the primary of the transformer will experience a virtual
short circuit. A large current will be discharged from C. This will be reflected at the secondary as a large
negative base current which will quickly and re-generatively turn OFF Q p.
During the time when Q p is OFF and Q 1 is ON, the primary current will rise and saturate the core. The
resistance R1 will limit the current through Q 1. The dot poles of the transformer are positive and therefore
Q p is maintained in the OFF-state. Now, if Q 1 is turned OFF, the voltage polarities across the windings
reverse due to inductance action and the stored energy in the core (air gap) will freewheel through the base
of Q p. This will provide the turn-ON energy to start the re-generative process for turn-ON of Q p.
The choice of nc:nb will depend on the hFE(min) of the power transistor Q p. Thus
nb ic
= = hFE(min) (2.11)
nc ib
The choice of np:nb will depend on the base–emitter breakdown voltage of Q p. This is because, when Q p is
OFF, the secondary voltage should not exceed the base–emitter breakdown voltage BVebo of Q p. Thus,
np Vcc
= (2.12)
nb BVebo
Chapter 02.indd 68 3/12/2009 4:02:25 PM

Drive Circuits 69
When Q 1 is ON, the current through the primary increases till it saturates the core. This saturation current
ipon, which is limited by R1, should provide the necessary turn-ON base drive current ib + for Q p. Thus,
⎛n ⎞
ipon = ib+ ⎜ b ⎟ (2.13)
⎝ np ⎠
Vcc
R1 = (2.14)
ipon
When Q 1 is turned ON in order to turn OFF Q p, the primary current ipoff , consists mainly of the reflected
collector current of Q p and the reverse base drive current ib – of Q p. Thus,
⎛n ⎞ ⎛n ⎞
ipoff = ic ⎜ c ⎟ + ib − ⎜ b ⎟ (2.15)
⎝ np ⎠ ⎝ np ⎠
This current of ipoff has to be supplied by the capacitor when Q 1 is turned ON. Thus, the capacitor should
have energy of at least
1
E c = CVcc2 = Vcc ⋅ ipoff ⋅ t off (2.16)
2
where toff is the turn-OFF time of Q p which is equal to (ts + tf ) of Q p. From Eq. (2.16), the value of
capacitor C is given by
2 ⋅ ipoff ⋅ t off
C= (2.17)
Vcc
One should note that when Q p is ON, the capacitor should charge to Vcc during this time. The charging
time constant is R1C. The capacitor will charge fully to Vcc in 5 R1C. This means that Q p should be on for at
least 5 R1C. Therefore, there is a minimum duty cycle limitation in this circuit. Thus, the duty cycle ranges
from 5R1C/Ts to almost 1, where Ts is the switching period.

This drive circuit is shown in Figure 2.18. This is almost identical to that shown in Figure 2.17. The only dif-
ference is that the charging time constant of C is made faster here by using a transistor Q 2. It can be shown
that the capacitor charging time constant is given by
R1C
τ= (2.18)
hFE(Q2)
It is evident from Eq. (2.18) that the charging time constant is significantly reduced if Q 2 is a high hFE
transistor. The hFE of Q 2 is generally chosen such that t is half of the turn-ON time of Q p that is (tr + td).
With this circuit, the duty cycle range is increased from almost 0 to almost 1.
Chapter 02.indd 69 3/12/2009 4:02:25 PM

Vcc
nc
R1 ic
Q2
Qp
D np nb
C
Vb R2
Q1
R3
2.2 MOSFET Drive Circuits
T he MOSFETs are voltage-controlled devices. As a consequence, the gate currents are not dependent on
the drain currents. The gate power required to maintain the MOSFETs in the ON condition is negligible.
This section discusses primarily the MOSFET drive circuits; however, as discussed in Chapter 1, the gate
portion of the IGBTs being functionally similar to the MOSFETs, these drive circuits that will be discussed
can also be applied for driving IGBTs.
Figure 2.19 shows the gate drive requirements for a MOSFET. It can be observed from the gate drive
requirements that the gate current required to maintain the MOSFET in the steady-ON condition is zero.
Therefore, the gate power required to maintain the MOSFET in the steady-ON condition is low. The gate
circuit energy is used only to turn-ON and turn-OFF the MOSFET. During turn-ON, a peak current of ig +
is applied which is used to turn-ON the MOSFET. During turn-OFF, a negative peak current of ig– is
provided for fast turn-OFF of the MOSFET.
During the turn-ON period (ton), it can be assumed that an equivalent constant current of igon is being
applied to the gate of the MOSFET. One can approximate the turn-ON surge current to be a right-angled
triangle with the peak of ig +. Thus
ig+
igon = (2.19)
2
The turn-ON and turn-OFF operation of the MOSFET is as explained in Chapter 1. To turn-ON a
MOSFET, a specific amount of gate charge Q G has to be supplied to the gate of the MOSFET. This amount
of Q G for a specific MOSFET is given in the manufacturers’ datasheets. Thus
Chapter 02.indd 70 3/12/2009 4:02:25 PM

Drive Circuits 71
ig+
ig
igon
ig−
Steady-ON
period
ton toff
Turn-ON Turn-OFF
interval interval
Figure 2.19 Gate drive requirement for MOSFETs.
Q G = igon t on (2.20)
It is evident from Eq. (2.20) that if one requires to turn-ON the MOSFET faster (i.e., smaller ton), the igon
required should be more. If the MOSFET can be switched slower, then a smaller igon would suffice. Consider
a MOSFET where 250 nC of charge is required to turn-ON the MOSFET. For a specific application, if it
is required that the MOSFET should be turned-ON in 1 μs, then the igon required would be 250 mA and ig +
required would be 500 mA. If on the other hand, the required turn-ON time is 2 μs, then igon required
would be 125 mA and ig + required would be 250 mA.
To allow a turn-ON gate drive surge current of ig +, only a limiting resistor R1 can be connected in series
with the gate of the MOSFET such that
Vcc
R1 = (2.21)
ig+
where Vcc is the gate drive supply voltage.
MOSFETs generally require a gate voltage of 15 V. Therefore, Vcc = 15 V in most cases. Various
MOSFET gate drive circuits will be now discussed in the following sections.
MOSFET Drive Circuit-1

The MOSFET drive circuit shown in Figure 2.20 is the simplest MOSFET drive circuit. The resistor R1
is used for limiting the turn-ON surge current and it is chosen based on Eq. (2.21). If the source Vg does
not have current-sinking capability, a resistor R2 is connected as shown and is used to provide a discharge
path for the input capacitance during turn-OFF of the MOSFET. R2 is usually chosen 10 times the value
of R1.
If the gate drive source Vg is capable of sinking current, then R2 need not be used. In such a case, as the
output resistance of the source Vg is low, the input capacitance will charge and discharge through R1.
Chapter 02.indd 71 3/12/2009 4:02:26 PM

id
Q
Vg R1
R2
Figure 2.20 MOSFET drive circuit-1.

MOSFETs can be driven directly from a complementary metal oxide semiconductor (CMOS) logic integrated
circuit (IC) as shown in Figure 2.21. For faster switching, igon should be increased. This is achieved by parallel-
ing buffers as indicated in Figure 2.21. Either the CMOS hex inverting buffer IC or non-inverting hex buffer
IC can be used to drive the MOSFETs. The input gate pulse signal source, Vg should have a polarity that
should be appropriate with respect to the type of buffer chosen, whether inverting or non-inverting. It should
be noted that here also as in all the MOSFET gate drive circuits that are discussed, the series resistor R1, which
is used to limit the turn-ON gate surge current, is selected based on Eq. (2.21).
id
Q
Vg R1
Chapter 02.indd 72 3/12/2009 4:02:26 PM

Drive Circuits 73
Vcc
R3
id
Q3
Q1
Qp
R4 R1
Vg
Q2
R2

The MOSFET drive can be made faster by using transistors to increase the turn-ON gate drive current, igon.
The MOSFET drive circuit shown in Figure 2.22 is similar in topology and operation to the BJT drive circuit
shown in Figure 2.10. The only difference is that the negative power supply is not used here. In Figure 2.10,
the negative power supply was used to provide the forward bias for Q 2 while turning OFF Q p. Here, the
input capacitance of the MOSFET provides the necessary forward bias to turn ON the PNP transistor Q 2
when Q 1 is turned OFF in order to switch-OFF the MOSFET Q p. This would imply that a negative power
supply is not needed.
However, if a lower voltage negative power supply Vss (~–5V) is used wherein the emitter of Q 2 is con-
nected to the negative pole of the Vss, then Q 2 would be ON even when the gate–source MOSFET capacitor
is fully discharged. This would ensure that the gate–source is connected by R1 which is almost like a short as R1
is a small resistance value. This would also ensure greater reliability in the sense that no spurious dv/dt on the
high-power drain side circuit will charge up the gate–source capacitance through the drain–gate capacitance
due to the fact that the gate is connected to source through R1 and Q 2.

Some circuits demand that the gate and source terminals of MOSFETs are floating with respect to those of
other MOSFETs in the circuit. This is typical of bridge topologies wherein the top MOSFET and the bottom
MOSFET need different drive grounds. Isolated drive to these MOSFETs can be provided either by using
opto-isolators or by using pulse transformers. Figure 2.23 shows a gate drive circuit using an opto-isolator.
This circuit is similar in topology and operation to the BJT opto-isolated circuit shown in Figure 2.13. One
should note that for all opto-isolated drive circuits whether BJT or MOSFET drive circuits, the secondary-side
circuitry (transistor side) should have a local power supply which is isolated from the primary-side circuitry
(diode side).
When Vg is positive, the diode in the primary side of the opto-isolator emits photons. These photons
impinge on the base region of the transistor in the secondary side. This causes photogeneration of carriers
Chapter 02.indd 73 3/12/2009 4:02:26 PM

Vcc
R4
R7 ic
Vg R8 R5 Q1 Qp
Q3 R1
R6
Q4 Q2
R3
sufficient to satisfy the base requirement for turn-ON of Q 4. When Q 4 turns ON, Q 3 is turned OFF. This
causes Q 1 to turn ON which supplies the required gate current to the MOSFET through R1 and turns ON
the MOSFET Q p.

The gate drive circuit shown in Figure 2.24 is another opto-isolated circuit. Here, the complementary pair
Q 1 and Q 2 shown in Figure 2.23 is replaced by CMOS buffer ICs either inverting or non-inverting type as
required by the polarity of Vg.
Vcc
id
R2 Qp
R5
R1
Vg R6 R3
Q1
R4
Q2
Chapter 02.indd 74 3/12/2009 4:02:26 PM

Drive Circuits 75
Vcc = 15 V
Vdclink = 300 V
D
R4 id
C
R2
Q1 Qp1
R1
R3
Q2
Vg R5
Q3 gnda
R6 Qp2
gndb

In most inverter and converter applications, the power converter is generally a half-bridge or a full-bridge config-
uration. The top MOSFET and the bottom MOSFET are switched at mutual exclusive time intervals to prevent
shorting the DC link or bus. In either case, the gate drive for the high-side MOSFET is generally isolated from
the low-side MOSFET by the use of opto-isolators or pulse transformers. Figure 2.25 gives a circuit wherein the
high-side and the low-side MOSFETs can be driven without the use of opto- or transformer isolation.
When Vg is zero, both Q 3 and Q 1 are OFF. Therefore, the MOSFET Q p1 is also in the OFF-state. But
during this period, the bottom MOSFET Q p2 is ON. The capacitor C will get charged to Vcc through the
diode D and the bottom ON MOSFET Q p2. When Q p2 is turned OFF and Vg is positive, Q 3 is ON, which
in turn switches Q 1 ON. The capacitor C now discharges through Q 1 into the input capacitance of the
MOSFET Q p1 thereby turning ON Q p1. When Vg is made zero again, Q 1 is turned OFF and Q 2 is turned
ON. The input capacitor of the MOSFET discharges through R1 and Q 2, thereby turning Q p1 OFF.
As the capacitor C needs to charge to Vcc during the time when Q p2 is ON, there is a minimum ON-time
limitation on Q p2. It should also be noted that when Q p1 is ON and Q p2 is OFF, the Q p1 drive capacitor’s nega-
tive pole which is connected to the source of Q p1 is at the DC-link voltage. At this time the diode D will have a
reverse voltage of (Vcc – VC – Vdclink ) and therefore should be rated to withstand the full DC-link voltage.

A transformer-isolated gate drive circuit is shown in Figure 2.26. One can note that this gate drive circuit is
exactly similar both in topology and operation to the BJT transformer-isolated base drive circuit shown in
Figure 2.14. The operation of the pulse transformer and the freewheeling circuit is similar to that discussed
for the transformer-isolated BJT base drive circuit. Here again, one can use a demagnetizing winding to perform
non-dissipative freewheeling as indicated in Figure 2.15 for the BJT base drive.
Chapter 02.indd 75 3/12/2009 4:02:26 PM

id
Vcc Qp
D1
Df R1
R2
R3
Vg R4
Q1
R5
id
Qp
D1 D2 R1
Vcc
Df
R3 Q1
R4
Vg R5
Q2
R6
For the MOSFET gate drive circuits discussed till now, the upper limit on the duty cycle is 1 (i.e., 100%).
But in the case of this drive circuit, the upper limit on the duty cycle is 0.5, that is, 50% only.

Another transformer-isolated gate drive circuit is shown in Figure 2.27. Here, the turn-OFF is due to active
pull down by the PNP transistor Q 1. This will, therefore, provide a faster turn-OFF of the MOSFET than
Chapter 02.indd 76 3/12/2009 4:02:27 PM

Drive Circuits 77
that provided by the gate drive circuit shown in Figure 2.26. When Vg is positive, Q 2 is turned ON. This
causes the dot poles of the pulse transformer to be positive with respect to the other poles. D1 and D2 will be
forward-biased and charge the input capacitor of the MOSFET and turn-ON Q p. During this time, as D2
is ON, the emitter–base junction of Q 1 is reverse-biased and therefore Q 1 is OFF. When Vg is made zero, Q 2
turns OFF. This causes the dot poles of the pulse transformer to become negative with respect to the other
poles. This will reverse bias D1 and D2, thereby switching OFF these diodes. As a consequence, the base of
Q 1 is pulled low through R3. The charge on the input capacitance of the MOSFET will make the emitter of
Q 1 positive with respect to the base and will therefore forward bias the emitter–base junction of Q 1 and turn
it ON. The input capacitance will then discharge through Q 1, thereby turning Q p OFF. One should note
that here also the upper limit for the duty cycle is 0.5, that is, 50%.

To obtain the full duty cycle range of 0 to 1 using transformer isolation, one can use the transformer-
isolated gate drive circuit shown in Figure 2.28. Here, two pulse transformers are used and they are
id
Vcc Qp
D1 R1
Df
R2
R3
T1
R6
Q1
R7
HF Carrier
osci.
Vcc
Carrier D2 R4
Vg gating
circuit
Df1
R5
T2
R8
Q2
R9
Chapter 02.indd 77 3/12/2009 4:02:27 PM

switched with 180° phase difference. The secondaries of the transformers are diode ORed to obtain the
desired gate pulse.
The gate drive pulse Vg is gated with a high-frequency carrier as indicated in Figure 2.28. Two switching
patterns are generated to switch the two pulse transformers. One signal is obtained by directly gating the
gate drive pulse Vg with the high-frequency carrier. The other signal is obtained by gating the gate drive
pulse Vg with the inverted high-frequency carrier. These two signals are used to switch the two pulse trans-
formers whose secondaries are diode ORed. In this manner, whatever be the duty cycle of the gate drive
waveform Vg, the transformers are always switching at 50% duty cycle thereby avoiding transformer core
saturation. In this manner, the duty cycle range of the MOSFET can be extended to 100%. One should
note that, in this case, the series resistor for the MOSFET is now connected in series with each transformer
before the ORing node. This is used to avoid any large circulating currents that may occur during turn-OFF
of one pulse transformer and turn-ON of the other pulse transformer.

A relatively simple transformer-isolated gate drive circuit with large duty cycle range capability (of 0 to
almost 100%) is shown in Figure 2.29. Here, when Q 2 is switched ON, the dot poles are positive with
respect to the other poles. This will charge the input capacitance of Q p through the body diode of the
MOSFET Q 1. When Q 2 is switched OFF, the dot poles become negative with respect to the other poles
and will switch ON Q 1. The input capacitance of the MOSFET Q p will discharge through Q 1 and thereby
turn-OFF Q p.
One should note that as MOSFETs do not need any gate current to maintain it in the ON-state, the
pulse transformer should be designed to transfer energy just sufficient to turn-ON Q p when Q p is required
to be turned ON and energy just sufficient to turn-ON Q 1 when Q p is required to be turned OFF. As a
consequence, the pulse transformer size will be small. The resistor R2 is used to limit the primary current of
the transformer when the transformer goes into saturation. However, one should note that in this circuit,
Vcc
id
Qp
R2 Q1
R1
Vg R3
Q2
R4
Chapter 02.indd 78 3/12/2009 4:02:27 PM

Drive Circuits 79
when Q p is OFF, the gate of Q p is floating. This will leave the MOSFET open to Miller turn-ON. Any large
dv/dt spikes on the drain side can easily charge up the gate–source capacitance through the drain–gate
capacitance and turn-ON Q p at an undesirable time.

This gate drive circuit, which is shown in Figure 2.30, provides inherent protection against overcurrents
through the drain of the MOSFET. The protection circuit discussed here can be used with any of the gate
drive circuits discussed till now. Here, a transistor Q 2 is used as a series-pass element which is normally ON.
When Q 2 is ON, the gate drive circuit will operate in the normal manner. If a large drain current flows
through the MOSFET when it is ON, then the series-pass transistor Q 2 is immediately cut off, thereby turning
OFF the MOSFET.
When Q p is ON, the voltage across it is the ON-state drop which is idRDS(ON). When the drain current
increases beyond a certain limit, the ON-state voltage will increase such that the “+” input of the comparator
A1 will be greater than the “−” input of the comparator. This will cause the output of A1 to go high. This will
enable the AND gate G3 which will switch-OFF the PNP series-pass transistor Q 2 (output of G2 is high at
this time). The threshold limit for the ON-state voltage can be set by adjusting the reference voltage to
the “−” input pin of the comparator. During the turn-ON process, when Q p is to be turned ON, the voltage
across Q p will be high. This will cause the comparator output to be high which will, in turn, cause Q 2 to
be OFF. Therefore, any gate drive pulse will not be passed on to the gate of Q p to turn it ON. To circumvent
this starting problem, a small delay of 1–5 μs is provided at turn-ON during which time the gate G3 is forcibly
Vcc
R5
G2 +
G1
R6 A1
D2
−
C G3
R4
R3
id
Vref Qp
Vg Q2
D1
R1
Q1
R2
Chapter 02.indd 79 3/12/2009 4:02:27 PM

disabled by the delay circuit of G1 and G2 such that Q 2 is ON till the MOSFET Q p turns ON. After the
initial period of 1–5 μs, G2 will go high and enable G3. But now A1 output will be low as Q p is now fully
ON. A1 will now go high only when overcurrents occur which turns OFF Q 2. This in turn will turn OFF
Q p. The gates G1, G2 and G3 should be Schmitt gates in order to avoid meta-stability problems.
2.3 Snubber Circuits
I n most power semiconductor devices, there is a danger of exceeding the voltage and the current ratings of
the devices during the turn-OFF and turn-ON instants, respectively. As was discussed earlier in Chapter 1,
when the BJT is being turned OFF, the voltage across the device Vce is determined by the external cir-
cuitry. If there is an inductive load in the collector or if there is significant amount of lead inductance associ-
ated with the collector or emitter leads, then when the BJT is being turned OFF, the current through the
device will fall rapidly to zero in a time corresponding to the fall time of the device. As a consequence,
a large voltage spike due to Ldi/dt will occur across the device and cause the Vce of the BJT to have a large spike
during the fall time. This may damage the device. Therefore, it becomes essential to limit the voltage spike across
the device during turn-OFF such that the voltage is within the Vceo rating of the device. Similar argument
applies to MOSFET and IGBT switches too. In general, during turn-OFF, the power switch should be
protected against overvoltage stress. The turn-OFF voltage stresses are reduced by using circuits called the
turn-OFF snubber circuits.
On the other hand, during turn-ON of the device, due to the presence of any capacitive load or parasitic
capacitance across the switch, there will be a huge surge current through the device which could damage
the device. Therefore, it is essential to limit the current spike through the device during turn-ON such that the
device current is within the peak rating of the device. The turn-ON current stresses are reduced by using
circuits called the turn-ON snubber circuits.
The snubber circuits, in general, modify the device switching characteristics and in doing so, reduce the
device transient stress. In fact, the transient voltage and current stress during turn-OFF and turn-ON,
respectively, are transferred to the snubber circuits. The snubber circuit action involves temporary energy
storage in either an inductor or a capacitor. In re-setting these passive components, it is usual to dissipate the
stored energy in a resistor as heat. As a consequence, the circuits with snubbers will be less efficient. At high
frequencies, these losses may become a limiting factor because of the difficulties associated with equipment
cooling. Instead of dissipating the snubber energy stored in the inductance and capacitance, alternately one
may recover the energy either back into the supply or into the load. There are both passive and active circuits,
available in the literature, which perform this energy recovery from the snubber. However, here the basic
concepts of the snubber action will be illustrated with the generic turn-OFF and turn-ON dissipative snubber
circuits. The discussion of the snubber circuits which will follow can be equally applied to BJTs, MOSFETs
and IGBTs.
Turn-OFF Snubber or Shunt Snubber

Consider a power semiconductor switch Q which could be a BJT or a MOSFET or an IGBT as shown in
Figure 2.31(a). Let the load be an inductive load. When the device Q is turned OFF, the load current iL will
fall to zero in a period of time corresponding to the fall time tf . This will cause the voltage across the transis-
tor Vce to shoot up to a large value due to the Ldi/dt phenomenon where L could be the lead inductances
associated with the collector and emitter leads. Therefore, it is required to limit the voltage across the collector–
emitter of the transistor and it should be made to rise gradually during the period tf as indicated in
Figure 2.31(c). To slow down the change in the voltage across the device, one can connect a capacitor C
Chapter 02.indd 80 3/12/2009 4:02:27 PM

Drive Circuits 81
iL iL
iL D ic
icap icap
Q
ic Vce Vcc
C Q ic R C
tf
(a) (b) (c)
Figure 2.31 Turn-OFF or shunt snubber.
across the device as indicated in Figure 2.31(a). As the capacitor C is connected in shunt with the device,
this type of snubber is also called the shunt snubber.
Without loss of generality, one can assume that the load current is falling linearly as shown in Figure 2.31(c)
during the fall time and the collector–emitter voltage across the device is rising linearly as shown in
Figure 2.31(c) during the fall time when the shunt snubber is used. Referring to Figure 2.31(a),
iL = ic + icap (2.22)
where ic is the current through the device and icap is the current through the capacitor. During the fall time
period of tf , the current ic through the device is given by
⎛ t⎞
ic = iL ⎜ 1 − ⎟ (2.23)
⎝ tf ⎠
From Eqs. (2.22) and (2.23), the current through the capacitance is given by
⎛ t⎞ ⎛t⎞
icap = iL − iL ⎜ 1 − ⎟ = iL ⎜ ⎟ (2.24)
⎝ tf ⎠ ⎝ tf ⎠
The voltage vce across the device is the same as the voltage across the capacitor C. Therefore,
1
v ce =i dt (2.25)
C cap
Use Eq. (2.24) in Eq. (2.25) and integrate within the fall time period. Then, apply the boundary condition
that at the end of the fall time period, the voltage across the device or the capacitor should be Vcc, that is, the
supply voltage. The capacitor value C is then given by
iLt f
C= (2.26)
2Vcc
One should note that though the circuit in Figure 2.31(a) will solve the problem of voltage spike during
turn-OFF process of the device, it will create a serious problem during the turn-ON of the device Q. When
the device Q is turned ON again, the capacitor will discharge through Q and will result in a large current
Chapter 02.indd 81 3/12/2009 4:02:27 PM

spike through the device which can damage the device. Therefore, to limit current through the device,
a resistor R is introduced in series with C. The resistor R should provide the function of current limiting only
during turn-ON of the device. However, during turn-OFF of the device, R is not needed. Therefore, to
reduce the dissipation in R during turn-OFF, a diode is placed across R as shown in Figure 2.31(b) so that R
comes into effect only during turn-ON when C discharges through R and the device.
Selection of R
When Q is turned ON, it should carry the following currents: The capacitive current discharge from C
which is equal to Vcc/R and the load current iL. Therefore,
Vcc
+ iL < I cm (2.27)
R
where Icm is the maximum collector-current rating of the transistor.
Re-arranging the inequality in Eq. (2.27), the following inequality is obtained:
Vcc
R> (2.28)
I cm − iL
It is also important to ensure that the capacitor discharges fully before the next charging when the transistor
turns OFF. Therefore, there is a minimum duration of time during which time the transistor should remain
ON so that the capacitor can fully discharge. The discharge time constant of the capacitor is RC. In five
times this time constant, the capacitor will be almost fully discharged. Therefore
Ton-min > 5RC (2.29)
where Ton-min is the minimum time for which the device Q should remain in the ON-state.
Re-arranging inequality (2.29), the following inequality is obtained:
Ton-min
R< (2.30)
5C
From inequalities in Eqs. (2.28) and (2.30), the range for choice of R is given by
Vcc T
< R < on-min (2.31)
I cm − iL 5C
One should note that when C is being charged, an energy of CVcc2 /2 is dissipated in R and when C is being
discharged, an energy of another CVcc2 /2 is dissipated in R. Therefore, in all, an energy of CVcc2 is dissipated
in R. As the function of R is to limit the current through the device Q during turn-ON only it is bypassed
using a diode D as shown in Figure 2.31(b). In this case, power is dissipated in R only during capacitor
discharge time. The power dissipated in R is now given by
1
PR = CVcc2 f s (2.32)
2
where fs is the switching frequency of the device Q.
Chapter 02.indd 82 3/12/2009 4:02:28 PM

Drive Circuits 83
Vcc
Vce Vcc
D
VL
L
R
ic iL
Q Vce
tr
(a) (b)
Figure 2.32 Turn-ON or series snubber.
Turn-ON Snubber or Series Snubber

The turn-ON snubber is the dual of the turn-OFF snubber. Consider a power semiconductor switch Q
which could be a BJT or a MOSFET or an IGBT as shown in Figure 2.32(a). Let the load be a capacitive
load. When the device Q is turned ON, the voltage across the device (Vce) will fall to zero in a period of time
corresponding to the rise time tr. This will cause the current through the transistor ic to shoot up to a large
value due to the C(dVcc/dt) phenomenon (capacitive discharge) where C could be the load capacitance or
any parasitic capacitance between the collector and emitter leads. Therefore, it is required to limit the current
through the transistor and it should be made to rise gradually during the period tr as indicated in Figure 2.32(b).
To slow down the change in the current through the device, one can connect an inductor L in series with the
device as indicated in Figure 2.32(a). As the inductor L is connected in series with the device, this type of
snubber is also called the series snubber. This snubber is an exact dual of the shunt snubber. Referring to
Figure 2.32(a), diode D and resistor R provide the freewheeling path for the inductance when the transistor
Q is turned OFF, so that voltage spike during turn-OFF process is limited.
Without loss of generality, one can assume that the voltage across the collector–emitter of the device
Q is falling linearly as shown in Figure 2.32(b) during the rise time and the current through the device is
rising linearly as shown in Figure 2.32(b) during the rise time when the series snubber circuit is used. Referring
to Figure 2.32(a),
Vcc = Vce + VL (2.33)
where Vce is the voltage across the device and VL is the voltage across the inductor. During the rise time
period of tr , the voltage Vce across the device is given by
⎛ t⎞
Vce = Vcc ⎜ 1 − ⎟ (2.34)
⎝ tr ⎠
Chapter 02.indd 83 3/12/2009 4:02:28 PM

From Eqs. (2.33) and (2.34), the voltage across the inductor is given by
⎛t⎞
VL = Vcc ⎜ ⎟ (2.35)
⎝ tr ⎠
The current ic through the device is the same as the current through the inductor L. Therefore
1
ic = iL = V dt (2.36)
L L
Use Eq. (2.35) in Eq. (2.36) and integrate within the rise time period. Then apply the boundary condition
that at the end of the rise time period, the current through the device, that is the current through the inductor,
should be iL (the load current value). The inductor value L is then given by
Vcc t r
L= (2.37)
2iL
Selection of R
When Q is turned OFF, it should withstand the following components of the voltage:
1. The voltage across the freewheeling components, that is D and R. This is equal to iLR + VD, where VD
is the diode forward drop.
2. The supply voltage Vcc.
Thus the following inequality should be satisfied for the device voltage rating.
Vcc + iL R + VD < Vceo (2.38)
where Vceo is the maximum collector–emitter voltage rating of the transistor.
Vceo − Vcc − VD
R< (2.39)
iL
It is also important to ensure that the magnetic energy in the inductor L discharges fully before the next
charging when the transistor turns ON. Therefore, there is a minimum duration of time during which time
the transistor should remain OFF so that the inductor energy can fully discharge. The discharge time constant
of the inductor is L/R. In five times this time constant, the inductor will be almost fully discharged.
Therefore,
L
Toff-min > 5 (2.40)
R
where Toff-min is the minimum time for which the device Q should remain in the OFF-state.
5L
R> (2.41)
Toff-min
From inequalities in Eqs. (2.39) and (2.41), the range for choice of R is given by
5L V − Vcc − VD
< R < ceo (2.42)
Toff-min iL
Chapter 02.indd 84 3/12/2009 4:02:28 PM

Drive Circuits 85
The power dissipated in R is given by

1 2
PR = Li f (2.43)
2 L s
where fs is the switching frequency of the device Q.
In this chapter we have discussed the requirements just by handling. They must be place on anti-static
for driving the BJTs and the MOSFETs giving a pads with the gate and source/collector shorted. In
flavor for the various types of drive circuits. It addition, a protective 15 V Zener diode is connected
should, however, be noted that by no means is the between gate and source/emitter to provide a measure
list of discussed drive circuits exhaustive. There are of clamping if the static induced voltage exceeds 15 V.
many ICs and hybrid circuits available commercially However, once the MOSFETs and IGBTs are mounted
that implement the drives circuits of all the switches onto the printed wiring boards or connected to the
of either half-bridge or full-bridge or three-phase drive circuits, they are usually very robust.
full-bridge as a single device. These integrated devices The key to reliable power switch performance
in addition provide features like overcurrent protec- is to ensure that the locus of the operating point is
tion and thermal protection. If a fault occurs, a fault always within the forward and reverse safe operat-
output is provided that can be used as an interrupt ing areas (SOARs). Any transition of the operating
signal to disable the drive pulses to the bridge circuit point locus across the boundary of the SOAR will
used in a converter or inverter application. Whether cause overstressing of the semiconductor bulk and
one uses a discrete drive circuit or an integrated the junctions. This will cause fast aging of the
drive circuit device for a specific application depends device leading to deterioration and premature fail-
on the cost, size, isolation feature, protection features ure. Most failures of the power switches are due to
and interface compatibility to microcontrollers or non-restriction of the operating point locus within
DSPs. However it should be borne in mind that the the SOAR during switching transitions. If such a
reliability and the performance of the power switch situation is even suspected, then appropriate snub-
is very much dependent on the drive circuit. It is not ber circuits must be incorporated to reduce the
without reason that the phrase “a power electronic device stresses even though it may be at the expense
product is as good as the drive circuit ” is central to the of efficiency.
performance of power electronic systems. Practice and practical are the essence of a good
BJTs are generally more robust during handling engineer. One must strive to both simulate and
whereas care must be taken in handling MOSFETs breadboard the BJT and MOSFET drive circuits
and IGBTs. The MOSFETs and IGBTs are voltage- discussed in this chapter and more by referring to
controlled devices. Their turn-ON depends on the literature. The next section provides few exercises
gate charge that is provided. The body of a person that can be simulated in spice and also implemented
contains sufficient static charge to charge up the gate by hardware breadboarding. The insights gained in
capacitance to a few hundred volts whereas the gate– obtaining experimental clarifications will be helpful
source or gate–emitter can handle only around 20 V. while designing DC–DC converters and DC–AC
Therefore, in many cases, the devices will get damaged inverters.
Chapter 02.indd 85 3/12/2009 4:02:28 PM

1. Consider the BJT drive test circuit shown in (c) From the tabulated values of ib, ic and Vce,
Figure 2.33. It consists of the test transistor Q p compute the product ic × Vce the instanta-
that is to be studied. Q p is used to switch a resis- neous power loss in the device. Plot ib, ic,
tive load Rc that draws power from a 15 V DC Vce and ic × Vce versus time.
supply. The base drive circuit comprises comple- (d) What is the effect of R1 on the above
mentary transistors Q 1 (NPN) and Q 2 (PNP) waveforms?
transistors that are connected as shown. Q 1 col- (e) At what value of ic and ib does Vce attain
lector is connected to Vcc positive pole and Q 2 saturation value of around 0.3 V?
collector is connected to Vee negative pole as (f ) At what value of ic and ib does Vce attain a
shown. The base pulse signal source Vb is used to value around 0.7 V which is just at the
provide the base drive pulse signal to switch the boundary of saturation?
power transistor Q p at a specific frequency. (g) Set R1 to an appropriate value. Set Vee = 5 V.
Mode of implementation: The above circuit Measure and tabulate ic, ib and Vce of Q p
can be studied by for various values of Vcc.
a. Simulation in Spice (h) From the tabulated values of ib, ic and Vce,
b. Hardware breadboarding compute the product ic × Vce the instanta-
neous power loss in the device. Plot ib, ic,
Tasks for study: Vce and ic × Vce versus time.
(a) Rig up the circuit/netlist as given in (i) What is the effect of Vcc on the above
Figure 2.33. waveforms?
(b) Set Vcc = 10 V, Vee = 5 V. Measure and ( j) Set R1 to an appropriate value. Set Vcc = 10 V.
tabulate ic, ib and Vce of Q p for various Measure and tabulate ic, ib and Vce of Q p
values of R1. for various values of Vee.
15 V
RC
Vcc
0−10 V ic
Q1
R3 ib
QP Vce
Vb R1
Q2
Vee
0−5 V 0−5 V
R2
Figure 2.33 BJT drive test circuit.
Chapter 02.indd 86 3/12/2009 4:02:28 PM

Drive Circuits 87
(k) From the tabulated values of ib, ic and Vce, (c) From the tabulated values of ib, ic and Vce,
compute the product ic × Vce the instanta- compute the product ic × Vce the instanta-
neous power loss in the device. Plot i b, ic, neous power loss in the device. Plot i b, ic,
Vce and ic × Vce versus time. Vce and ic × Vce versus time.
(l) What is the effect of Vee on the above (d) What is the effect of R2 on the above wave-
waveforms? forms?
2. Consider the BJT drive test circuit shown in (e) Set Vcc = 10 V. Set R1 to provide steady-
Figure 2.34. It is the same as the test circuit of ON base-current value appropriate for the
Figure 2.33 but for the inclusion of the speed collector current. Set R2 to provide the i b+.
up circuit. Measure and tabulate ic, ib and Vce of Q p
for various values of Vee.
Mode of implementation: The above circuit (f ) From the tabulated values of ib, ic and Vce,
can be studied by compute the product ic × Vce the instanta-
a. Simulation in Spice neous power loss in the device. Plot i b, ic,
b. Hardware breadboarding Vce and ic × Vce versus time.
Tasks for study: (g) What are the values of ib+ and ib – at which
turn-ON and turn-OFF is best and switching
(a) Rig up the circuit/netlist as given in power dissipation is least?
Figure 2.34. (h) On incorporating another BJT in
(b) Set Vcc = 10 V, Vee = 5 V. Set R1 to provide Darlington configuration with the output
steady-ON base-current value appropriate for power BJT Q p as shown in Figure 2.35,
the collector current. Measure and tabulate ic, what is the effect on the base drive
ib and Vce of Q p for various values of R2. requirements?
Ensure in all cases that R2 is less than R1/5.
15 V
RC
Vcc
0−10 V R2 C ic
Q1
R3 ib
QP Vce
Vb R1
Q2
Vee
0−5 V 0−5 V
R2
Figure 2.34 BJT drive test circuit.
Chapter 02.indd 87 3/12/2009 4:02:28 PM

15 V Mode of implementation: The above circuit

can be studied by
Rc a. Simulation in Spice
ic
Tasks for study:
ib
QP1 (a) Rig up the circuit/netlist as given in
QP Vce Figure 2.36.
RP1
(b) Set Vcc = 15 V, Vee = 5 V. Measure and
tabulate id, ig and Vds of Q p for various
RP2 values of R1.
(c) What is the effect of R1 on the turn-ON
Figure 2.35 Darlington configuration of the and turn-OFF times? From the tabulated
output power transistor. (All other values of id, ig and Vds estimate the gate
portions of the drive are the same charge required to turn-ON the MOSFET
as in Figure 2.34.) to full enhancement.
3. Consider the MOSFET drive test circuit shown (d) Set R1 to an appropriate value. Set Vcc = 15 V.
in Figure 2.36. It consists of the test MOSFET Measure and tabulate id, ig and Vds of Q p
Q p that is to be studied. Q p is used to switch a for various values of Vee.
resistive load Rc that draws power from a 15 V (e) What is the effect of Vee on the above
DC supply. The gate drive circuit comprises waveforms?
complementary transistors Q 1 (NPN) and Q 2 4. Consider the MOSFET shunt snubber test cir-
(PNP) transistors that are connected as shown. cuit shown in Figure 2.37. It consists of the test
Q 1 collector is connected to Vcc positive pole and MOSFET Qp that is used to switch a resistive load
Q 2 collector is connected to Vee negative pole as Rd that draws power from a DC supply, Vdd. The
shown. The gate pulse signal source Vg is used to inductance L1 indicates the track inductance of
provide the gate drive pulse signal to switch the the inductive effects in the load. The shunt snub-
power MOSFET Q p at a specific frequency. ber circuit comprising R, C and D is connected
15 V
RC
Vcc id
0−15 V
Q1
R3 QP Vds
ig
Vg R1
Q2
Vee
0−5 V 0−5 V
R2
Figure 2.36 MOSFET drive test circuit.
Chapter 02.indd 88 3/12/2009 4:02:29 PM

Drive Circuits 89
Vdd
Rd
Vcc(15 V) L1
R
MOSFET QP
Vg drive
circuit R1 D
C
Figure 2.37 MOSFET shunt snubber test circuit.
as shown. The gate pulse signal Vg is used to pro- (b) Set Vdd = 100 V, Rd = 20 Ω. Measure and
vide the gate pulse signal to switch the power tabulate id and Vds of Q p for various values
MOSFET Q p at a specific frequency. The pulse of C.
signal from Vg is passed through an appropriate (c) What is the effect of C on the Vds waveform?
MOSFET drive circuit to provide the necessary (d) What is the effect of R on the turn-ON
gate charge to switch the MOSFET Q p. drain current of the MOSFET?
(e) Measure the rms current through R and
Mode of implementation: The above circuit compute the power dissipation in R with
can be studied by and without D.
a. Simulation in Spice (f ) How does the modified shunt snubber cir-
b. Hardware breadboarding cuit as shown in Figure 2.38 operate? What
Tasks for study: is the effect on the id and Vds waveforms?
What is the role of RC time constant on
(a) Rig up the circuit/netlist as given in the Vds waveform? What should be the
Figure 2.37. value of R and C for a given load?
Vdd
Rd
L1
Vcc(15 V)
D
MOSFET QP
C R
Vg drive
circuit R1
Figure 2.38 Modified shunt snubber circuit.
Chapter 02.indd 89 3/12/2009 4:02:29 PM

5. Consider the MOSFET series snubber test cir- passed through an appropriate MOSFET drive
cuit shown in Figure 2.39. It consists of the test circuit to provide the necessary gate charge to
MOSFET Q p that is used to switch a capaci- switch the MOSFET Q p.
tive load. The series snubber circuit comprising
R, L and D is connected as shown. The gate
can be studied by
pulse signal Vg is used to provide the gate pulse
signal to switch the power MOSFET Q p at a a. Simulation in Spice
specific frequency. The pulse signal from Vg is b. Hardware breadboarding
Vdd
D
L
R
Vcc(15 V)
MOSFET QP
Vg drive
circuit R1
Figure 2.39 MOSFET series snubber test circuit.
Vdd
L
R C
Vcc (15 V)
MOSFET QP
Vg drive
circuit R1
Figure 2.40 Modified series snubber circuit.
Chapter 02.indd 90 3/12/2009 4:02:29 PM

Drive Circuits 91
Tasks for study: (e) Measure the rms current through R and
compute the power dissipation in R.
(a) Rig up the circuit/netlist as given in Figure
2.39. (f ) How does the modified series snubber
circuit as shown in Figure 2.40 operate?
(b) Set Vdd = 15 V. Measure and tabulate id
What is the effect on the id waveform and
and Vds of Q p for various values of L.
Vds waveform? What is the role of RC time
(c) What is the effect of L on the id waveform? constant on the id and Vds waveforms?
What should be the value of R and C for a
(d) What is the effect of R on the turn-OFF
given load?
drain–source voltage of the MOSFET?

1. BJT is a device. 11. If the collector current , then the
transistor will come out of saturation and oper-
2. In BJT switching, the base currents are
ate in the active region.
on the collector currents.
12. In applications where the load current can vary
3. During the turn-ON period, the base current
in a wide range, the base current should vary
must be than the final or steady-ON
to the collector current. Such a base
value.
drive circuit is called .
4. During the turn-OFF period, a base
13. Isolation between the higher power collector
current should be provided to quickly remove
side and the lower power base side of the power
the stored charges in the transistor.
transistor is provided either by using
5. For the BJT drive circuit-1, R2 is used to pro- or by using .
vide the path for removing the
14. In the case of opto-coupler isolation, the duty
charges during turn-OFF.
cycle can be varied from 0 to 1, but
6. In BJT drive circuit-2, the two transistor stages, power supplies for the diode side and transistor
Q 2 and Q 1, provide the gain to suit- side of the opto-coupler are required.
ably drive Q p.
15. In the case of transformer isolation, the duty
7. In BJT drive circuit-3, the turn-OFF time of cycle can be varied only from .
the power transistor is improved by using an
16. MOSFET is a .
.
17. In MOSFET switching, the gate currents are
8. During turn-OFF, the negative base current for
on the drain currents.
fast turn-OFF in BJT drive circuit-6 is given
by . 18. The gate current is during the steady-
ON period when the MOSFET is ON.
9. If the transistor is operated to be just out of
, then the stored base charge is small 19. Compared to the BJT, the MOSFET drive
and the turn-OFF times will be very fast. circuit power requirement is .
10. If the collector current , then the 20. The benefits of the dual transformer ORed
transistor will be overdriven into saturation for circuit as given in MOSFET drive circuit-9 are
the same base drive. , and .
Chapter 02.indd 91 3/12/2009 4:02:29 PM

21. In MOSFET drive circuit-10, when the output 23. limits the voltage across the device
power MOSFET is OFF, the gate is during turn-OFF process.
and open to due to large dv/dt spikes
24. Turn-OFF snubber is also called .
on the drain side.
25. If there is a capacitive load or parasitic capaci-
22. If there is an inductive load in the collector or
tance across the device, then a large
if there is significant amount of lead inductance
flows through the device at turn-ON.
associated with the collector or emitter leads,
then when the BJT is being turned OFF, the 26. Turn-ON snubber circuit limits the
current through the device will fall rapidly to the device during turn-ON process.
zero in a time corresponding to the fall time of
27. Turn-ON snubber is also called .
the device leading to large stress
on the device. 28. The circuits with snubbers will be
efficient.
1. Discuss the effect of the base drive waveshape on 12. Explain the operation of the opto-coupler-based
the turn-ON and turn-OFF speeds of the device. BJT drive circuit-8.
2. Discuss the effect of the base drive waveshape 13. Explain the operation of the transformer-based
on the switching power dissipation. BJT drive circuit-9.
3. What are the requirements of a good base drive? 14. Explain the operation of the BJT drive circuit-12.
Illustrate with the base-current waveform.
15. What is the difference between BJT drive
4. Explain the operation of the BJT base drive circuit-12 and BJT drive circuit-13?
circuit-1.
16. Explain the operation of MOSFET drive
5. Explain the operation of the BJT base drive circuit-1. What is the function of resistor R2?
circuit-2. Under what constraints should R2 be used?
6. How does the operation of BJT drive circuit-3 17. In MOSFET drive circuit-2, why are the buffers
differ from that of BJT drive circuit-4? connected in parallel?
7. In BJT drive circuit-5, the R2-C is the speed up 18. Explain the operation of MOSFET drive
circuit that ensures fast turn-ON. Explain? circuit-3.
8. Explain the difference between the BJT drive 19. Explain the difference between the opto-
circuit-5 and BJT drive circuit-6. isolated BJT drive circuit-8 and MOSFET
drive circuit-4.
9. What is Baker’s clamp? Where and why is it
used? 20. “The MOSFET drive circuit-6 is used to drive
the top MOSFET of bridge arms.” Explain its
10. Explain the operation of a non-isolated
operation.
proportional-base drive circuit.
21. How can the duty cycle range be improved
11. What is the function of D1 in the proportional-
with transformer isolation between the collec-
base drive circuit as given in BJT drive circuit-7?
tor side and the base drive side?
Chapter 02.indd 92 3/12/2009 4:02:29 PM

Drive Circuits 93
22. In the MOSFET drive circuit-9, what is the func- 24. The protection circuit discussed in MOSFET
tion of the secondary-side resistors R1 and R4? drive circuit-11 needs to be disabled at turn-ON.
Why?
23. Explain the operation of MOSFET drive
circuit-10.
| PROBLEMS
1. A BJT has to switch a load of 1 A. The base turn-ON is 2 A, then calculate the turn-ON
drive power is derived from 5 V voltage source. time.
Calculate the values of resistors R1 and R2 for
8. For Problem 7 above, what is the peak current
the BJT having saturation hFE as 100.
requirement for a turn-ON time of 1 μs? Calcu-
2. A BJT has to switch a load of 10 A that is late the gate resistance that needs to be connected
connected to a 400 V DC link/bus. The BJT in series.
drive circuit-3 is used to drive the power tran-
9. What should be the voltage and current rating
sistor. The power transistor has a saturation
of Q 2 in MOSFET drive circuit-3?
hFE of 80. The base drive is powered from a 10
V DC source. Calculate the values of R1, R2 10. A bridge arm is used as part of a converter. The
and L. DC-link voltage is 400 V. The gate power
supply is derived from a 15 V power supply.
3. A load of 10 A is to be switched by a BJT. The
The MOSFET drive circuit-6 is employed to
BJT base drive circuit is powered from 10 V DC
drive the top MOSFET of the bridge arm. The
source. The BJT selected for the application has
bridge MOSFETs are switching a drain cur-
the following specifications: hFE(sat) = 100,
rent of 10 A. The RDS(ON) of the bridge MOS-
tr = 2 μs and tf = 2 μs. If the BJT drive circuit-5
FETs is 0.1 Ω. When the top MOSFET is
is used in the application, then calculate the
ON, what is the reverse voltage across the
values of R1, R2 and C.
diode D?
4. In BJT drive circuit-10, consider the time
11. In the protection circuit of MOSFET drive
period when the transistor Q 1 is ON. What is
circuit-11, a drain current of 20 A flows
the voltage across the freewheeling diode Df
through the power MOSFET that has a RDS(ON)
during this time period?
of 0.1 Ω. What is the voltage at the “+” terminal
5. In the BJT drive circuit-11, the PNP transistor of amplifier A1?
Q 1 is used to ensure fast turn-OFF of the power
12. A MOSFET is switching a 20 A inductive load
transistor. What should be the Vce and the col-
from a 100 V DC source. The fall time of the
lector-current rating of the transistor Q 1?
device is 0.5 μs. Calculate the shunt snubber
6. For a varying load current that has a maximum capacitance value.
value of 10 A, design a proportional-base drive
13. For Problem 12, the peak current rating of the
circuit as discussed in BJT drive circuit-13.
MOSFET is specified as 30 A. The converter,
7. A particular MOSFET requires 400 nC to take where the MOSFET is used, operates at a
it to full enhancement. The MOSFET is driven switching frequency of 20 kHz. The range of
from a 15 V source. If the peak gate current during duty cycle is from 0.2 to 0.7. What is the value
Chapter 02.indd 93 3/12/2009 4:02:30 PM

of the shunt snubber resistor? What is the power time of the device is 2 μs. Calculate the series
dissipation in the shunt snubber resistor? snubber inductance value. The voltage rating
of the device is specified as 150 V. The con-
14. For Problem 13, what is the snubber diode
verter where the IGBT is used operates at a
current and peak inverse voltage rating?
switching frequency of 20 kHz. The range of
15. An IGBT is switching a capacitive load. The duty cycle is from 0.2 to 0.7. What is the
load current is 20 A drawn from a 100 V DC value of the series snubber resistor? What is
source. To protect against the turn-ON cur- the power dissipation in the shunt snubber
rent surges, a series snubber is used. The rise resistor?
| ANSWERS
Fill in the Blanks
1. current-controlled 12. in proportion; proportional- isolated power supplies on
2. dependent base drive circuit the secondary side
3. higher 13. opto-couplers; transformers 21. floating; turn-ON
4. negative 14. isolated 22. voltage
5. discharge; stored base 15. 0 to 0.5 23. Turn-OFF snubber circuit
6. current 16. voltage-controlled device 24. shunt snubber
7. inductor 17. not dependent 25. current surge
8. (Vcc + Vbesat)/R1 18. zero 26. current through
9. saturation 19. lesser 27. series snubber
10. decreases 20. isolation; full range of duty 28. less
11. increases cycle; no requirement for
Chapter 02.indd 94 3/12/2009 4:02:30 PM

Rectifiers
Learning Objectives
CHAPTER
3
understand the operation of rectifiers and the effect of the various loads on rectifier
functioning.
draw current and voltage waveforms at various points in the circuit.
understand and design capacitor-filter rectifier circuits.
understand the operation of controlled rectification.
T his chapter deals primarily with converting the AC voltage that is available from the mains to the DC
voltage which is required for most electronic products. The AC voltage is first converted to a pulsating
DC voltage by using diodes or thyristor. This pulsating DC is then filtered to provide smooth DC voltage.
The rectifiers may be broadly classified as
1. uncontrolled rectifiers;
2. controlled rectifiers.
The uncontrolled rectifiers use diode as the semiconductor power switch. As the turn-ON and turn-OFF of the
diode is uncontrolled (discussed in Chapter 1), such rectifier topologies using diodes as the power switch are called
uncontrolled rectifier. On the other hand, if any of the controllable power semiconductor switches like the bipolar
junction transistors (BJTs), metal oxide semiconductor field effect transistors (MOSFETs), insulated gate bipolar
transistors (IGBTs) and thyristors are used, then such rectifiers are called controlled rectifiers.
The uncontrolled rectifiers are discussed first in this chapter. The various single- and three-phase rectifier
configurations are discussed. This is followed by a discussion on the popular capacitor input filter rectifier
wherein the operation and design of capacitor input filters are dealt. The inrush current at rectifier startup is
also addressed by including some methods to solve this problem. The discussion on the capacitor input filter
is followed by the inductance input filter or the LC-filter rectifier. The output ripple content of the LC filter
is analyzed. The effect of LC filter on the currents in the rectifier is also discussed.
After discussing uncontrolled rectifiers, the controlled rectifiers are addressed. The phase-controlled
converters, viz., single-phase half-wave controlled converters, single-phase full-wave converters, three-phase
half-wave controlled converters, three-phase full-wave half-controlled converters and three-phase full-wave
full-controlled converters are considered. The firing sequences and the corresponding waveforms at the rectifier
outputs are also discussed.
Chapter 03.indd 95 3/12/2009 3:49:15 PM

3.1 Uncontrolled Rectifiers
E lectrical power generation and distribution is usually accomplished in the form of AC currents and
voltages. However, many types of electronic and electrical equipments operate from DC sources. The
AC voltage must therefore be rectified and filtered to provide a DC output voltage at a specified current or
power level. Depending on the load power requirements, the AC input may be obtained from
1. single-phase supply for low-to-medium power requirements;
2. three-phase supply for medium-to-high power requirements.
A transformer may be used in between the AC input supply and the rectifier input to provide
1. voltage scaling;
2. isolate the input from the rectifier output.
In such cases, the transformer should be capable of handling the entire load power and should be designed
for the mains frequency. This would increase the cost, size and weight of the rectifier. Frequently, the
input voltage is directly rectified by using a transformer and the filtered rectifier output is then switched
at a high frequency across a load or transformer supplying the load. A typical example of the latter case is
switching power supplies and step-up frequency converters. In either case, the magnetics operate at a
higher frequency for size, weight and cost reduction. These are classified as DC–DC converters that are
discussed in Chapter 5.
In high-power systems, regulation is frequently accomplished by controlling the phase or conduction
time of the AC wave by thyristors (silicon-controlled rectifiers; SCRs). In low-power systems, such as
AC–DC power supplies, regulation is usually achieved by either a transistor post regulator like in linear
regulator or by pulse-width modulation for switched-mode regulation.
3.2 Rectifier Circuits
T he uncontrolled rectifier circuits consist primarily of diode semiconductor switches. The diodes are
connected in various topological configurations. The source for the rectifier circuits is considered to be
a sinusoidal voltage source such as the mains/grid that is the most common source in many applications.
Based on the type of the input source for the rectifiers, they are broadly classified as
1. single-phase rectifier circuits;
2. three-phase rectifier circuits.
As the names imply, the single-phase rectifier circuit topologies are designed for applications wherein the
input source is the single-phase grid and likewise, the three-phase rectifier circuit topologies are designed for
applications with three-phase grid as the input source.
Single-Phase Circuits
In this category of rectifier topologies, there are three basic rectifier configurations that are popular:
1. half-wave rectifier;
2. full-wave center-tapped rectifier;
3. full-wave bridge rectifier.
These three basic configurations will now be discussed.
Chapter 03.indd 96 3/12/2009 3:49:16 PM

Rectifiers 97
Half-Wave Rectifier
The single-phase half-wave rectifier circuits for various load conditions are shown in Figure 3.1. The recti-
fier topology of Figure 3.1(a) is applicable for resistive and/or capacitive loads. On the other hand, the
rectifier topology of Figure 3.1(b) is applicable for inductive and/or resistive–inductive loads. The diode D2
across the load is needed for providing a path for the trapped kinetic energy in the inductor to discharge
when the diode D1 is reverse-biased. As the diode D2 provides a path for the inductor current to freewheel
through it when D1 is OFF, D2 is called the freewheeling diode. The mains/grid voltage Vin is applied at the
primary of the transformer. The voltage at the secondary of the transformer is nVin. The choice of the turns
ratio n depends on the load voltage requirements.
Whenever the dot poles of the transformer are positive, diode D1 will conduct and the load voltage VL
will follow the secondary voltage Vs. During this time the diode D2 in the case of circuit of Figure 3.1(b)
will be reverse-biased and therefore OFF. When Vs goes negative with respect to the dot pole, D1 will be
reverse-biased and will switch OFF. If the load is inductive, then the inductive current will freewheel
through diode D2.
Figure 3.2 shows the voltage and current waveforms for a resistive load. In this case the diode D2 is not
operative. Referring to the waveforms shown in Figure 3.2, the load voltage is a pulsating half-sinusoid as
indicated. It is given by
nVm
VLav = (3.1)
π
where VLav is the average voltage across the load resistor RL; Vm is the peak voltage of the input sine wave; n
is the transformer turns ratio.
D1 IL
1:n
Load
Vin Vs VL (R or RC)
(a)
D1
1:n IL
Load
Vin Vs D2 VL (L or RL)
(b)
Figure 3.1 Half-wave rectifier configuration: (a) For resistive–capacitive

loads; (b) for resistive–inductive loads.
Chapter 03.indd 97 3/12/2009 3:49:16 PM

Vm
Vin
0 t
nVm
VL
VLav
0 t
Im /n
IL
ILav
0 t
Figure 3.2 Half-wave rectifier: Waveforms for resistive load.
The average load current is given as

nVm
I Lav = (3.2)
π RL
where ILav is the average load current and the remaining symbols have their usual meaning. The power
delivered to the load is given as
1
T /2
⎛ nVm sin ω t ⎞
Po =
T ∫ (nVm sin ω t ) ⎜⎝ RL ⎟ dt
⎠
0
n 2Vm 2 ⎛ nVm ⎞ ⎛ nVm ⎞ (3.3)

= =⎜ ⎟⎜ ⎟
4 RL ⎝ 2 ⎠ ⎝ 2 RL ⎠
= VLrms I Lrms
where VLrms is the root mean square (rms) voltage across the load resistor and ILrms is the rms current flowing
through the load resistor.
The current through the secondary of the transformer is also same as that of the load and hence the rms
secondary current is given as
I srms = I Lrms (3.4)
The primary or line side rms current is given as

I prms = nI srms (3.5)
Chapter 03.indd 98 3/12/2009 3:49:16 PM

Rectifiers 99
It can be seen from the waveforms of Figure 3.2 that the maximum peak inverse voltage (PIV) that Dl
should withstand when OFF is Vm. Likewise in the case of inductive loads, D2 is operative. When D2 is
OFF, D1 is ON. The maximum PIV that D2 should withstand when OFF is also Vm.
The output voltage is only a half sine wave. Some form of output filtering is essential to obtain a low
ripple output voltage. This topology can also be used for direct line rectification by removing the costly
transformer. However, the half-wave circuit is usually limited to low-power, poor output ripple applications.
This topology is used to obtain the DC bus voltage for inputs of linear regulators in power supplies.
Full-Wave Center-Tapped Rectifier

The full-wave center-tapped rectifier circuit is shown in Figure 3.3. Here the use of a transformer is manda-
tory. The turns ratio n is defined as the ratio of the number of windings in the secondary between the dot pole
and the center-tap to the number of windings in the primary. Referring to Figure 3.3, diode D1 will conduct
when the dot poles of the transformer are positive and VL will follow the positive half of Vs. When the dot
poles are negative, diode D2 will conduct resulting in a full wave rectified waveform at the output. Here, the
peak-reverse voltage for the diodes is twice Vs, thus demanding a higher reverse voltage rating for a given VL
requirement.
In the case of the center-tapped rectifier circuit, there is no need for a separate freewheeling diode. This
is due to the fact that one of either D1 or D2 is always ON. The voltage and current waveforms are shown in
Figure 3.4.
Referring to the waveforms shown in Figure 3.4, the load voltage is a pulsating full sinusoid as indicated.
It is given by
2nVm
VLav = (3.6)
π
where VLav is the average voltage across the load resistor RL; Vm is the peak voltage of the input sine wave;
n is the transformer turns ratio. The average load current is given as
2nVm
I Lav = (3.7)
π RL
where ILav is the average load current.
D1
1:n IL
Vs = nVin VL Load
Vin
Vs
D2
Figure 3.3 Full-wave center-tapped rectifier.
Chapter 03.indd 99 3/12/2009 3:49:17 PM

Vm
Vin
0 t
nVm
VLav
VL
0 t
Im/n ILav
IL
0 t
Im/n
ID1
0 t
Im/n
ID2
0 t
Im
Iin
0 t
Figure 3.4 Center-tapped full-wave rectifier: Waveforms for resistive load.
The power delivered to the load is given as

T /2
∫ (nVm sin ω t ) ⎜⎝
2
Po = ⎟ dt
T 0
RL ⎠
(3.8)
n Vm 2 ⎛ nVm ⎞ ⎛ nVm ⎞
2
= =⎜ ⎟⎜ ⎟ = VLrms I Lrms
2 RL ⎝ 2 ⎠ ⎜⎝ 2 RL ⎟⎠
where VLrms is the rms voltage across the load resistor and ILrms is the rms current flowing through the load
resistor.
Chapter 03.indd 100 3/12/2009 3:49:17 PM

Rectifiers 101
The current through the secondary of the transformer is a half-wave rectified waveform. The portion of
the transformer secondary winding above the center-tap carries current only when D1 is ON as indicated in
ID1 waveform of Figure 3.4. On the other hand, the portion of the secondary winding below the center-tap
carries current only when D2 is ON as indicated in ID2 waveform of Figure 3.4. Thus the secondary winding
currents are half-wave rectified waveforms. The rms value of the secondary winding current is given as
nVm I Lrms
I srms = = (3.9)
2 RL 2
The primary or line side current is however a full sine waveshape as this is the algebraic sum of the reflected
secondary winding currents. This is indicated in the Iin waveform of Figure 3.4. The rms primary or line side
current is given as
I prms = nI Lrms (3.10)
Unlike the half-wave rectifier topology, the output voltage here is a full-wave rectified waveform. This means
that the output filter requirement is less stringent as compared to the half-wave rectifier output.
Full-Wave Bridge Rectifier

The full-wave bridge rectifier circuit is shown in Figure 3.5. This is one of the most popular rectifier configu-
rations used in many practical circuits. Here four diodes are used as indicated. When the dot poles of the
transformer are positive, then D1 and D4 will conduct. When the dot poles are negative, then D3 and D2 will
conduct thus resulting in a full-wave rectified voltage waveform at the output. It should be noted that the
full-wave bridge topology can also be used for direct line rectification wherein one can avoid using the input
transformer.
The bridge circuit is normally used for medium-to-high-voltage output applications. Here, the peak-
reverse voltage for all the diodes is Vs itself, thus requiring diode with PIV ratings that are half that
required for the center-tapped configuration for a given output voltage requirement. However, it should
be noted that two diodes at any given instant will conduct and are in series. Hence there will be two diode
drops in the bridge rectifier topology as compared to the half-wave or full-wave center-tapped rectifier
topologies.
IL
D1 D3
1:n
VL Load
Vin Vs = nVin (R, L, C)
D2 D4
Figure 3.5 Full-wave bridge rectifier.
Chapter 03.indd 101 3/12/2009 3:49:17 PM

In the case of the bridge rectifier circuit also, there is no need for a separate freewheeling diode. This is
due to the fact that diode pair D1, D2 and diode pair D3, D4 act as freewheeling paths for any inductive load.
The voltage and current waveforms are shown in Figure 3.6.
Referring to the waveforms shown in Figure 3.6, the load voltage is a pulsating full sinusoid as indicated.
2nVm
VLav = (3.11)
π
where VLav is the average voltage across the load resistor RL; Vm is the peak voltage of the input sine wave;
n is the transformer turns ratio.
Vm
Vin
0 t
nVm VLav
VL
0 t
Im/n ILav
IL
0 t
Im/n
ID1, ID4
0 t
Im/n
ID2, ID3
t
Im
Iin
0 t
Figure 3.6 Bridge full-wave rectifier: Waveforms for resistive load.
Chapter 03.indd 102 3/12/2009 3:49:18 PM

Rectifiers 103
The average load current is given as

2nVm
I Lav = (3.12)
π RL
where ILav is the average load current. The power delivered to the load is given as
2
T /2
Po =
T ∫ (nVm sin ω t ) ⎜⎝ RL ⎟ dt
⎠
0
n Vm 2 ⎛ nVm ⎞ ⎛ nVm ⎞
2 (3.13)
= =⎜ ⎟⎜ ⎟ = VLrms I Lrms
2 RL ⎝ 2 ⎠ ⎜⎝ 2 RL ⎟⎠
where VLrms is the rms voltage across the load resistor and ILrms is the rms current flowing through the load
resistor.
The current through the secondary of the transformer is a full-wave rectified waveform. The rms value of
the secondary winding current is given as
I srms = I Lrms (3.14)
The primary or line side current is a full sine waveshape that is the reflected secondary winding current. This
is indicated in the Iin waveform of Figure 3.6. The rms primary or line side current is given as
I prms = nI Lrms (3.15)
The currents through the diodes are half-wave rectified waveforms as indicated in Figure 3.6. As in the
case of the center-tapped full-wave rectifier topology, the output voltage here is a full-wave rectified wave-
form. This means that the output filter requirement is less stringent as compared to the half-wave rectifier
output.
Three-Phase Circuits
Three-phase power is used in medium-to-high-power applications and may be applied to an off-line rectifier,
say to produce a nominal 560 V DC bus from a 400 V AC line, or to a transformer whose secondary or
secondaries are rectified in various manners. The advantages of using three-phase power as compared to
single-phase power are: (a) higher output voltages, (b) lower output ripple, (c) higher input power factor and
(d) reduced harmonic distortion of the input current.
Figures 3.7 and 3.8 show the Y–Y and Δ–Y half-wave rectifier circuit configurations. The RYB
secondary-side line-to-line voltages, the load voltage and the load currents for a resistive load are shown
in Figure 3.9. It can be observed that the load voltage ripple is considerably reduced as compared to that
of the single-phase rectifier configurations. Referring to Figure 3.9, it can be observed that the load voltage
ripple swing is from the peak voltage nVm to nVmcos(60°), that is, nVm/2. Referring to Figure 3.9 and
considering the voltage reference axis to be at the peak of the sinusoid, the load voltage contribution
from each phase is
π /3
1
2π ∫ nVm cos θ dθ (3.16)
− π /3
where n is the secondary to primary line-to-line turns ratio.
Chapter 03.indd 103 3/12/2009 3:49:18 PM

D1
VL Load
Y
D2
B
D3
Figure 3.7 Three-phase half-wave rectifier, Y–Y.
D1
VL Load
D2
B
D3
Figure 3.8 Three-phase half-wave rectifier, Δ–Y.
Each of the three phases contribute to the total load average voltage equally in a period and hence the
average load voltage is given as
π /3
3 3nVm π /3 3 3nVm
VLav =
2π ∫ nVm cos θ dθ =
2π
sin θ − π /3 =
2π
(3.17)
− π /3
The average load current for a resistive load RL is given as
VLav 3 3nVm
I Lav = = (3.18)
RL 2π RL
Chapter 03.indd 104 3/12/2009 3:49:18 PM

Rectifiers 105
R Y B
nVm
VS
0 t
60˚
nVm/RL
IL
0 t
nVm/RL
ID1
0 t
Figure 3.9 Load voltage and current waveforms for half-wave rectifier
configurations of Figures 3.7 and 3.8.
The rms value of the load voltage is given as
π /3
1
VLrms =
2π ∫ (nVm cos θ )2 dθ × 3
− π /3
3n 2Vm 2 ⎛ π /3 ⎞
π /3 π /3
3n 2Vm 2 sin 2θ
=
2π ∫ cos 2 θ dθ =
4π ⎜⎝
⎜ θ − π /3 +
2
⎟
⎟
− π /3 − π /3 ⎠
⎛ 1 3 3⎞ n 2Vm 2 ⎛ 3 3 ⎞
= n 2Vm 2 ⎜ + ⎟ = ⎜1+ ⎟
⎝ 2 8π ⎠ 2 ⎝ 4π ⎠
nVm 3 3
VLrms = 1+ (3.19)
2 4π
Similarly, the rms value of the load current for a resistive load is given as
nVm 3 3
I Lrms = 1+ (3.20)
RL 2 4π
Chapter 03.indd 105 3/12/2009 3:49:19 PM

⎡ 1 π /3 ⎛ nV cos θ ⎞ ⎤
Po = ⎢ ∫
⎢⎣ 2π −π /3
(nVm cos θ ) ⋅ ⎜⎜ m ⎟⎟ dθ ⎥ × 3
⎝ RL ⎠ ⎥⎦
π /3 π /3
3n 2Vm2 3n 2Vm2
=
2π RL ∫ cos 2 θ ⋅ dθ =
4π RL ∫ (1 + cos 2θ ) ⋅ dθ
−π /3 −π /3
3n 2Vm2 ⎛ π /3 sin 2θ ⎞
π /3
= ⎜θ + ⎟
4π RL ⎜ −π /3 2 ⎟
⎝ −π /3 ⎠
3n 2Vm2 ⎛ 2π 3⎞
= ⎜ + ⎟
4π RL ⎜⎝ 3 2 ⎟⎠
n 2Vm2 ⎛ 1 3 3 ⎞
Po = ⎜ + ⎟ (3.21)
RL ⎝ 2 8π ⎠
Referring to Eqs. (3.19) and (3.20), one can observe that the output power Po is
Po = VLrms I Lrms (3.22)
Diodes D1, D2 and D3 carry current for only a third portion of each period. The current in diode D1 is
shown in Figure 3.9. The average and the rms values of the diode currents are given as
3nVm
I Dav = (3.23)
2π RL
nVm 1 3
I Drms = + (3.24)
RL 2 3 4π
The secondary windings of the transformer carry the same currents as that flowing through the diodes.
Therefore the secondary winding average and rms currents are the same as that of the diodes. The primary
winding currents are turns ratio (n) times the secondary currents. In the case of star-connected primary, the
winding and line currents are same. In case the primary windings are delta-connected, the line currents are
3 times the winding currents.
Figure 3.10 shows a full-wave rectification with a three-phase bridge diode rectifier connected to the
secondary as indicated in Figure 3.10. It can be observed that the availability of the neutral point is not
essential for rectification in this topology. However, if the neutral point is available then a positive output at
the positive output rail with respect to the neutral and a negative output at the negative output rail with
respect to the neutral can be obtained in this configuration.
Figure 3.10 shows the D–Y full-wave bridge rectifier circuit configurations. The RYB secondary-side
line-to-line voltages, the load voltage and the load currents for a resistive load are shown in Figure 3.11.
Here the full-bridge topology performs a full-wave rectification. For example, when the secondary-side
R phase is positive, diode D1 conducts and when the R phase goes negative, diode D4 conducts. In a similar
Chapter 03.indd 106 3/12/2009 3:49:19 PM

Rectifiers 107
R
D1 D2 D3
VL
Load
Y
D4 D5 D6
Figure 3.10 Three-phase full-wave bridge rectifier, Y–Y.
R Y B
Vsec
nVm
0
t
nVm/RL
iL
30°
0
t
nVm/RL
iD1
0
t
Figure 3.11 Load voltage and current waveforms for bridge rectifier
configurations of Figure 3.10.
Chapter 03.indd 107 3/12/2009 3:49:19 PM

manner the diode pairs (D2, D5) and (D3, D6) bridge arms also operate. It can be noted that the load voltage
ripple is further improved as compared to the three-phase half-wave rectifier circuit. Referring to Figure 3.11,
it can be observed that the load voltage ripple swing is from the peak voltage nVm to nVmcos(30°), that is,
3nVm / 2. There are six ripples in one period. This means that each phase contributes two of the six ripples.
Referring to Figure 3.11 and considering the voltage reference axis to be at the peak of the sinusoid, the load
voltage contribution from each phase is equal to
π /6
1
2
2π ∫ nVm cos θ dθ (3.25)
− π /6
where n is the secondary to primary line-to-line turns ratio.

Each of the three phases contribute to the total load average voltage equally in a period and hence the
average load voltage is given as
π /6
3 3nVm π /6 3nVm
VLav = 2
2π ∫ nVm cos θ dθ =
π
sin θ − π /6 =
π
(3.26)
− π /6
The average load current for a resistive load RL is given as
VLav 3nVm
I Lav = = (3.27)
RL π RL
The rms value of the load voltage is given as
π /6
1
2π −π∫/6
VLrms = (nVm cos θ )2 dθ × 6
3n 2Vm 2 ⎛ π /6 sin 2θ ⎞
π /6 π /6
3n 2Vm 2
⎜θ ⎟
=
2π ∫ cos 2 θdθ =
2π ⎜ −π /6
+
2 ⎟
−π /6 ⎝ −π /6 ⎠
⎛1 3 3⎞ n 2Vm 2 ⎛ 3 3 ⎞
= n 2Vm 2 ⎜ + ⎟= ⎜1 + ⎟
⎜ 2 4π ⎟ 2 ⎜⎝ 2π ⎟⎠
⎝ ⎠
nVm 3 3
VLrms = 1+ (3.28)
2 2π
Similarly, the rms value of the load current for a resistive load is given as
nVm 3 3
I Lrms = 1+ (3.29)
RL 2 2π
⎡ 1 π /6 ⎛ nV cos θ ⎞ ⎤
Po = ⎢ ∫
⎢⎣ 2π −π /6
(nVm cos θ ) ⋅ ⎜ m
⎝ RL
⎟ dθ ⎥ × 6
⎠ ⎥⎦
Chapter 03.indd 108 3/12/2009 3:49:19 PM

Rectifiers 109
π /6
3n 2Vm2
=
π RL ∫ cos 2 θ ⋅ dθ
−π /6
2 2 π /6
3n Vm
=
2π RL ∫ (1 + cos 2θ ) ⋅ dθ
−π /6
π /6
3n 2Vm2 ⎛ π /6 sin 2θ ⎞
= ⎜θ + ⎟⎟
2π RL ⎜⎝ −π /6 2 −π /6 ⎠
3n 2Vm2 ⎛ π 3⎞
= ⎜ + ⎟
2π RL ⎝ 3 2 ⎠
n 2Vm2 ⎛ 1 3 3 ⎞
Po = ⎜ + ⎟ (3.30)
RL ⎝ 2 4π ⎠
Referring to Eqs. (3.28) and (3.29), one can observe that the output power Po is
Po = VLrms I Lrms (3.31)
Referring to Figure 3.11, it can be seen from the representative D1 current waveshape that each diode carries
the load current during two ripples in a period or carry an equivalent of one of the six load current ripples
in half a period. The D1 diode current is shown in Figure 3.11. The average and the rms values of the diode
currents are given as
π /6
1 nVm π /66 nV
I Dav =
π RL ∫ nVm cos θ dθ =
π RL
sin θ − π /6 = m
π RL
(3.32)
− π /6
nVm 1 3
I Drms = + (3.33)
RL 2 3 2π
3.3 Capacitor Input Filter
T he capacitor input filter is an inexpensive and one of the most popular filters that is used for almost all
applications and loads that require a DC bus. Capacitor input filters are the most volumetrically effi-
cient means of filtering rectified sine waves and storing energy. The schematic of the capacitor input filter is
shown in Figure 3.12. The capacitor charges up to the peak value of the input voltage and tries to maintain
this value as the full-wave rectified input drops to zero. The capacitor will discharge through the load until
the input full-wave rectified voltage again increases to a value greater than the capacitor voltage. At this
point, the diode rectifier will again recharge the capacitor.
The ripple voltage across the filter capacitor is a function of the filter capacitance value, the input fre-
quency and the load current. Considerable importance is given to calculations of ripple amplitude because
this parameter influences other design parameters for downstream power conversion devices. In the case of
linear power supplies, the minimum capacitor voltage at low line (i.e., minimum input voltage) must be
equal to the output voltage plus the minimum voltage which the pass regulator can tolerate while maintain-
ing a regulated output. At high line, the voltage across the pass regulator increases and the regulator must
dissipate substantial power.
Chapter 03.indd 109 3/12/2009 3:49:20 PM

1φ
or C RL Vo
3φ
AC
input
Rectifier
Figure 3.12 Capacitor input filter.
In switch-mode power supplies, higher ripple voltage may be tolerated since the pulse-width modulator
will correct for the DC bus ripple variations without an increase in power dissipation. The filter capacitance
may be chosen (for a desired output ripple) or the output ripple may be decided (for a desired capacitance).
In many cases, the power supply can operate with a 25% peak-to-peak ripple voltage across the input filter
capacitor and a line variation of ±15%.
Design of Capacitor Input Filter Rectifier

Figure 3.13 shows a single-phase full-wave bridge rectifier with capacitor input filter. This is the most
common type of AC-to-DC converter used in most applications. The design of the circuit of Figure 3.13 is
now discussed in detail.
The input filter capacitor C determines the peak-to-peak ripple voltage component DVr. At low-line
voltage, DVr determines the minimum input voltage Vm-min which is an important design consideration and
defines the transformer turns ratio required to achieve the specified output voltage at maximum load condi-
tions or maximum duty cycle of any downstream-controlled converter or inverter.
If the input filter capacitor is too small, the resulting large ripple voltage will require the post regulator
to have a large line-regulation requirement. This would result in poor transformer utilization and higher
ratings for the post regulator power devices. If the input filter capacitor is larger than necessary, it will not
only cost more but the recharging capacitive current pulses drawn from the line will also be narrower and
id i
D1 D3 ic io
Vmsin wt RL Vo
D2 D4
Figure 3.13 Single-phase rectifier–capacitor input filter.
Chapter 03.indd 110 3/12/2009 3:49:20 PM

Rectifiers 111
larger in amplitude. This adversely affects the line power factor and also increases the conducted electromag-
netic interference (EMI). The higher rms input line current causes increased losses in the line, diodes and
filter capacitor, thus decreasing the efficiency and reliability. Therefore a reasonable rule of thumb is to com-
promise on a ripple voltage of about 15–30% of the minimum peak line voltage resulting in acceptable
capacitor size, weight and cost. It is expected that the downstream converter or inverter will take care of the
ripple and line regulation.
Figure 3.14 shows the voltage and current waveforms for the rectifier-filter circuit of Figure 3.13. The
capacitor charges only during the period corresponding to the angle δ shown in Figure 3.13. During the
remaining period the capacitor is discharging to the load. The current waveforms are approximated to
pulsed waveforms as shown in Figure 3.13 without loss of generality from the point of view of design of
diodes and capacitor as the current rating obtained for the components would be a conservative value.
During the positive cycle the diode pair (D1, D4) conducts and during the negative cycle the pair (D3, D2)
conducts. The capacitor gets fully charged after four to five cycles. Referring to Figure 3.14, it can be
V1
Vo
V2 ΔVr
0
wt
a p−a
i
Im
Io
0
wt
id
Im
0
wt
Figure 3.14 Waveforms of voltage and currents for C-filter rectifier as shown in Figure 3.13.
Chapter 03.indd 111 3/12/2009 3:49:20 PM

observed that diodes D1 and D4 will get forward-biased when the input voltage increases more than V2.
Similarly, diodes D3 and D2 get forward-biased when the input voltage goes below V2 in the negative half-
cycle. Only when the rectifier diodes are forward-biased will there be a current flow to charge the capacitor C.
The charging up of the capacitor is reflected as an increase in voltage across the capacitor as indicated in the
waveform shown in Figure 3.14. Once the capacitor charges to V1, the input voltage begins to fall wherein
the rectifier diodes will become reverse-biased. Now the capacitor discharges into the load with average load
current Io and as a consequence the output voltage Vo will decrease as shown. This process repeats cycle by
cycle resulting in the output voltage to have a ripple of DVr. Observe from Figure 3.14 that the capacitor
charging current flows only during period α as indicated.
Rectifier Component Selection

In order to select the components of the rectifier, the following parameters must be specified a priori.
1. Output voltage, Vo.
2. Output ripple, DVr.
3. Maximum load power, Po.
4. Nominal input rms voltage, Vin,rms ± % tolerance.
5. Input voltage variation tolerance, %tol.
6. Frequency of input, fs.
At steady state, the energy gained by the capacitor is equal to the energy lost to the load. In a half-period, the
capacitor energy that is given to the load εo is
1
εo = C (V12 − V2 2 ) (3.34)
2
The output load energy in a half-period of T/2 when the capacitor discharges is given as
⎛π −α ⎞ T
εo = Po ⎜ ⎟ (3.35)
⎝ π ⎠ 2
From Eqs. (3.34) and (3.35), the output capacitor value is given as
π −α 2 Po
C= (3.36)
π (V1 − V2 2 ) f r
2
where fr is the frequency of the rectified waveform and is equal to 2/T. Equation (3.36) is the capacitor selec-
tion equation. With reference to this equation, the following comments are important. If fs is the source or
line frequency, then
1. for single-phase half-wave rectifiers, fr = fs = 50 Hz;
2. for single-phase full-wave rectifiers, fr = 2fs = 100 Hz;
3. for three-phase full-wave rectifiers, fr = 6fs = 300 Hz.
The output power Po is a design specification that is determined from the load requirements. V1 is the peak
value of the input voltage. It should correspond to the minimum peak value the input voltage can reach.
This is because the capacitor should be selected such that it is capable of providing the specified output
energy even under low input voltage situations. Thus,
V1 = Vm − min (3.37)
Chapter 03.indd 112 3/12/2009 3:49:21 PM

Rectifiers 113
where
⎛ %tol ⎞
Vm − min = 2Vin-rms ⎜ 1 −
⎝ 100 ⎟⎠
V2 is given as
V2 = V1 − ΔVr (3.38)
where DVr is the peak-to-peak ripple voltage and is usually specified at 15–30% of the minimum input
voltage.
In Figure 3.14, the currents through the diodes and the capacitors are shown approximated as pulse cur-
rents of peak amplitude Im and conducting for a period corresponding to the angle α. As the pulse approxi-
mation of the current waveshapes totally enclose the actual current waveshapes any component selection
(diode and capacitor) made based on the pulse current waveshapes will definitely work for the actual wave-
shapes also. It should be noted that approximations based on such engineering judgment will be encoun-
tered frequently in circuit design. Such approximations will enable one to obtain close form solutions to
many design parameters without compromising on the component ratings.
The peak, average and the rms currents that flow through the diodes and the rms current through the
capacitor are required to the calculated so that properly rated diodes and output capacitor can be selected.
Referring to Figure 3.14, it can be observed that
V2 = V1cos α
and therefore
V
α = cos −1 2 (3.39)
V1
Let Io be the average value of the load current, then
P
Io = o (3.40)
Vo-avg
where Vo,avg = Vo-avg = (V1 + V2 ) / 2.
Referring to Figure 3.14, the current i flows through the capacitor and also through the load. As the
capacitor current has zero average value under steady state, the load current Io is the average value of the
current i. Therefore,
⎛α⎞
Io = Im ⎜ ⎟ (3.41)
⎝π⎠
Substitution of Eq. (3.40) into Eq. (3.41) and re-arranging gives
2π Po
Im = (3.42)
α (V1 + V2 )
Diode Selection
1. The diode peak current rating should be greater than the value calculated by Eq. (3.42).
2. The rms value of current through diodes is given by I d,rms = I m α / 2π .
3. The average value of current through diodes, Id,avg = Ima /2p.
4. The PIV rating for the diodes should be greater than the maximum peak value that the input voltage
will reach. This is given as
Chapter 03.indd 113 3/12/2009 3:49:21 PM

⎛ %tol ⎞
PIV > V1− max = 2Vin-rms ⎜ 1 +
⎝ 100 ⎟⎠
5. The average and the rms currents through the diodes are needed to estimate the power dissipated in the
diodes. This will reflect in the selection of heat sinks for the diodes.
Capacitor Selection
1. The capacitor value selected should be greater than that calculated by Eq. (3.36). As the capacitor
charges up only in one direction, the voltage across the capacitor is unidirectional. Therefore an electro-
lytic capacitor should be selected.
2. The rms value of current through capacitor is
α ⎛ π −α ⎞
I Crms = ( I m − I o )2 + I o2 ⎜ (3.43)
π ⎝ π ⎟⎠
Equation (3.43) is obtained by referring to Figure 3.14. It can be observed that (Im – Io) current flows
through the capacitor during a out of p periods and Io value of current flows through the capacitor
during the remaining p − a out of p periods. Squaring the currents and integrating between the angle
limits mentioned above and taking the mean over p gives ICrms. The rms value of the capacitor current
is used to select the equivalent series resistance (ESR) rating for the capacitor.
3. The voltage rating of the capacitor should be greater than V1-max.
Remarks on the rectifier–capacitor input filter design:
1. The specifications for the design of the rectifier–capacitor input filter circuit consists of input voltage
minimum (Vm-min) and maximum (Vm-max) limits, desired output voltage peak-to-peak ripple (DVr),
output power (Po) and the input frequency (fs ).
2. The capacitor value is evaluated according to Eq. (3.36). It is essential to take care that the capacitor
value is evaluated for the minimum input voltage value.
3. The capacitor is an electrolytic capacitor as the voltage it supports is unidirectional. The voltage rating
of the capacitor selected should be greater than Vm-max. Either an aluminum electrolytic or a tantalum
capacitor is normally chosen.
4. The rms current rating required for the capacitor is calculated using Eq. (3.43). From the manufacturers’
datasheets, a capacitor having an rms current rating that is 1.5 to 2 times that calculated is
selected.
5. The average, peak and rms current rating of the rectifier diodes are calculated using Eqs. (3.40)–(3.42).
From the manufacturers’ datasheets, a diode having current ratings that are 1.5 to 2 times the values
calculated is selected.
6. The PIV seen by the rectifier diodes is Vm-max. Therefore, diodes having PIV ratings that are greater than
the value of Vm-max should be selected.
One should note that the design of rectifier-filter circuit with three-phase inputs is similar to that outlined
above. In the case of three-phase inputs, the rectifier is a three-phase full-wave bridge rectifier as shown in
Figure 3.15. The output Vo charges to the peak value of the line-to-line voltage. If Vm is the peak value of the
line-to-neutral voltage, then Vo will charge up to 3Vm . The capacitor design equation is same as Eq. (3.36)
except that fr is now six times fs. Thus, it is evident that for a given output power and peak-to-peak ripple,
the size of the capacitor is smaller with three-phase input source than that for a single-phase input source.
Chapter 03.indd 114 3/12/2009 3:49:21 PM

Rectifiers 115
3f R
line-to-
Y C R Vo
line
voltages B
Figure 3.15 Three-phase capacitor-filter rectifier circuit.
Many products are made for the global markets. The two most common types of mains input that the
AC–DC converter may be used with are 115 V and 230 V line inputs. A circuit diagram of a universal or
dual input range rectifier filter is shown in Figure 3.16. The charges on the capacitors C1 and C2 are balanced
by equalizing resistors connected across each capacitor. For 230 V line operation, the switch S is set at
position “b”. The input rectifiers are now configured as a normal full-wave bridge circuit. For 115 V line
operation, the switch S is set at position “a”. The input rectifier is now configured as a voltage doubler such
that Vo will have the same value as though operating from a 230 V line. While it is technically possible to
operate the input section as a bridge at both 230 V and 115 V, the post regulator, which in most cases is a
switching regulator, will have to be designed to operate over a much larger input voltage swing which would
significantly increase the cost. Therefore, by adopting the above strategy as indicated in Figure 3.16, the post
regulator needs to be designed only for 230 V line input.
io
C1
b
Vm sinwt C RL Vo
a S
C2
Figure 3.16 Dual input range rectifier filter.
Chapter 03.indd 115 3/12/2009 3:49:21 PM

Hold Time
Let Vh be the minimum voltage required by the load/post regulator to function. This implies that for Vo
greater than Vh, the load/post regulator will function as per specification. However, when Vo becomes less
than Vh, the load will not get the required minimum voltage and therefore will not become operational.
When the input power goes OFF, the time taken by the rectifier output voltage to come down to Vh from
the moment the input power went OFF is called the hold time (th). The concept of hold time can be visual-
ized from Figure 3.17.
If the rectifier output is powering up a digital controller board or any intelligent microcontroller or
microprocessor or digital signal processor board, during the hold time the digital controller will perform safe
shut down operations. Therefore, sometimes the output capacitor needs to be calculated based on the mini-
mum hold time requirement of the load. The worst case hold time occurs when the input mains voltage is
at a minimum, that is, Vm-min and the power goes OFF when the output capacitor voltage is at V2 value.
During the hold time, the output capacitor must supply a maximum amount of energy given by
ε h = Pot h (3.44)
1
ε h = C (Vm- min 2 − Vh2 ) (3.45)
2
where
⎛ %tol ⎞
Vm- min = 2 ⋅Vin-rms ⎜ 1 −
⎝ 100 ⎟⎠
and Vh is specified as per load requirements. From Eqs. (3.44) and (3.45) the capacitor value for a specified
hold time is given as
2 Pot h
C= (3.46)
Vm-min − Vh2
2
The greater of the two values, as calculated from Eqs. (3.36) and (3.46), should be used for selection of the
capacitor value such that both the ripple and hold time requirements are met.
Vo
Vh
th
Power
OFF
Figure 3.17 Illustration of hold time.
Chapter 03.indd 116 3/12/2009 3:49:22 PM

Rectifiers 117
Turn-ON Currents and Surge Limiting

In a capacitor input filter, the current flow is quite narrow and sharp (as indicated in Figure 3.14). If the AC
power is turned ON at or just before the peak of the input voltage waveform, the peak current is limited
only by line resistance (Rline), ESR of the capacitor(RESR ), rectifier diode dynamic resistance and the input
line and/or the leakage reactance of the transformer if present (XL).
If an input side transformer is used, then the total equivalent resistance of the transformer looking into
the secondary is
⎛N2⎞
Rsec + ( Rline + Rpri ) ⎜ s2 ⎟
⎜⎝ N p ⎟⎠
where Rsec and Rpri are the winding resistances of the secondary and primary windings of the transformer,
respectively; Rline is the line or conductor track resistance of the primary side; Ns and Np are the secondary
and the primary number of turns, respectively. The total series resistance Rs is given as
N s2
Rs = Rsec + ( Rline + Rpri ) + 2rd + RESR (3.47)
N p2
where rd is the dynamic resistance of the diodes. During the capacitor charge duration, two of the four
diodes of the full-bridge will conduct. Therefore, two diode dynamic resistances are included in series.
The total series inductance Ls is given as
⎛N2⎞
Ls = Lline ⎜ s2 ⎟ + Lleakage (3.48)
⎜⎝ N p ⎟⎠
Lline is the conductor inductance on the primary or the line side and Lleakage is the equivalent transformer
leakage inductance as seen from the secondary. It should be noted that if the transformer is not present then
Rsec and Rpri values will be zero and Ns/Np ratio will be unity in Eqs. (3.47) and (3.48). The rectifier–capacitor
circuit along with the series impedance is shown in Figure 3.18.
The total source inductance Ls is a function of the inductance in the input line and the leakage reactance
of the transformer referred to the secondary. Figure 3.18 shows an off-the-line bridge rectifier with the AC
voltage applied when its voltage value corresponds to the maximum. The dynamic equation during the time
the capacitor is charging is given as
di 1
dt C ∫
Vac = Rsi + Ls + idt (3.49)
From the above equation it may be observed that the startup current i drawn from the source is dependent
on the instantaneous amplitude of the input source at startup, the equivalent series circuit resistance Rs, the
equivalent series circuit inductance Ls and the output capacitor C. At startup as the capacitor voltage is not
yet built up, the current is primarily dependent on the component values of Rs and Ls. In off-the-line bridge
rectifier circuits wherein there is no input transformer, the startup current surges could be very high in low
Ls circuits. Especially for high-power circuits, the line inductance is very small due to the use of wide con-
ductors. In such cases, the startup surge current will depend primarily on only Rs. This could be detrimental
to the components of the rectifier–capacitor circuit. However, in many cases, the inductance of an EMI
filter at the input provides sufficient surge protection for the off-the-line rectifier circuit. The startup peak
Chapter 03.indd 117 3/12/2009 3:49:22 PM

i
Ls io
Rs
Vm sinwt C RL Vo
(a)
Vo
0 t
i First
cycle
inrush
(b)
Figure 3.18 (a) Rectifier filter with source inductance and resistance; (b) inrush current
in the first cycle.
current may be reduced by a factor of 10 if the power supply is turned ON at the zero crossing of the input
wave, as compared to turning ON the power supply at the peak of the input wave.
The substantial startup current surge may be more economical to limit than choosing higher current
diodes. Figure 3.19 shows few methods of limiting the startup surge current. In the step-start circuit shown
in Figure 3.19(a), Rs is in series with R1 through which current flows when the input switch is closed,
thereby limiting the inrush current. The energizing time of contactor, K1, is typically from one-to-three
cycles of the input frequency which allows the filter capacitor to become charged, after which R1 is shorted
by K1 contacts to eliminate power dissipation and voltage drop due to inclusion of R1.
Chapter 03.indd 118 3/12/2009 3:49:22 PM

Rectifiers 119
A negative temperature co-efficient power thermistor, as shown in Figure 3.19(b), can also be used to
limit the inrush currents. At turn-ON, the “several ohms” resistance in TH1 limits the inrush current which,
in turn, dissipates power in the thermistor. This power dissipation raises the temperature of the device and the
resistance drops to a low value for normal operation. However, the thermal time constant of the thermistor
must be considered. If a power supply has been operating at near no load for some time, the resistance of TH1
will be higher than at full load. When full load is applied, the voltage drop across TH1 will cause the output
to drop substantially or in the case of a regulated supply, the output may go out of regulation. If the power
supply is operating at full load where TH1 is very low in resistance and a short power interruption occurs
(long enough for the input filter capacitor to discharge), then a high inrush current will occur when the mains
voltage returns. This is because the thermistor resistance has not yet recovered to its high cold state value.
R1
C RL Vo
k1
(a)
TH1
C RL Vo
(b)
Soft start
circuit C RL Vo
(c)
Figure 3.19 Surge current-limiting circuits.
Chapter 03.indd 119 3/12/2009 3:49:23 PM

D1 D3
C RL
Vo
D5
Vm sinwt
D6
R1
Q1
D2 D4
R2 C1
(d)
Figure 3.19 Continued.

The circuit shown in Figure 3.19(c) utilizes a TRIAC Q1. The soft-start control circuit is phase con-
trolled at turn-ON to limit inrush current. Initially, the conduction period of each half-cycle is very short,
but then increases with time until full conduction is reached. Three precautions should be observed with
this circuit. First, the conduction period should increase linearly every cycle at turn-ON to prevent a DC
component to be applied to the transformer primary if a transformer is used. In such cases, it may be better
to locate the TRIAC on the secondary side. Second, the control circuit should have mains-interruption sens-
ing, with a time constant sufficient to permit soft-start operation when the mains again returns; otherwise,
a high inrush current will occur. Third, the control circuit should provide continuous turn-ON signal to Q1
during desired full conduction. Otherwise, Q1 will turn-OFF when the input current, which is discontinu-
ous, goes to zero.
An input rectifier and capacitor filter is shown in Figure 3.19(d) with an N-channel power MOSFET
inserted in the DC return line. Diodes D1–D4 rectify the input to charge C and supply power to RL when
Q1 is ON. The MOSFET has two main advantages over a bipolar transistor. First, the high input impedance
of the gate allows high-resistance values for the biasing resistor, thus minimizing power dissipation. Second,
the MOSFET is immune to forward-biased secondary breakdown, although the safe operating area (SOA)
must still be observed. At turn-ON, C1 charges exponentially through R1 to a final steady-state value deter-
mined by R2/(R1 + R2). C1 charges to a few volts before the gate threshold voltage is reached, at which time
Q1 begins to conduct in the active region and act like a resistor. C reaches full charge in a few cycles, deter-
mined by the time constant of R1R2C1/(R1 + R2). The addition of diodes D5 and D6 serves to decouple C1
from C during operation and allows C1 to discharge through R2 when input power is removed. If D5 and D6
were omitted and R1 were connected to C, the voltage across C (as C discharges) would maintain a charge
on C1 even though input power is momentarily removed. Thus, when input power is again applied, the
charge on C1 could be sufficient to bias Q1 into full conduction. Complete discharge of C1 during power
interruptions could be achieved by paralleling C1 with a PNP transistor such that the base is biased by a
voltage divider from D5, D6 to DC return. With the power ON, the transistor would be reverse-biased but
with power OFF, the transistor would conduct to rapidly discharge C1. Inserting a small resistor between the
source terminal of Q1 and the negative terminal of C1 can act as a current limiter. As source current increases,
Vgs decreases thus increasing Vds thereby limiting the inrush current.
Chapter 03.indd 120 3/12/2009 3:49:23 PM

Rectifiers 121
3.4 Power Factor
P ower factor ( pf ) gives the quality measure of a circuit. If the pf is unity, the input power drawn from the
source will be the load power and the power that gets dissipated in the various components. However, if
the pf is low, the input source should be rated for a much higher power than the required load power. This
would imply higher line losses resulting in lower efficiencies. Referring to Figure 3.14, the input voltage
waveform is sinusoidal, but the current is not. It should be noted that the average current required by the
load is Io but the peak current demand from the source is Im which is much higher than Io. This implies a
low pf for the rectifier–capacitor circuit. Further, as the line current flows only for a period corresponding to
a, there is a voltage drop in the line impedance only during this period. This leads to distortion in the input
applied voltage. For such cases, what is the pf ? And how is it computed and measured?
Consider a voltage source of an arbitrary waveshape that is periodic with period T. It is connected to a
pure resistive load R as shown in Figure 3.20. The current waveshape will exactly resemble the voltage wave-
shape. The entire power drawn from the source is given to the load R. This means that no power is used for
kinetic storage (in inductor) or potential storage (in capacitor) or returned back to the input. The power that
is given to load R is called the active power and the power that is either stored or returned back to the input
is called the reactive power. In the case of the circuit as described above the reactive power is zero. The power
that is given to the resistive load, Pres, is given by
T
1
T ∫0
Pres = v × i × dt (3.50)
where v and i are the instantaneous values of the current and voltage across the load resistor R.
If the input source is a voltage source, the reference waveshape for the source currents is that correspond-
ing to the voltage source, and if the input source is a current source, the reference waveshape for the source
voltage is that corresponding to the current source. For now, consider that the input source is a voltage
source and its waveshape is defined. Let this voltage source of defined waveshape be connected to an arbi-
trary load. The power delivered from the source is given as
T
1
T ∫0 i L
Pload = v × i × dt (3.51)
where vi is the instantaneous value of input source voltage that is across the arbitrary load; iL is the instanta-
neous value of the current flowing through the source for the specified arbitrary load.
Measure or estimate the peak current Im that flows from the source for the specified load. Replace the
load with a resistive load Rref that has a value which results in the same peak current through the source and
load. For a resistive load, the peak current occurs at the peak of the input voltage waveshape, Vm. This would
V R
Figure 3.20 A voltage source connected to a resistive load.
Chapter 03.indd 121 3/12/2009 3:49:24 PM

be the reference load for the input voltage source for a specified peak current Im. Thus the value of the refer-
ence load resistor R that should be selected for Im is
Vm
Rref = (3.52)
Im
where Vm is the peak of the input source voltage; Im is the measured or estimated peak current flowing
through the source for the specified arbitrary load.
The power that is delivered from the input source with the reference resistive load as determined above
is given as
T
1
T ∫0 i R
Pref = v × i × dt (3.53)
where Pref is the reference power delivered from the input source with the reference load; vi is the instanta-
neous value of input source voltage that is across the arbitrary load; iR is the instantaneous value of the cur-
rent flowing through the source for the reference resistive load.
The pf is defined as the ratio of the power delivered to the arbitrary load to the reference power delivered
to the reference resistive load wherein the same input peak current is maintained. Thus pf is given by
T T
1
T ∫0 i L ∫ vi × iL × dt
v × i × dt
Pload
pf = = T
= T0 (3.54)
Pref 1
T ∫0 i R ∫ vi × iR × dt
v × i × dt
0
The pf gives a measure of the departure of the arbitrary load from the resistive load. If pf is unity, the arbi-
trary load is equivalent to the resistive load. If pf is less than unity, it implies that there are either kinetic or
potential energy storage components in the arbitrary load.
EXAMPLE 3.1 RLC Load

Consider a series RLC load connected to a sinusoidal voltage source as shown in
Figure 3.21.
i R L
V C
Figure 3.21 Series RLC load.
Chapter 03.indd 122 3/12/2009 3:49:24 PM

Rectifiers 123
The current waveform leads or lags the voltage waveform by an angle q. One
can apply Eq. (3.54) to obtain the pf for this RLC load supplied from a sinusoidal
voltage source. Here
v i = Vm sin(ω t )
iL = I m sin(ω t − θ )
where
⎛ ωL⎞ ⎛ 1 ⎞
θ = tan −1 ⎜ ⎟ − tan −1 ⎜
⎝ R ⎠ ⎝ ωCR ⎟⎠
The reference load resistor Rref is given by
Vm
Rref =
Im
The pf for the series RLC load can be estimated using Eq. (3.54). This is given by
T T
∫ Vm sin(ωt )I m sin(ωt − θ )dt Vm I m ∫ sin(ωt )sin(ωt − θ )dt

pf = T 0 = 0
T
= cos(θ )
∫ Vm sin(ωt )Vm sin(ωt ) / Rref ]dt Vm I m ∫ sin(ωt )sin(ωt )dt
0 0
EXAMPLE 3.2 Rectifier–Capacitor Filter Load

Consider the rectifier–capacitor filter load that is supplied by the sinusoidal voltage
source. Here
v i = Vm sin(ω t )
iL is the discontinuous waveform i as shown in Figure 3.14. This current flows every
half-cycle for a period corresponding to a. The peak value Im of this current wave-
form can be measured using an oscilloscope.
From the estimate of the peak current that flows through the input source, the
reference load resistor Rref is given by
V
Rref = m
Im
and then
iR = I m sin(ω t )
From Eq. (3.54), the pf for the rectifier–capacitor filter load can be obtained as follows:
T
1
T ∫0 m
V sin(ω t )idt
Po + Ploss 2( Po + Ploss ) (3.55)
pf = T
= =
1 Vm I m / 2 Vm I m
T ∫0 m
V sin(ω t )I m sin(ω t )dt
Chapter 03.indd 123 3/12/2009 3:49:24 PM

where Ploss is the consolidated losses of the diodes, ESR of capacitor and the transformer
losses if transformer is present. In higher power circuits where the losses are negligible
compared to the output power, the pf for the rectifier–capacitor filter circuit reduces to
2 Po
pf = (3.56)
Vm I m
The value of Im has to be measured. However, one can obtain an approximate esti-
mate of the pf by using the relationship for Im given by Eq. (3.42) in Eq. (3.56):
α (V1 + V2 )
pf =
πV m
where
⎛V ⎞
α = cos −1 ⎜ 2 ⎟
⎝ V1 ⎠
Observe from the above representation that as the output ripple decreases, the time duration for which the
diodes conduct, a, will reduce. This implies that for the same output power, the current peak Im will be
larger, resulting in lower pf. The pf can be improved by
1. introducing an inductor in series;
2. increasing the diode conduction angle, by-means of controlled switching of power switches. This is
called the unity pf converter.
The former method will be discussed in the next section. However, the latter method will be discussed in
Chapter 10 after discussion of the control principles.
3.5 Rectifier–LC Filter
T he rectifier–LC filter consists of an inductor that is placed between the rectifier and the capacitor filter.
Since current cannot change instantaneously in the inductor, inrush currents at turn-ON are reduced.
Further, as the inductor has a smoothening effect on the current, the output voltage ripple is reduced as com-
pared to the rectifier–capacitor filter. Most importantly, introduction of the inductor widens the conduction
angle of the diodes, thereby improving the pf significantly. Alternately one may argue that the capacitor is a
potential energy storage element and the inductor being a kinetic energy storage element offsets the effect of
the capacitor such that the input source sees an impedance that is closer to the resistive load.
However, the major drawback is the inductor itself. The inductor will have to be designed for the supply
frequency of 50 Hz. This will make the size and cost of the inductor prohibitively large. This significant
drawback has prevented this topology from becoming popular in commercial equipments.
A single-phase full-wave rectifier with LC filter is shown in Figure 3.22(a). During the peak portions of the
voltage waveform, energy is stored in the inductor and during the valley portion of the voltage waveform, it is
transferred to the capacitor and load. The waveforms for the rectifier–LC filter are shown in Figure 3.22(b).
Referring to Figure 3.22(b), the rectified waveform contains an average voltage component and an AC
component. As the rectified waveform has even wave symmetry, it can be deduced that the AC component
contains only even harmonics. The Fourier series of the rectified waveform is given as
2Vm ⎛ 2 2 2 ⎞
Vrect =
π ⎜⎝ 1 − 3 cos 2ω t − 15 cos 4ω t − 35 cos 6ω t − …⎟⎠ (3.57)
Chapter 03.indd 124 3/12/2009 3:49:24 PM

Rectifiers 125
L i
i0
Vm sinwt
Vo = Vdc
(a)
V
Vm V0
2Vm/π
0
(b)
V0
Vm
2Vm /π
0 Idc-min i0
(c)
Figure 3.22 (a) Rectifier–LC filter circuit; (b) voltage waveform; (c) output voltage versus load.
where Vm is the peak value of the rectified sinusoidal voltage; w = 2pf is the fundamental radian frequency
and f is the line frequency.
The rectified waveform Fourier component representation given in Eq. (3.57) can be re-written as
Vrect = VDC-part + V AC-part (3.58)
Chapter 03.indd 125 3/12/2009 3:49:25 PM

where
2Vm
VDC-part = (3.59)
π
2Vm ⎡ ∞ ⎛ −2 ⎞ ⎤
V AC-part = ⎢ ∑ ⎜ 2 ⎟ cos nω t ⎥ (3.60)
π ⎢⎣ n = 2 ⎝ n − 1⎠ ⎥⎦
where n is the even harmonic number.

The LC-filter circuit is an averaging circuit or a low-pass circuit. It will remove the AC components and
pass only the DC part. Therefore, the output voltage Vo primarily contains only the average or DC part of
the rectified waveform.
2Vm
VDC = VDC-part = (3.61)
π
The AC part of the rectified waveform gets attenuated due to the capacitive reactance. Thus the AC part of
the rectifier waveform that appears at the output is given as
⎛ X C RL ⎞
Vripple = V AC-part ⎜ ⎟ (3.62)
⎝ X L + X C RL ⎠
Thus
Vo = VDC + Vripple
For a good LC-filter circuit, XC << RL and XC << XL. Therefore, the output ripple voltage given in Eq. (3.62)
can be approximated as
⎛X ⎞
Vripple = V AC-part ⎜ C ⎟ (3.63)
⎝ XL ⎠
The discussion above assumes that the current through the inductor is continuous. The desired output voltage
ripple is achievable only if the inductor current is continuous. As the load resistance RL increases, the DC or
the average part of the current through the inductor that flows through RL reduces. As RL is further increased
towards open circuit, the inductor current gradually reduces and becomes discontinuous. When this happens,
the output voltage will tend to rise toward the peak input voltage, Vm, as shown in Figure 3.22(c). Current
Idc-min is the load value at which the inductor current becomes discontinuous and each rectifier diode con-
ducts for less than 180o. The load Idc-min is the minimum load current for which the inductor is in continuous
conduction.
If the minimum load is specified, then the inductor value can be designed such that the inductor current
is continuous even at minimum load and the output voltage is as given by Eq. (3.61). The inductor has a
DC component and an AC component of current. For the current to be continuous, the DC component
2Vm / π RL-max must be equal to or greater than the AC component. As the capacitive reactance XC << XL,
the AC component of the inductor current that comprises predominantly of the second-harmonic compo-
nent is 4Vm / 3π X L . Thus
2Vm 4Vm
≥ (3.64)
π RL 3π X L
Chapter 03.indd 126 3/12/2009 3:49:25 PM

Rectifiers 127
where X L = 2ω L . From the above inequality, the following inductor value selection criterion is obtained.
RL-max
L≥ (3.65)
3ω
where RL-max corresponds to the load resistance value at minimum load. The input source considered for the
LC-filter circuit is a single-phase sinusoidal source. On rectification, the number of rectified pulses per
period is two. If a three-phase source is rectified then the number of rectified pulses per period is six. If such
a rectified output is fed to the LC filter, then the general equation for inductance value of polyphase rectifi-
ers can be obtained along similar lines as discussed and is given as
2 RL-max
L≥ (3.66)
p( p 2 − 1)ω
where p is the number of rectified pulses per fundamental line period.
From Eq. (3.63) it can be observed that a larger L will result in a smaller output voltage ripple. From
Eq. (3.66) it can be noted that when the load is light, RL is large and a large value of L is needed if the ripple
is not to be excessive. This means that the series inductance should have a large value at no load and may
be allowed to decrease as the load increases. Reactors can be designed wherein the inductance decreases as
the DC current through inductor increases by making use of the non-linear B–H magnetic characteristic of
the inductor core material. This can be a solution to the bulky and expensive constant inductance chokes
that need more iron to avoid saturation at larger currents.
Alternately, a bleeder resistor could be added across the output filter capacitor such that the minimum
load is guaranteed. In this case, the power dissipation and rating of the bleeder resistor and reduction in over-
all efficiency must be considered. The value of the bleeder resistor for a single-phase full-wave rectifier is
VDC
RB = (3.67)
I dc-min
Referring to Figure 3.22(c), the current Idc-min is the value of the load current at which the current in the
inductor becomes discontinuous or in other words the DC component of the inductor current is equal to
the peak of the AC component of the inductor current. At still lighter loads, the DC component becomes
lesser than the peak AC component and leads to discontinuous currents in the inductor. Therefore, by
introducing a bleeder resistor RB across the output capacitor C the rectifier filter will always see a minimum
load that is just enough to make the current in the inductor continuous. From Eq. (3.65) it can be argued
that the bleeder resistor that will provide the minimum load for continuous conduction of the inductor
current is
RB = 3ωL (3.68)
and for polyphase input sources, the general equation for the bleeder resistor is
p( p 2 − 1)ω L
RB = (3.69)
2
where p is the number of rectified pulses per fundamental line period.
Output Ripple
Equation (3.63) gives the amount of ripple voltage in the output. The inductor has a high reactance and
the capacitor has a low reactance at the harmonic frequencies and the following discussion considers only
Chapter 03.indd 127 3/12/2009 3:49:25 PM

the predominant second-harmonic effect, with negligible error in analysis. As the second harmonic is the
dominant effect with respect to the ripple voltage, the ripple waveshape can be considered to be sinusoidal.
Referring to Eqs. (3.60) and (3.63), the rms value of the output ripple voltage is given as
4Vm X C Vm
Vripple-rms = = (3.70)
3 2π X L 3 2πω 2 LC
The preceding equation provides the LC product, but individual component values must still be determined.
For a specified value of the minimum load, the value of L can be calculated. From the LC product, the value
of C can then be estimated.
Turn-ON Current
An important consideration with inductor input filters is the frequently under-damped nature of the circuit
at turn-ON. This is especially true of soft-start switch-mode power supplies because the input filter capacitor
will charge to a voltage higher than the peak input voltage and the “downstream” switching transistors will
experience a higher-than-normal voltage when switching commences. This effect has to be taken into
account while designing post regulators.
Design Summary
In the design of the rectifier–LC filter circuit, the input voltage, the minimum load requirements and the
output rms ripple are specified. Based on these inputs:
1. The LC value is evaluated using Eq. (3.70).
2. The value of L for minimum load is calculated using Eq. (3.65).
3. The value of C is calculated from LC product and the value of L. Ensure that XC << XL at the second
harmonic.
4. The rectifier diodes should be rated to carry the inductor current. Considering the second harmonic as
the dominant effect, this is given as
2Vm 4Vm
Id = +
π RL-min 6πω L
5. The PIV ratings for the diodes should be greater than Vm.
3.6 Controlled Rectifiers
I n the previous sections, it is seen that the rectified DC output voltage is dependent on the amplitude of
the AC input voltage and the load. The output voltage is therefore unregulated. As the input voltage
varies, so does the output voltage. The output voltage increases as the load approaches an open circuit. This
section presents phase-control topologies that may be employed to provide a regulated output. Line and
load regulation are achieved by using thyristors family of power switches to control the conduction angle of
the sine wave input in response to some desired control. TRIACs are used to control AC load voltages, as in
light dimmer assemblies. For high-power AC loads, parallel back-to-back SCRs can be used. Power supplies
utilize rectifiers and TRIACs or SCRs as the control element to provide a desired DC output from single-
phase or three-phase inputs.
Chapter 03.indd 128 3/12/2009 3:49:25 PM

Rectifiers 129
Single-Phase Power Circuits

Few topologies of single-phase power circuits are shown in Figure 3.23. The full-wave center-tap with resistive
load is used primarily for low-voltage outputs, as there is only one forward SCR drop. The resistive load of
Figure 3.23(a) is replaced by an inductor filter in Figure 3.23(b). In this case, a freewheeling diode D is
inserted to provide a path for the inductor current when the SCRs turn OFF. The bridge circuits of Figures
3.23(c)–(f ) are used primarily for higher output voltage applications. Figures 3.23(a)–(d) have the advantage
of common cathodes that permit firing pulse to be applied to both SCRs without need for isolation between
the two SCR gate drive circuits. They also permit simultaneous application of trigger pulses from a pulse
transformer with a single secondary if isolation is required. However, individual gate resistors to each SCR are
recommended. The anode terminal is normally the case of the SCR in stud-mounted devices. The circuit of
Figure 3.23(e) allows the SCRs to be mounted to a common heat sink without isolation, while the diodes
(cathode is normally the case of the device) are mounted to another heat sink without isolation. The two SCRs
are connected in series in Figure 3.23(f ), which has the advantage of eliminating the freewheeling diode since
this function is performed by the two bridge diodes in series. However, isolated trigger pulses to the SCRs
must be provided for the circuits in Figures 3.23(e) and (f ). In Figure 3.23(g), four SCRs are utilized. Here,
output polarity reversal is possible.
Resistive Load
Figure 3.24 shows the output voltage waveforms at various firing angles for resistive loads in single-phase
circuits given in Figure 3.23. Referring to Figure 3.24, the average value or the DC value of the rectified
waveform, triggered at firing angle a is given as
Vs = Vm sinwt RL Vdc
Vp
(a)
L
Vs = Vm sinwt C RL Vdc
Vp
(b)
Figure 3.23 Single-phase, phase-control circuits.
Chapter 03.indd 129 3/12/2009 3:49:26 PM

1φ
AC Vdc RL
input
(c)
1φ
AC C RL Vdc
input
(d)
1φ
AC Vdc RL
input
(e)
Chapter 03.indd 130 3/12/2009 3:49:26 PM

Rectifiers 131
1φ
AC Vdc C RL
input
(f)
1φ
AC Vdc C RL
input
(g)
1φ
AC
RL
input
(h)
Chapter 03.indd 131 3/12/2009 3:49:26 PM

π
1 V
VDC =
πα∫ Vm sin θ dθ = m (1 + cos α )
π
(3.71)
The outputs of circuits shown in Figures 3.23(a), (c), (e) have an average value as given in Eq. (3.71). The out-
puts of the rectifier for circuits in Figures 3.23(b), (d), (f ) also deliver an average value given above. Here the
outputs are filtered using an LC filter to remove the AC component in the rectified output. Circuits of
Figures 3.23(b), (d), (f ) have a freewheeling diode across the rectifier. Because of this, the rectifier output will
contain only positive portions, as the freewheeling diode will clamp the output to zero when it conducts.
Input
waveform
α = 0°
α
α = 60°
α = 90°
α = 120°
Figure 3.24 Voltage waveforms after rectification for various firing

angles for circuits given in Figures 3.23(a)–(f).
Chapter 03.indd 132 3/12/2009 3:49:27 PM

Rectifiers 133
Inductive Load
The circuit shown in Figure 3.23(g) has no freewheeling diode. Therefore, SCR commutation will take place
only when the inductor current goes below the holding current value. The waveforms for the phase-control
circuit of Figure 3.23(g) for various trigger angles are shown in Figure 3.25.
The output voltage with an inductive load is given as
π +α
1 2Vm
VDC =
π ∫ Vm sin θ dθ =
π
cos α (3.72)
α
Input
waveform
α = 0°
α
α = 60°
α = 90°
α = 180°
Figure 3.25 Voltage waveforms after rectification for various firing

angles for circuit given in Figure 3.23(g).
Chapter 03.indd 133 3/12/2009 3:49:27 PM

Figure 3.23(h) provides a topology wherein two SCRs are connected back to back. This topology is a useful
circuit for voltage control of resistive loads like heating applications and light dimmer applications. In low-
power applications, the two SCRs may be replaced by a single TRIAC to control the voltage to the load as
in conventional light dimmers. For high-power applications including inductive loads, the parallel back-to-
back SCRs offer additional current capability and overcome the problem of low-commutating dv/dt rating
associated with TRIACs. If the resistive load is replaced by a transformer whose secondary voltage is rectified
and filtered, a phase-controlled power supply results and regulation is achieved by controlling the firing
angle via feedback circuitry from the DC output to the TRIAC or SCR gates.
Three-Phase-Controlled Rectifier Circuits

Various three-phase-controlled power circuits are shown in Figure 3.26. The half-wave circuits of Figures
3.26(a) and (b) are seldom used due to poor utilization of transformer VA ratings. The bridge circuits of
Figures 3.26(c)–(f ) are more common.
A
1
B Vdc RL
2
C
3
(a)
L
A
1
B C Vdc
C
3
(b)
Figure 3.26 Three-phase, phase-controlled circuits.
Chapter 03.indd 134 3/12/2009 3:49:27 PM

Rectifiers 135
1 3 5
3φ
AC RL Vdc
2 4 6
(c)
1 3 5
3φ Vdc
C RL
AC
2 4 6
(d)
1 3 5
3φ RL Vdc
AC
2 4 6
(e)
Chapter 03.indd 135 3/12/2009 3:49:28 PM

1 3 5
3φ C RL Vdc
AC
2 4 6
(f)
Half-Wave Control
The half-wave circuits of Figures 3.26(a) and (b) give output voltage waveforms as shown in Figure 3.27. Referring
to Figure 3.27, the average or the DC component of the half-wave control circuit output voltage is given as
π
1 3V
VDC = 3 ∫
2π (π /6 )+α
Vm sin θ dθ = m cos(α + 30o )
2π (3.73)
AN BN CN AN BN CN
α = 0°
α AN BN CN AN BN CN
α = 30°
Figure 3.27 Voltage waveforms after half-control rectification for various firing
angles for circuits given in Figure 3.26(a) and (b).
Chapter 03.indd 136 3/12/2009 3:49:28 PM

Rectifiers 137
AN BN CN AN BN CN
α
α = 60°
AN BN CN AN BN CN
α
α = 90°
Reference
waveform
VAN
0
30° α
SCR 1 120°
SCR 2 120°
SCR 3 120°
Figure 3.28 Sequence of SCR firing.
Chapter 03.indd 137 3/12/2009 3:49:28 PM

where Vm is the line-to-neutral peak voltage. The sequence of SCR firing for the circuits of Figures 3.26(a) and (b)
is shown in Figure 3.28.
Full-Wave, Half-Control
The circuits in Figures 3.26(c) and (d) are full-wave bridges with half-control (only three SCRs are required)
and are suitable for resistive load where no regeneration is needed. The SCR cathodes are common which
allows a single non-isolated firing stage for triggering the SCRs. However, the output ripple frequency is
reduced from six times the lines frequency to three times the line frequency at delays angles a > 0 (refer Figure
3.29). The rectifier output voltage waveforms for full-wave half-control circuits are shown in Figure 3.29 for
various firing angles a. The average value of the output voltage for the full-wave half-control circuit is given as
3Vm
VDC = (1 + cos α ) (3.74)
2π
α
AB AC BC BA CA CB AB AC BC BA CA
α = 30°
α
α = 60°
Figure 3.29 Voltage waveforms after rectification for full-wave half-control for various firing
angles of circuits given in Figures 3.26(c) and (d).
Chapter 03.indd 138 3/12/2009 3:49:29 PM

Rectifiers 139
α
α = 90°
Reference
waveform
VAB
0
60° α
SCR 1 120°
SCR 3 120°
SCR 5 120°
Figure 3.30 Firing sequence for the SCRs.
where Vm is the line-to-line peak voltage. The firing sequence for the SCRs is as shown in Figure 3.30. The
diodes D2, D4 and D6 will conduct depending on the most negative line (A or B or C line) at any given
instant of time.
Full-Wave, Full-Control
Full-wave bridges with full control (six SCRs) are shown in Figures 3.26(e) and (f ). The firing sequence is
more complicated than that of half-control topologies since the SCRs require two gate signals each cycle.
Chapter 03.indd 139 3/12/2009 3:49:29 PM

The rectifier output voltage waveforms for various firing angles, a, are shown in Figure 3.31. The average
value of the output voltage for the full-wave full-control resistive load is given as
3Vm
VDC = cos α (3.75)
π
where Vm is the line-to-line peak voltage.

The firing sequence for the full-wave, full-control circuits of Figures 3.26(e) and (f ) is given in Figure
3.32.
Overlap and Overlap Angle

In the discussions till now, instantaneous commutations among SCRs were assumed. However, in practice
this will not be the case. In series with each SCR there will be an inductance. This inductance exists because
α
α = 30°
α
α = 60°
Figure 3.31 Voltage waveforms after rectification for full-wave full-control for
various firing angles of circuits given in Figures 3.26(e) and (f).
Chapter 03.indd 140 3/12/2009 3:49:29 PM

Rectifiers 141
α
α = 90°
Reference
waveform
VAB
0
60° α
SCR 1 120°
SCR 3 120°
120°
SCR 5
120°
SCR 2
120° 120°
SCR 4
60° 120°
SCR 6
Figure 3.32 Firing sequence for the full-wave, full-control circuits of Figures 3.26(e) and (f).
Chapter 03.indd 141 3/12/2009 3:49:30 PM

of source, line, load, etc. Owing to the presence of this series inductance, during commutation, the SCR
that is turning OFF will turn-OFF only after the current through its inductance reduces below the holding
current value. The current in the SCR that is turning ON will rise slowly depending on the value of the
inductance. During a small interval of the time both the commutating SCRs will be ON. This is called con-
duction overlap. The period in angle for which the conduction overlap occurs is called the overlap angle.
During overlap the output voltage reduces by the inductance voltage drop equal to L(di/dt) and this increases
with load current. It should be borne in mind that the presence of these series parasitic inductances that
cause the conduction overlap will result in loss of output regulation.
The rectifiers and especially the capacitor-filter rectifi- (EMI) issues. Therefore due to regulations, more
ers are rather ubiquitious as far as power electronic sys- and more power electronic systems are incorporat-
tems are concerned. Most power electronic systems ing unity pf correction circuits into the frontend
draw the power from the AC grid. They are first con- AC–DC rectifier circuits. The more popular fron-
verted to DC by the rectification mechanisms discussed tend boost converter for pf correction and the fron-
in this chapter to form the DC bus or the DC link for tend converters for inverter applications are
the system. In this respect, the AC–DC rectifiers play a discussed in Chapter 10.
pivotal role in most power electronic applications. Apart from the above problems, the issue of the
Though the capacitor-filter rectifier is volumet- turn-ON surge currents that is prevalent in capaci-
rically the most efficient and also least expensive, it tor-filter rectifier circuits is a serious drawback. Extra
suffers from low pf issues. This implies that the grid circuits have to be used to circumvent the problem
power from which the power is drawn has to be of startup surge currents. Notwithstanding the
rated many more times than the actual active load drawbacks mentioned above, the capacitor-filter
power. This leads to significant line loss. Further as rectifier is still the most simple and popular AC–DC
the peak currents flow in a small duration near the rectifier circuit that is in use till date. The rectifier
voltage peaks, the line drops due to the current LC filter is not as popular due to the bulky and
flow distorts the voltage waveforms causing a flat- costly inductor. In higher power circuits, controlled
tening of the sinusoidal voltage waveshape near the rectification followed by capacitor-filter circuits are
peaks. This leads to higher total harmonic distor- used. Here the trigger angle or the firing angle gives
tion and conducted electromagnetic interference a measure of control on the startup surge currents.
1. Consider the capacitor-filter rectifier circuit as b. Simulation in SciLAB
shown in Figure 3.33. The input is a single-phase c. Hardware bread-boarding
source that is derived from a 230 V mains. The
Tasks for study:
input may be given from an autotransformer to
(a) Rig up the circuit as shown in Figure 3.33
obtain different input voltage amplitudes.
and plot Vi and i versus time.
Mode of implementation: The above circuit (b) Why is there ringing on the current wave-
can be studied by form?
a. Simulation in Spice (c) What is the series impedance of the circuit?
Chapter 03.indd 142 3/12/2009 3:49:30 PM

Rectifiers 143
c iR
+ i
1 3
ic
Vi ∼ a
+
Cf RL Vo
b
∼
4 2
−
Figure 3.33 Single-phase capacitor-filter rectifier circuit.
(d) Plot Vi and Vo versus time. What is the (j) Under what conditions do you get maxi-
ripple? How does ripple depend on load, mum peak current to flow through the
capacitor Cf and frequency of input wave- diodes?
form? (k) Measure the input voltage and current and
(e) Measure the current and voltage waveform estimate the power factor.
across the rectifier diode. (l) What is the effect of change in Cf value on
(f ) Estimate the average and rms currents the input current and power factor?
through the rectifier diode. Calculate the
2. Consider the three-phase capacitor-filter recti-
diode power dissipation.
fier circuit as shown in Figure 3.34. The input is
(g) What should be the peak current rating of
a three-phase source that is derived from a 400 V
the diode?
mains. The input may be given from a three-
(h) Change the initial charge voltage on the
phase autotransformer to obtain different input
capacitor Cf. What is the effect on the
voltage amplitudes. The circuit shown consists
input surge current?
of a basic single-phase full-bridge rectifier.
(i) Change the phase angle of the input Vi at
When switches “S1” are put in position 1, then
start up. Observe the effect on current i.
one more diode arm gets connected. This trans-
What happens and why?
forms the circuit into a three-phase full-bridge
0
i iR
S1 + ic +
1 C1
a 0
b c RL Vo
Va Vb Vc
S2 +
1 C2
0
S1 −
S1
1 0 1
Figure 3.34 Three-phase capacitor-filter rectifier circuit.
Chapter 03.indd 143 3/12/2009 3:49:30 PM

rectifier circuit. Note that when “S1” are in (j) What is the effect of the S2 switch on the
position 0, Va and Vc are disconnected by S1 diode currents for a given load?
thereby applying only a single-phase voltage (k) What is the effect of unequal C1 and C2 on
across the b-c diode bridge. During single-phase the output voltage?
operation of the circuit, switch “S2” gives a (l) How is charge equalization done for C1
doubler effect. and C2? [Hint: Connect resistors across
each capacitor and observe.]
can be studied by 3. Consider the single-phase LC-filter rectifier
a. Simulation in Spice circuit as shown in Figure 3.35. The input is a
b. Simulation in SciLAB single-phase source that is derived from a 230 V
c. Hardware bread-boarding mains. The input may be given from an auto-
Tasks for study: transformer to obtain different input voltage
(a) Keep S1 in position 0 and S2 in position 0. amplitudes. Ls and Rs are the equivalent circuit
Observe Vbc, i, iR, iC and Vo. series inductance and resistance, respectively,
(b) Keep S1 in position 0 and S2 in position 1. that represent the non-idealities; Lf is the filter
Observe Vbc, i, iR, iC and Vo. What is the inductance of the LC filter.
effect of switch S2? What is its application? Mode of implementation: The above circuit
(c) Keep S1 in position 1 and S2 in position 0. can be studied by
Observe Va, Vb, Vc, i, iR, iC and Vo. a. Simulation in Spice
(d) What is the ripple frequency of the output b. Simulation in SciLAB
when S1 is in position 1 and in position 0? c. Hardware bread-boarding
(e) When S1 is in position 1, what is the worst
case ripple? Tasks for study:
(f ) What should be the peak current rating of (a) Rig up the circuit as shown above and plot
the diode? Vi and i versus time.
(g) Change the initial charge voltage on the (b) Plot the rectified Vi and Vo versus time.
capacitors. What is the effect on the turn- (c) Vary the load resistor RL and plot Vo versus
ON surge current? load Io.
(h) Observe the diode currents when S1 is in (d) What is the minimum load current for
position 1 and in position 0. which the inductor current is continuous?
(i) Estimate the average and rms currents of (e) Incorporate a bleeder resistor across the
the diode when S1 is in position 1 and in capacitor and plot the efficiency versus load
position 0. with and without the bleeder resistor.
LS RS Lf iR
+ iL
1 3 ic
Vi ∼ a +
Cf RL Vo
∼ b
4 2
−
Figure 3.35 Single-phase LC–rectifier circuit.
Chapter 03.indd 144 3/12/2009 3:49:31 PM

Rectifiers 145
(f ) What is the ripple? How does a ripple (i) What should be the peak current rating of
depend on load, inductor, capacitor and the diode?
frequency of input waveform? (j) Compare the turn-ON surge current with
(g) Measure the current and voltage waveforms the capacitor-filter rectifier for a given load.
across the rectifier diode. (k) Measure the input voltage and current and
(h) Estimate the average and rms currents estimate the power factor.
through the rectifier diode. Calculate the (l) What is the effect of change in Lf and Cf
diode power dissipation. value on the input current and power factor?

1. Single-phase rectifier circuit topologies are 11. In the case of the bridge rectifier circuit, there
designed for applications wherein the input is no need for a separate diode.
source is the .
12. The load ripple frequency for a half-wave recti-
2. Three-phase rectifier circuit topologies are fier system is the input source
designed for applications wherein the input frequency.
source is a .
13. The load ripple frequency for a full-wave recti-
3. For inductive loads, an additional diode called fier system is the input source
is needed to provide a path for the frequency.
inductor current when the rectifier diode is OFF.
14. In a three-phase half-wave rectifier circuit, the
4. Use of a is compulsory for the full- secondary windings of the transformer carry
wave center-tapped rectifier. the currents as that flowing through
5. The PIV for the diodes of a full-wave center- the diodes.
tapped rectifier is the PIV for the 15. The load ripple frequency for a three-phase
diodes of the half-wave or full-bridge rectifiers half-wave rectifier system is the
for a given load voltage requirement. input source frequency.
6. In a full-wave center-tapped rectifier, the cur- 16. The load ripple frequency for a three-phase
rent through the secondary of the transformer full-wave rectifier system is the input
is a wave rectified waveform. source frequency.
7. In a full-wave center-tapped rectifier, the current 17. Capacitor input filters are the most
through the rectifier diodes of the transformer is efficient means of filtering rectified sine waves
a wave rectified waveform. and storing energy.
8. In the full-wave bridge rectifier there will be 18. The output voltage ripple for a single-phase
diode drop in the bridge rectifier capacitor-filter rectifier is the input
topology as compared to the half-wave or full- frequency.
wave center-tapped rectifier topologies.
19. The output ripple frequency for three-phase
9. In a full-wave bridge rectifier with transformer, half-wave capacitor-filter rectifier is .
the current through the secondary of the trans-
former is a waveform. 20. The output ripple frequency for three-phase
full-wave capacitor-filter rectifier is .
10. In a full-wave bridge rectifier, the current
through the rectifier diodes is a wave 21. In a three-phase capacitor-filter rectifier circuit
rectified waveform. that is supplied from three-phase mains, the
Chapter 03.indd 145 3/12/2009 3:49:31 PM

peak capacitor voltage is time the 29. In LC–rectifier filters, when the load is light, a
input mains phase voltage. value of L is needed if the ripple is
to be small.
22. The turn-ON surge current in the capacitor-
filter rectifier is limited by the line resistance, 30. Power supplies utilize TRIACs or SCRs as the
of the capacitor, rectifier diodes device to provide a desired dc output
and the input line and/or the leak- from single-phase or three-phase inputs.
age of the transformer if present. 31. In single-phase half-controlled SCR bridge
23. Low power factor implies line rectifier configurations, the rectifier output will
losses. contain only portions.
24. Unity power factor occurs when the load is 32. In single-phase full-controlled SCR bridge rec-
purely . tifier configuration, the SCR commutation will
take place only when the inductor current goes
25. In a capacitor-filter rectifier circuit, lower the below the value.
power factor, will be the
33. The reference waveform for sequencing the firing
current drawn from the input source.
of the SCRs for the three-phase half-wave half-
26. The rectifier–LC filter does not have the prob- controlled rectifier is the waveform.
lem of turn-ON due to the presence 34. The reference waveform for sequencing the
of the . firing of the SCRs for the three-phase full-
27. The major drawback of the LC-filter rectifier is wave half-controlled rectifier is the line-to-line
the that has to be designed for the waveform.
supply frequency. 35. Three-phase full-wave full-controlled rectifier
28. For a good LC-filter circuit, and consists of SCRs in bridge configu-
Xc << XL. ration.
1. Where are rectifiers used? 7. Explain the operation of the three types of
single-phase rectifier circuits with illustrative
2. How are rectifiers classified?
waveforms.
3. What are the three basic rectifier configurations?
8. A delta-star transformer is used for a three-
4. For a half-wave rectifier, draw the load voltage phase full-wave rectifier system. In the
and current waveforms for an inductive load secondary of the three-phase transformer, if the
and derive the load power equation. neutral point is available, draw and explain the
schematic to generate positive and negative
5. For a full-wave center-tapped rectifier, draw
voltages with respect to the neutral point using
the load voltage and current waveforms for an
a single three-phase bridge-rectifier topology.
inductive load and derive the load power
equation. 9. Discuss the operation of the capacitor input
filter rectifier.
6. For a full-wave bridge rectifier, draw the load
voltage and current waveforms for an inductive 10. What are the factors that affect the output
load and derive the load power equation. ripple of the capacitor-filter rectifier?
Chapter 03.indd 146 3/12/2009 3:49:31 PM

Rectifiers 147
11. What is hold time? How does one design the 16. Discuss the design considerations for the
value of capacitor for a specified hold time? inductor of an LC–rectifier filter.
12. Discuss the turn-ON surge current problem in 17. Explain the terms minimum load current
capacitor-filter rectifier circuits. What are the and bleeder load resistance with respect to a
parameters affecting the amplitude of the turn- rectifier–LC filter.
ON surge current? How is this problem solved
18. What is the ripple content in the output volt-
in practice?
age of an LC–rectifier filter?
13. What are the methods for improving the power
19. Discuss the operation of half- and full-control
factor of AC–DC rectifiers?
of single-phase controlled rectifiers.
14. What are the benefits of the LC–rectifier filter?
20. Distinguish between half-wave half-control,
15. What are the disadvantages of the LC–rectifier full-wave half-control and full-wave full-con-
filter? trol in three-phase-controlled rectification.
| PROBLEMS
1. A half-wave rectifier is supplying 50 W to a secondary to primary phase winding turns ratio
resistive load by drawing power directly from a is 0.5. What are the average load voltage and
230 V rms mains grid. What are the average current? What are the rms load voltage and
load voltage and current? What are the rms current? What are the average and rms currents
load voltage and current? of the rectifier diodes?
2. A full-wave center-tapped rectifier is supplying 7. A 1000 W resistive load is supplied by a star–
50 W to a resistive load by drawing power from star transformer-isolated three-phase full-wave
a center-tapped transformer that is connected rectifier. The rectifier is drawing power directly
to the 230 V rms mains grid. The turns ratio of from the 400Vrms three-phase mains grid. The
the center-tapped transformer is unity. What secondary to primary phase winding turns ratio
are the average load voltage and current? What is 0.5. What are the average load voltage and
are the rms load voltage and current? current? What are the rms load voltage and
current? What are the average and rms currents
3. For Problem 2, calculate the rms value of the
of the rectifier diodes?
primary and secondary winding currents.
8. In a capacitor-filter rectifier that is supplied
4. A full-wave bridge rectifier is supplying 50 W
from 230 V, 50 Hz mains, a 100 μF output
to a resistive load by drawing power directly
capacitor discharges to 250 V every half-cycle.
from the 230 V rms mains grid. What are the
What is the energy given to the load by the
average load voltage and current? What are the
capacitor every input voltage cycle?
rms load voltage and current?
9. Calculate the capacitor value of a capacitor-
5. For Problem 4, calculate the rms value of the
filter rectifier that is supplied from 230 V, 50 Hz
primary and secondary winding currents.
mains wherein the peak-to-peak ripple is 50 V
6. A 1000 W resistive load is supplied by a star– and the load is 500 W.
star transformer-isolated three-phase half-wave
10. For Problem 9, calculate the load current and
rectifier. The rectifier is drawing power directly
the peak, rms and average values of the current
from the 400Vrms three-phase mains grid. The
Chapter 03.indd 147 3/12/2009 3:49:31 PM

through the rectifier diodes. Calculate also the are triggered at firing angles of 45°, 90°, 120°
rms current through the capacitor. and 180°?
11. For Problem 9, estimate the power factor of the 17. A single-phase full-controlled SCR bridge
capacitor-filter rectifier circuit. rectifier is supplying an RL load from a single-
phase 230 V mains. What is the average value
12. For an LC–rectifier filter that is supplied from
of the load voltage if the SCRs are triggered at
the 230 V, 50 Hz mains, the load power is 500 W,
firing angles of 45°, 90°, 120° and 180°?
what is the output voltage if the current
through the inductor is continuous? 18. For a three-phase half-wave half-controlled recti-
fier which is supplying a load from a three-phase
400 V rms line-to-line mains grid, find the trigger
the 230 V, 50 Hz mains, if the maximum value
angle at which the average load voltage is 75V.
of the load resistance is 50 Ω, estimate the
minimum value of the filter inductor required. 19. For a three-phase full-wave half-controlled
rectifier which is supplying a load from a three-
phase 400 V rms line-to-line mains grid, find
the three-phase 400 V mains, if the maximum
the trigger angle at which the average load
value of the load resistance is 50 Ω, estimate the
voltage is 400 V.
minimum value of the filter inductor required.
20. For a three-phase full-wave full-controlled
15. For Problem 13, if the rms value of the output
rectifier which is supplying a load from a three-
ripple is not to exceed 50 V, calculate the
phase 400 V rms line-to-line mains grid, find
capacitor value of the LC filter.
the trigger angle at which the average load
16. A single-phase half-controlled SCR bridge voltage is 400 V.
rectifier is supplying a resistive load of 1000 W
from a single-phase 230 V mains. What is the
average value of the load voltage if the SCRs
| ANSWERS
Fill in the Blanks
1. single-phase grid 14. same 25. higher; peak
2. three-phase grid 15. thrice 26. surge currents; inductor
3. freewheeling diode 16. four times 27. inductor
4. transformer 17. volumetrically 28. Xc << RL
5. twice 18. twice 29. large
6. half 19. 150 Hz 30. control
7. half 20. 300 Hz 31. positive
8. two 21. 3 32. holding current
9. full 22. equivalent series resistance; 33. line-to-neutral A phase
10. half dynamic resistance; reactance 34. VAB
11. freewheeling 23. higher 35. six
12. same as 24. resistive
13. twice
Chapter 03.indd 148 3/12/2009 3:49:31 PM

DC–DC Linear Regulators
Learning Objectives
CHAPTER
4
understand the operating principle of linear regulators.
design and apply the various types of linear regulators.
analyze and characterize the linear regulators.
T he DC−DC linear regulators convert the input DC voltage to an output DC voltage wherein the
output voltage is regulated for variations in input voltage, temperature and output load. The regulation
is achieved by operating the power semiconductor devices in the linear region rather than as switches. This,
of course, leads to large power dissipation in the power devices and consequent reduction in efficiency of the
regulator.
There are, in general, two component sets in the operation of the DC−DC linear regulators, namely, (a)
the voltage scaling circuit and (b) the output voltage regulation circuit. However, the operations of these two
component sets are interlinked and dependent on each other. Hence the designs of these two component
sets are not decoupled from each other.
The output DC voltage value is always lesser than the input unregulated DC voltage value, that is, the
scaling circuit will only attenuate the input voltage but will not amplify the input voltage. In other words, the
gain in the scaling circuit is less than unity. Even though the linear regulators have low efficiency, the quality
of the output voltage with respect to the variations in parameters like input voltage, temperature and load are
excellent and by far better than that offered by any switched-mode converter. The output voltage can be
designed to be almost ripple-free in the case of the linear regulators. The major disadvantage of the linear
regulators is in term of efficiency and as a consequence the very low resulting volumetric power density.
4.1 Generic Linear Regulator
T he schematic of a generic linear regulator is shown in Figure 4.1. It consists of two main resistive com-
ponents: (a) Rs, the series component that performs the function of dropping the voltage across itself to
obtain a specified scaling at the output; (b) Rsh, the shunt component that generally performs the function
of the regulation. Both Rs and Rsh are variable resistances.
Rs is an essential component is all linear regulators. However, Rsh may not be present in all linear regula-
tor topologies. If the output voltage regulation is performed by varying Rs, such regulators are classified as
series regulators and if it is performed by varying Rsh then such regulators are classified as shunt regulators.
Chapter 04.indd 149 3/12/2009 3:50:01 PM

Io
Iin
Rs Ish
Vi Rsh RL Vo
Linear regulator
Figure 4.1 Schematic of a generic linear regulator.
Operating Principle
The operation of the generic linear regulator depends on whether it is the series regulator type or the shunt
regulator type. In the series regulator, as the output voltage Vo increases, the series resistance Rs is increased.
The drop across Rs increases thereby bringing down the Vo. Likewise as Vo decreases, Rs is decreased. The drop
across Rs decreases and brings up Vo. In the shunt regulator, the shunt component is used for regulating Vo.
Referring to Figure 4.1, one can observe that
Vo Vo
I in = I sh + I o = + (4.1)
Rsh RL
Vo = Vi − I in Rs (4.2)
Let Pi be the power drawn from the input source. Using Eqs. (4.1) and (4.2), Pi is given as
Pi = Vi I in = (Vo + I in Rs )I in
= Vo I in + I in2 Rs
2
⎛ R + RL ⎞ 2 ⎛ Rsh + RL ⎞
= Vo I o ⎜⎜ sh ⎟⎟ + I o ⎜⎜ ⎟⎟ Rs (4.3)
⎝ Rsh ⎠ ⎝ Rsh ⎠
2
⎛ R ⎞ 2⎛ R ⎞
= Vo I o ⎜⎜ 1 + L ⎟⎟ + I o ⎜⎜ 1 + L ⎟⎟ Rs
⎝ Rsh ⎠ ⎝ Rsh ⎠
The output power (Po) delivered to the load is
Po = Vo I o (4.4)
The efficiency of the linear regulator is defined as the ratio of the power delivered to the output load to the
power drawn from the input source. This is given as
Chapter 04.indd 150 3/12/2009 3:50:02 PM

DC–DC Linear Regulators 151
Po Vo I o
η= = (4.5)
Pi Vo I o [1 + ( RL / Rsh )] + I o2 [1 + ( RL / Rsh )]2 Rs
If Rsh is very high or tending to infinity, then Iin = Io. From Eq. (4.5) it can be observed that
Vo
η=
Vi
4.2 Linear Regulator Topologies

Shunt Regulator
Figure 4.2 shows the circuit schematic of a Zener shunt regulator. The shunt component is a Zener diode
that is operated in the reverse breakdown region. The series component is a fixed resistor Rs.
Ideal Zener
Figure 4.3 shows the static characteristic of the Zener diode. The Zener diode is the shunt element that is
connected across the load. According to the idealized static characteristic of the Zener diode as shown in
Figure 4.3(a), one can observe that the voltage across the Zener diode is a constant at Vz. Thus Vo is clamped
to Vz irrespective of the changes in the input voltage and the load. Thus Vo is regulated by virtue of the
Zener diode being operated in the breakdown region. Referring to Figure 4.2,
I in = I z + I o (4.6)
Vi − Vz
I in = (4.7)
Rs
Vo = V z (4.8)
From Eq. (4.8), it is seen that Vo is independent of the load Io and the input Vi. Thus the load voltage is reg-
ulated as Vz is constant. Under no-load condition, Io = 0 and the whole of Iin flows into Iz. As the load
increases, a part of Iin commutates from the Zener to the load, that is, Iz decreases and Io increases. Likewise,
if Io decreases, Iz will increase to compensate for the decrease in Io, thus regulating the output. The Zener
Ii Io
Rs Iz
Vi Vz RL Vo
Figure 4.2 Zener shunt regulator.
Chapter 04.indd 151 3/12/2009 3:50:02 PM

i i
Vz
Vz
0 u 0 u
1/rz Iz
(a) (b)
Figure 4.3 (a) Ideal static characteristic of the Zener diode;

(b) piece-wise linear static characteristic of the Zener diode.
current Iz is maximum under no-load condition when Io is zero. Under this condition, the input current Iin
flows through the Zener. Thus,
Vi-max − Vz
I z-max = (4.9)
Rs
Iz-max is limited only by the power rating of the Zener diode. The other limiting condition is when the input
current completely flows through the load under full-load condition. During this condition,
I o = I in and I z = 0 (4.10)
The Zener should be maintained in the breakdown region even under minimum input voltage condition.
Therefore, the maximum load current allowable is
Vi-min − Vz Vz
I in = I o-max = = (4.11)
Rs RL-min
Problem 4.1
Consider a shunt regulator circuit wherein the input voltage Vi varies between 9 and 13 V and the Zener diode
used is a 5 V, 400 mW device. What is the range of the load resistance?
Solution
The maximum possible Zener current is
Pz 400 mW
= = 80 mA
Vz 5V
Chapter 04.indd 152 3/12/2009 3:50:03 PM

The operating input current should be less than 80 mA. Let 60 mA be the maximum Zener operating cur-
rent Iz-max. Then from Eq. (4.9),
V − Vz 13 − 5
Rs = i-max = ≈ 133 Ω
I z-max 60 mA
The power dissipated in the series resistor Rs is
(13 − 5)2
PRs = ≈ 0.48 W
133
From Eq. (4.11), the minimum value of the load resistance RL should be calculated such that even at the
lowest input voltage and load, the Zener is in reverse breakdown. Thus,
RsVz 133 × 5
RL-min = = ≈ 166 Ω
Vi-min − Vz 9−5
On the other hand, the maximum value of the load resistance can be infinite, that is, open circuit when all
the input current flows through the Zener diode. Thus the range of the load resistance is limited to
166 ≤ RL < ∞
Non-Ideal Zener
The Zener diode does not have ideal static characteristics as shown in Figure 4.3(a). The piece-wise linear
approximation of the actual static characteristic of the Zener diode shown in Figure 4.3(b) is much closer to
reality. The Zener has a non-zero operating resistance rz in the breakdown region as shown in Figure 4.3(b).
This first-level non-ideality is introduced into the shunt regulator circuit as indicated in Figure 4.4.
Owing to the presence of rz, the output voltage is no longer identical to the Zener breakdown voltage Vz.
It is now given as
Vo = Vz + I z rz (4.12)
From Eq. (4.12) it is evident that the output voltage is dependent on Iz. However, Iz varies from almost zero
at full-load condition to full load current value during no-load operation. Therefore, the regulation of the
output with respect to variations in the load is poor due to the voltage drop across rz.
Iin Rs Iz Io
Vi rz RL Vo
Vz
Ro = rz //Rs
Figure 4.4 The Zener diode is replaced by first-level non-idealities.
Chapter 04.indd 153 3/12/2009 3:50:03 PM

Io /b
Q
Rs
Iz
Vbe Io
Vi Vz
RL Vo
Figure 4.5 Modified shunt regulator.
Modified Shunt Regulator

The regulator shown in Figure 4.5 tries to reduce the effect of the Zener diode non-ideality rz on the output
voltage regulation.
For the modified shunt regulator of Figure 4.5, the output voltage Vo is the Zener voltage Vz minus the
drop across the base−emitter junction of the transistor Q. Thus,
Vo = Vz − Vbe (4.13)
In this circuit, the maximum Zener current is limited to I o / β instead of Io for the circuit of the
Figure 4.4. As the Zener current variation is reduced by b, the drop across the Zener resistance rz is also very
much diminished. Thus, the output voltage including the Zener non-ideality is given as
Vo = Vz + I z rz − Vbe (4.14)
where maximum value of Iz is Io/b.

In this topology, even though the output voltage regulation with respect to variations in the load
is much better as compared to the circuit of Figure 4.4, the outp ut voltage is dependent on Vbe of the
transistor Q. The Vbe of the transistor varies with temperature at the rate of about −2 mV/°C rise in
temperature. However, if the Zener is chosen to have a positive temperature coefficient then this would
compensate for the variations in Vbe due to temperature, thereby making the output voltage Vo independent
of temperature.
Series Regulator
The series regulator is a linear regulator wherein the series component is controlled. To control the series
component, a controllable resistor has to be used as the series-pass element. A transistor or bipolar junction
transistor (BJT) operating in the linear region can be used as a series-pass element whose base current can be
controlled to change the resistance presented to the input source. Figure 4.6(a) shows the basic schematic of
the series regulator with the series-pass component being a transistor Q.
Figure 4.6(b) shows the series regulator wherein the base drive for the series-pass transistor Q is being
controlled by the operational amplifier (op-amp). The output voltage Vo is attenuated by two resistors R1
and R2 and fed to the “−” terminal of the op-amp. The “+” of the op-amp is fed from a Zener voltage refer-
ence. The output of the op-amp drives the base of the transistor Q. If the output voltage increases, the “−”
terminal of the op-amp will increase. As the “+” terminal is at a constant reference potential, the output of
the op-amp will decrease the drive to the base of the transistor Q. This will increase the resistance presented
Chapter 04.indd 154 3/12/2009 3:50:03 PM

Vi RL Vo
(a)
Q
Rz R1
−
Vi RL Vo
+
R2
Vz
(b)
Io
Q1
Rb
R1
Rz
Vi Q2 RL Vo
R2
Vz
(c)
Figure 4.6 Series regulator: (a) Series regulator using transistor as the series pass; (b) the
series-pass transistor is controlled by an op-amp; (c) the series pass is controlled by
another transistor amplifier.
by Q to the input and thereby increase the drop across the collector to emitter of the transistor Q. As a con-
sequence, the output voltage will decrease and be brought back to the equilibrium state. All these events
occur simultaneously at an instant. Likewise if the output voltage were to decrease, the drive to Q will
increase thereby decreasing the drop across the collector−emitter of Q. This will tend to increase the output
voltage and bring it back to the equilibrium state.
Chapter 04.indd 155 3/12/2009 3:50:04 PM

The circuit of Figure 4.6(c) is similar to that of Figure 4.6(b) except that the op-amp is replaced by a
transistor (BJT) amplifier Q2 as shown. The Zener voltage reference is fed to the emitter terminal of Q2 and
the attenuated output voltage is fed to the base terminal of Q2. The transistor Q1 acts as the series-pass transis-
tor. The base of the transistor Q1 is at potential (Vo + Vbe1). Thus, the current through the bias resistor Rb is
Vi − Vo − Vbe1
I Rb = (4.15)
Rb
IRb can be considered as a current source that diverts the current between the base of Q1 and the collector of
Q2 depending on the output voltage.
If the output voltage decreases, then the base−emitter potential (Vbe2) of Q2 decreases. This will reduce
the collector-current flow of Q2. As a consequence, the portion of IRb that will be diverted to the base of Q1
will increase, resulting in reduced collector−emitter resistance of Q1. This will decrease the drop across the
collector−emitter of Q1 and will make the output voltage to increase and be brought back to the equilibrium
state. In a similar manner, the output voltage regulation will occur when the output voltage increases.
The resistances R1 and R2 are chosen based on the following governing equation:
⎛ R2 ⎞
Vo ⎜ ⎟ = Vz + Vbe2 (4.16)
⎝ R1 + R2 ⎠
Allow a current of about 1% of the full load current to flow through R1 and R2. This constraint and Eq. (4.16)
can be used to decide the values of R1 and R2. Rz is chosen such that the current flowing through the Zener
diode biases it in the breakdown region of the static characteristic curve. Rb is chosen based on Eq. (4.15) with
the nominal value of the input voltage Vi.
Negative Voltage Regulator

In the series regulator of Figure 4.6, the negative rail is used as the circuit ground. However, if one has to use
the positive rail as the circuit ground, then the schematic is modified as shown in Figure 4.7. The positive
output voltage series regulators have the series-pass component in the positive rail and the voltages are mea-
sured with reference to the negative rail as indicated in Figure 4.7(a). The NPN series-pass transistor Q is
replaced by an equivalent PNP series-pass transistor as shown in Figure 4.7(b). In this figure, the series-pass
element is included in the negative rail and the voltages are measured with respect to the positive rail.
Therefore the voltages are negative in the circuit of Figure 4.7(b).
Vi RL Vo
Vi RL Vo
(a) (b)
Figure 4.7 Series regulator with (a) negative rail as circuit ground; (b) positive rail as circuit ground.
Chapter 04.indd 156 3/12/2009 3:50:04 PM

The negative regulator of Figure 4.7(b) is implemented in Figure 4.8(a). The schematic of the negative
output series regulator shown in Figure 4.8(a) is similar to the positive output regulator of Figure 4.6 except
that the transistors are replaced by PNP transistors to handle reverse current flow direction. The Zener volt-
age reference is also reverse as indicated.
Q1 −
Rb
R1
Rz
−
Vi Q2 RL Vo
+ Io
− R2
Vz
+
+
(a)
Vi1
Common negative rail or
+Vo1 positive regulator
Common positive rail or

negative regulator
−Vo2
Vi2
(b)
Figure 4.8 (a) Series regulator with negative output voltage; (b) dual power supply linear regulator.
Chapter 04.indd 157 3/12/2009 3:50:04 PM

The operation of the negative output voltage regulator of Figure 4.8(a) is also very similar to that of the
positive voltage regulator of Figure 4.6. Here too, the current through Rb can be assumed to be a current
source as the base of Q1 is at potential −Vo − Vbe1 and the other end of Rb is at −Vi potential. Thus,
Vo + Vbe1 − Vi
I Rb = (4.17)
Rb
If the output voltage becomes more negative, then the potential at the base of Q2 decreases as compared to
the emitter of Q2. This increases the drive for the PNP transistor and the contribution of collector current
of Q2 to IRc increases, thereby decreasing the contribution of the base current of Q1 to IRb. As this decreases
the base drive for Q1, the emitter−collector voltage drop of Q1 increases and brings Vo to the equilibrium
state. Similar regulation action occurs when Vo becomes less negative.
The positive rail regulators and the negative rail regulators can be used together to obtain both positive
and negative power supplies with a common ground. A typical dual supply linear regulator circuit is shown
in Figure 4.8(b).
IC Linear Regulators
The linear regulators are available commercially in the form of integrated circuits (IC). The IC regulators can
be classified into three major categories, namely,
1. fixed regulators;
2. variable regulators;
3. variable regulators with current boost.
The fixed regulator types are available as three-pin regulators. The more popular among them are the 78xx
and 79xx series. Here xx represents the output voltage of the regulator. For example, 7805 is a 5 V positive
voltage regulator, 7812 is a 12 V positive voltage regulator and so on. The 79xx series are negative voltage
regulators that are complementary to the 78xx series.
The variable regulators are also available as three-pin regulators with three pins, viz. input, output and
adjust. The “adjust” pin is used for varying and setting the output voltage of the regulator. The 317 regulator
is a popular variable voltage regulator. The 350 is also another three-pin adjustable regulator like the 317 but
with higher load current capability. The 337 is a negative voltage regulator that is complementary to the 317
positive voltage regulator.
There is a class of IC regulators that gives flexibility both in output voltage setting and also in the cur-
rent-carrying capability by giving facility to include an external boost transistor. The 723 is a very popular
linear regulator IC in this class.
Example: 317 IC Regulator

The schematic of the 317 IC linear regulator is shown in Figure 4.9. The characteristic feature of this regula-
tor is the voltage across the output pin (OUT) and the adjust pin (ADJ). The voltage between these two pins
is maintained at 1.25 V for this regulator. The output voltage Vo is decided by the ratio of R1 and R2.
Referring to Figure 4.9, the output voltage is given as
⎛ R ⎞
Vo = 1.25 ⎜ 1 + 2 ⎟ (4.18)
⎝ R1 ⎠
Chapter 04.indd 158 3/12/2009 3:50:05 PM

Io
IN OUT
317
ADJ
R1
1.25 V
Vi RL Vo
R2
Figure 4.9 Basic schematic of 317 IC linear regulator.
R1 is recommended by the manufacturer to be around 240 Ω. R2 can be calculated from the output voltage
requirement and Eq. (4.18). The maximum input voltage that can be applied is 40 V. The 317 IC regulator
needs a minimum differential voltage of 3 V between its input and output pins. Thus the maximum output
voltage can be 37 V. The 317 IC is rated to handle 1.5 A and the 350 IC is rated to handle 3 A. All other
aspects of 350 IC are similar to the 317 IC.
The circuit of Figure 4.9 is sufficient for the basic operation of the regulator. However there are a few
components that need to be included to improve the reliability of the circuit. The complete circuit sche-
matic of the 317 IC regulator circuit is shown in Figure 4.10.
The regulation of the 317 IC is best when the input and the output load are pure DC. However, there
will be ripples on the input side and the load may be switching loads like digital circuit loads. In such cases
there will be AC components present. The regulator performance deteriorates for the AC components.
Therefore, the AC components in the load side that are generated due to digital loads should not be passed
D2
Io
IN OUT
317
ADJ
D1 R1
Vi C1 1.25 V
C3 RL
0.1 to 10 to
1 μF 20 μF
R2 C2
10 μF
Figure 4.10 317 IC linear regulator complete circuit schematic.
Chapter 04.indd 159 3/12/2009 3:50:05 PM

through the regulator to the input side. Likewise the AC ripples present in the input side should not be
passed to the load side through the regulator. The AC components should be bypassed by using capacitors
that will provide low impedance paths for the AC components on either side. C1 and C3 are the input side
and the output side AC bypass capacitors, respectively. A capacitor C2 is connected between the ADJ pin of
the regulator and the circuit ground. This capacitor acts as a buffer to stabilize the fluctuations in the voltage
at the ADJ pin. This provides increased input ripple rejection. However, during turn OFF of the power
supply, the output capacitor C3 will discharge its charge to the load. The capacitor C2 will try to discharge
through the regulator. However, a large discharge current through the ADJ pin will damage the 317 IC. To
provide an alternate discharge path for the capacitor C2, diode D1 has been introduced. Capacitor C2 can
discharge through D1 into the output load. Diode D2 is a protection diode. During situations when the
input voltage suddenly becomes zero during power outage, the input output voltage differential across the
317 IC will become negative and will be equal to −Vo. This will damage the regulator IC. To prevent this,
D2 will ensure that the drop across the input and output of the regulator IC will be clamped to −0.7 V if
ever the output voltage increases beyond the input voltage thereby protecting the regulator IC.
Practical Tips
One of the main benefits of the linear regulators is the high quality of performance. However, there are
integration issues that can have adverse effects on the performance of the regulation. One such crucial issue
is the layout of the input side capacitor. The input side capacitor carries high peak currents as shown and
also discussed in the previous chapter. If the capacitor is laid out as shown in Figure 4.11(b), then the output
voltage is given as
Vo = Vreg + ic Rlead
Regulator
RL Vo
(a)
Regulator
a
Vreg RL Vo
Rlead
(b)
Figure 4.11 Loss of regulation due to incorrect layout of input capacitor

(a) correct; (b) incorrect.
Chapter 04.indd 160 3/12/2009 3:50:06 PM

The term ic Rlead in the output voltage equation will cause de-regulation with load. Therefore care should be
taken to layout the input side capacitor such that it does not interfere with the regulation. The layout of
Figure 4.11(a) is correct and will not deteriorate the output regulation.
Another important layout issue that affects the output regulation is shown in Figure 4.12 with respect to
the 317 IC regulators. In Figure 4.12(a), the lead resistance Rlead comes within the regulation loop of the IC
regulator and in Figure 4.12(b), the layout is such that the lead resistance through which the load current flows
is out of the regulation loop of the IC regulator. For the layout of Figure 4.12(a), the output is given as
⎛ R ⎞ ⎛ R ⎞ ⎛ R ⎞
( )
Vo = 1.25 − I o Rlead ⎜ 1 + 2 ⎟ = 1.25 ⎜ 1 + 2 ⎟ − I o Rlead ⎜ 1 + 2 ⎟
⎝
(4.19)
R1⎠ ⎝ R1⎠ ⎝ R1 ⎠
For the layout of Figure 4.12(b), the output is given as
⎛ R ⎞
Vo = 1.25 ⎜ 1 + 2 ⎟ − I o Rlead (4.20)
⎝ R1 ⎠
From Eqs. (4.19) and (4.20) it is evident that in the case of layout of Figure 4.12(b), the term affecting regu-
lation is only I o Rlead, whereas the term affecting regulation in the case of layout of Figure 4.12(a) is
317
ADJ Rlead
R1
1.25 V
Vi RL Vo
R2
(a)
317
ADJ Rlead
R1
Vi RL Vo
R2
(b)
Figure 4.12 Effect of lead resistance in 317 IC or any variable IC regulator.
Chapter 04.indd 161 3/12/2009 3:50:06 PM

I o Rlead [1 + ( R2 R1 )] which is higher than that for Figure 4.12(b). Therefore, the layout of Figure 4.12(b) is
the preferred layout.
Four-Wire Connection
Another important consideration in the linear regulators is the issue of high current loads. In the case of
high current loads, especially if they are located remotely, the lead resistance Rlead (Figure 4.13) will signifi-
cantly affect the output voltage regulation. Referring to the circuit schematic shown in 4.13, Vreg is the volt-
age that is regulated by the regulator. However, as the actual load is located remotely, the actual output
voltage is across the load resistor RL and is given as
Vo = Vreg − I o Rlead (4.21)
From Eq. (4.21), it is evident that the actual output voltage is different from the regulated voltage Vreg by a
load-dependent term I o Rlead . This will make the output voltage regulation for changes in the load poorer.
A solution for this is the four-wire connection as shown in Figure 4.14. The regulator provides four
output terminals 1−4 as indicated in Figure 4.14, viz., (a) positive rail or + power terminal; (b) positive
io
Rlead
−
Vi Vreg RL Vo
+
Vref
Figure 4.13 Effect of lead resistance in high current regulators.
1+ io
2 Rlead
100 Ω
S+
− Vreg RL Vo
Vi
+
Vref
3
100 Ω
S− Rlead
4−
Figure 4.14 Four-wire connection for high current linear regulators.
Chapter 04.indd 162 3/12/2009 3:50:06 PM

of the output voltage sense or S+ terminal; (c) negative of the output voltage sense or S− terminal and
(d) negative rail or − power terminal.
The load current flows through the positive rail terminal, through the load and finally returns through the
negative rail terminal. The actual output voltage Vo is sensed by the sensing leads that are connected to termi-
nals 2 and 3. The currents through the sensing leads are very small as compared to the load current Io. There-
fore, the sense terminals will essentially measure the actual remote output voltage Vo. In this way, even the
drops across the lead resistances in the positive rail and negative rail are included inside the regulation loop,
thereby regulating the actual output Vo. Sometimes terminals 1 and 2 and terminals 3 and 4 are connected
with resistances of about 100 Ω as shown in Figure 4.14. This will enable either two-wire connections or four-
wire connections to be used depending on the remoteness of the load, wiring layout and lead resistances.
Protection
Protection circuitry is an integral part of any system for improving the availability of the system. The linear
regulator should, in general, be protected against the following:
1. over current;
2. over voltage;
3. reverse voltages;
4. reverse voltage drops across the series component.
For the shunt linear regulators, the series component is a fixed resistor. The power rating of the resistor is
normally chosen to be
(Vi- max − Vz )2
PRs =
Rs
However, if the output gets short circuited, then the entire output voltage will drop across the series resistor.
The dissipation will exceed the rated power dissipation and damage the series component. To take care of
this extreme situation, the series resistor Rs in the case of the shunt regulator should have a power rating of
Vi-2max
PRs =
Rs
For series regulators, Figure 4.15 shows a very simple and effective current protection scheme. Two compo-
nents, transistor Q s and current resistor Rs are used for current protection. The current sense resistor is
chosen such that when the load current limit is reached, the voltage drop across it is 0.5 V. Any further
increase in the load current will increase the drop across Rs which in turn increases the base−emitter voltage
of Q s. This increased base−emitter voltage will make Q s to draw increased collector current. As the current
through Rb is a current source, the increase in the collector current drawn by Q s is at the expense of the base
current of Q1. Thus, as the drive to Q1 decreases, Q1 presents a higher resistance to the input source and the
drop across Q1 increases thereby decreasing the output voltage and load current. In this manner, the combi-
nation of Rs−Q s will provide over current limiting in linear series regulators.
The over voltage protection is mainly to ensure that the load is not damaged. The over voltage may occur
if the series pass of the linear regulator is shorted or due to an electrostatic coupling at the output. The over
voltage protection is implemented by means of the crow-bar protection circuit as shown in Figure 4.16.
Under normal operating conditions, (a) the thyristor, Th1, is OFF, (b) the Zener diode Dcb is reverse-biased
and in OFF region and (c) the potential of the thyristor gate is zero. If for some reason the output voltage
goes beyond a particular limit, the Zener diode Dcb breaks down and a potential of [(Vo − Vz-cb ) / ( R1b + R2b )]R2b
Chapter 04.indd 163 3/12/2009 3:50:07 PM

Rs Io
Q1
Rb
R1
Q3
Rz
Vi RL Vo
Q2
R2
Figure 4.15 Current protection.
D1
Fuse
Q1
Dcb Rcb
Vi Dr Vo
R1b Th1
R2b
Figure 4.16 Voltage protections.
is applied at the gate of the thyristor Th1. The thyristor turns ON and connects a small value resistance Rcb
across the output Vo, effectively shorting the output with Rcb. This will bring down the output voltage
within safe limits. The thyristor switches OFF only after its current is brought below the holding current
value. For crowbar protection, a fuse is included in series with the input source. When the output is short
circuited due to the action of the Th1, the current from the input source increases and blows the fuse. This
will bring the current in the thyristor below the holding current value and switches OFF the thyristor Th1.
Alternately, this is done by switching OFF the input source Vi by turning OFF a series switch that can be
included in series with the input source or the current protection scheme of Figure 4.15 can be used. If Th1
turns ON due to over voltage condition, the current through the sense resistor Rs of Figure 4.15 will increase
and clamp at 0.7 V. Transistor Q s will be in ON-state and divert all the current from IRb to its collector. This
Chapter 04.indd 164 3/12/2009 3:50:07 PM

will cut-off Q1 which will bring down the thyristor current below the holding value and turn OFF the thy-
ristor. However, it should be noted that if the output over voltage has occurred due to Q1 collector−emitter
short, then the current limit circuit will not work. In such cases either the fuse in series with the input will
blow or the collector−emitter of Q1 that was shorted will open.
Diode Dr is connected as shown in the figure to prevent any negative voltage being applied to the load.
Similar reverse diode may be connected at the input side to prevent any accidental reverse voltage being
applied to the positive rail regulator, thereby preventing any damage to the linear regulator and the load.
Diode D1 is used to prevent any negative voltage drop occurring across the series-pass element. This can
happen when the input is turning OFF and the output capacitor is still holding charge. Under these condi-
tions, diode D1 will conduct and clamp the voltage across the series-pass element to −0.7 V.
Current Regulation
Till now, the linear regulators for output voltage regulation are discussed. However, similar principles can be
used for regulating the output current to obtain a current source. The main principle in a current regulator
is shown in Figure 4.17(a). Here the voltage across a resistor VR is held constant by means of a voltage regu-
lator that may be a shunt or a series regulator. This makes the current VR/R through that resistor constant
which is directed to the output as the load current.
Figure 4.17(b) shows the circuit implementation of the current regulator. A modified shunt regulator
with a PNP transistor is used in the circuit schematic of the current regulator. The voltage across the emitter
resistor Re is given as Vz − Vbe. The emitter current is therefore given as
Vz − Vbe
I Re = (4.22)
Re
From Eq. (4.22), it can be observed that Vz and Vbe are constants and hence the current through Re is a con-
stant. For a high b transistor, the current through the collector of the transistor Q is almost the same as IRe.
This current flows through the connected load resistor RL as a constant current developing an output voltage
of
⎛ V − Vbe ⎞
Vo = I L RL = ⎜ z ⎟ RL (4.23)
⎝ Re ⎠
+
Re
Vz
−
Reg
VR / R Q
VR R Vi
IL
Vi IL
Rz
RL RL Vo
(a) (b)
Figure 4.17 (a) Current regulation principle; (b) current regulator circuit schematic.
Chapter 04.indd 165 3/12/2009 3:50:07 PM

The maximum output voltage that can develop is given by
Vo-max = Vi-max − Vz + Vbe (4.24)
4.3 Parameters of the Linear Regulator
T he following four parameters of the linear regulator determine the quality of the regulator:
1. efficiency;
2. line regulation;
3. load regulation;
4. temperature regulation.
The first parameter is related to the power delivered to the output with respect to the power consumed from
the source. This primarily affects the volumetric power density of the linear regulator. The power dissipated
in the series-pass element determines the size of the heat sink that needs to be employed to ensure proper
heat flow to the ambient. The remaining three parameters provide quantitative measures for the perfor-
mance quality of the output voltage. This performance quality of the linear regulator is measured by regula-
tion of the output voltage for (a) variations in the line, (b) variations in load and (c) variations in temperature.
This is mathematically represented as
∂Vo ∂V ∂V
ΔVo = ΔVi + o ΔI o + o ΔT (4.25)
∂Vi ∂I o ∂T
where ∂Vo / ∂Vi is the line-regulation coefficient with Io and T constant; ∂Vo / ∂I o the load-regulation coef-
ficient with Vi and T constant; ∂Vo / ∂T the temperature-regulation coefficient with Vi and Io constant. For
any linear regulator selection, these four important parameters are essential and need to be quantified and
specified. The analysis of the regulators will also focus on these parameters.
4.4 Analysis of a Series Regulator
T ill now, the regulator circuits were discussed in an operational sense wherein one could design and
implement the regulator circuits. However, to obtain greater insight into the operation of the regulator
so that one may improve the circuits, a more systematic method of analysis has to be performed. In this sec-
tion, a systematic analysis will be performed on a series linear regulator circuit (shown in Figure 4.18). The
analysis will bring out the problems in this topology and will also suggest remedial measures that can be
taken to address the issues. Without loss of generality, a similar approach may be used to analyze other linear
regulator circuits.
First-Level Modeling
To analyze the regulator shown in Figure 4.18, the transistor and the Zener diode are replaced by the first-
level idealized model. The first-level model for the transistor and the Zener diode are shown in Figure 4.19.
For the NPN transistor shown in Figure 4.19(a), the first-level model is shown in Figure 4.19(b). The collec-
tor current is a dependent current source of value bib. The first-level model of a Zener diode is a DC source
of value Vz as shown in Figure 4.19(d).
Chapter 04.indd 166 3/12/2009 3:50:08 PM

Io
Q1
Rb
R1
Rz
Vi RL Vo
Q2
R2
Vz
Figure 4.18 Linear series regulator.
ic ie
bib = ic
ie
ib ib
(a) (b)
Vz Vz
(c) (d)
Figure 4.19 (a) NPN transistor; (b) NPN transistor first-level model;
(c) Zener diode; (d) first-level model of Zener diode.
Referring to Figure 4.18, both the transistors Q1 and Q2 are replaced by their first-level model. Similarly,
the Zener diode is also replaced by its first-level idealized model. The resulting equivalent circuit is shown in
Figure 4.20.
By applying Thevenin’s theorem to the base portion of the transistor Q2 of Figure 4.20, the modified
equivalent circuit representation of the series regulator is obtained as shown in Figure 4.21.
Referring to Figure 4.21, the base current of Q2 is given as
Vo [ R2 / ( R1 + R2 )] − Vz Vo ⎛ R + R2 ⎞
ib2 = = − Vz ⎜ 1 ⎟ (4.26)
R1 //R2 R1 ⎝ R1R2 ⎠
Chapter 04.indd 167 3/12/2009 3:50:08 PM

b1ib1
Io
Rb ib1
irb
Rz R1
b2ib2
Vi RL Vo
ib2
Vz R2
Figure 4.20 Equivalent circuit of the series regulator using the first-level
models of the transistor and the Zener diode.
b1ib1
Rb ib1
irb
Rz
b2ib2
Vi RL Vo
R1//R2
ib2
R2
Vz Vo
R1 + R2
Figure 4.21 Equivalent circuit of the series regulator.
The current through Rb is given as

V i − Vo
irb =
Rb (4.27)
Chapter 04.indd 168 3/12/2009 3:50:08 PM

ib1 = irb − β2ib2

Vi − Vo ⎧⎪ Vo ⎛ R + R2 ⎞ ⎫⎪
= − ⎨β2 − β2Vz ⎜ 1 ⎟⎬ (4.28)
Rb ⎪⎩ R1 ⎝ R1R2 ⎠ ⎭⎪
The regulating action of the series regulator is defined by Eq. (4.26). If the output voltage Vo increases,
then ib2 increases. An increase in ib2 will cause a decrease in ib1 as seen from Eq. (4.28). This in turn will
reduce the base drive for Q1 which will present a higher resistance to the input source. As a consequence,
the collector−emitter of Q1 will drop more voltage and the output voltage will reduce and return back to
the equilibrium state. A similar regulating action occurs when the output voltage Vo decreases.
The output voltage is given as
Vo = (ib1 + β1ib1 )RL = (β1 + 1)ib1RL

⎡ V − Vo V ⎛ R + R2 ⎞ ⎤
= (β1 + 1)RL ⎢ i − β2 o + β2Vz ⎜ 1 ⎟⎥ (4.29)
⎢⎣ Rb R1 ⎝ R1R2 ⎠ ⎥⎦
The output relationship as given in Eq. (4.29) is re-arranged such that it is of the form
Vo = k1Vi + k2Vz (4.30)
where
( β1 + 1)RL / Rb
k1 =
[( β1 + 1)RL / Rb ] + [( β1 + 1)β2 ( RL / R1 )]+ 1
and
( β1 + 1)β2 RL [( R1 + R2 ) / R1R2 ]
k2 =
[( β1 + 1)RL / Rb ] + [( β1 + 1)β2 ( RL / R1 )] + 1
It should be noted that in the above derivation for Vo, the load current Io is taken as ( β1 + 1)ib1. However, Io
is actually ( β1 + 1)ib1 − iR1 where iR1 is the current through R1. Referring to Figure 4.20, this is given as
(Vo − Vz ) / R1 . As R1 is large, iR1 is neglected in the above first-level derivation for Vo. However, in later dis-
cussions, the contribution of iR1 to the load current is also considered.
Referring to Eq. (4.30), there are a few conclusions that one can draw at this point of the analysis. If Vo
is to be independent of the input source Vi, then k1 should be zero and k2 should be a constant. The factor
k1 can be written as 1
k1 =
1 + β2 ( Rb / R1 ) + [ Rb / ( β1 + 1)RL ]
In this equation, only the term β2 ( Rb / R1 ) is greater than unity. If the transistor Q2 is chosen as a high gain
transistor, then b2 will be high. In fact, b2 will be much higher than the ratio of Rb/R1. As a consequence,
this term will be much greater than 1 and thus
k1 1 (4.31)
The factor k2 in Eq. (4.30) can be written as
β2 [( R1 + R2 ) / R1R2 ]
k2 =
(1 / Rb ) + β2 (1 / R1 ) + [1 / ( β1 + 1)RL ]
Chapter 04.indd 169 3/12/2009 3:50:09 PM

Due to the b2 terms both in the numerator and the denominator, the value of k2 is significant and
k2 k1 (4.32)
From the inequalities in Eqs. (4.31) and (4.32), one can say that the term k1Vi is negligibly small. Thus
Vo ≈ k2Vz
which is a regulated voltage.
Effect of Current through Rb

From Figure 4.21, it can be observed that the current through the base of the series-pass transistor Q1 is
ib1 = irb − β2ib2
where
Vo ⎛ R + R2 ⎞
ib2 = − Vz ⎜ 1 ⎟
R1 ⎝ R1R2 ⎠
If R1, R2, Vz and Vo are constants, then ib2 will be a constant. For a given temperature, b1 can also be consid-
ered to be a constant. Therefore, the term β2ib2 in the ib1 equation is a constant and does not contribute to
loss of regulation. The other term in the ib1 equation is the current through Rb. This is given as
V i − Vo
irb = (4.33)
Rb
In Eq. (4.33), one can observe that irb is a function of both Vi and Vo. If Vi varies, irb also varies and hence
ib1 varies which in turn causes Vo to vary as
Vo = ( β1 + 1)ib1RL (4.34)
Therefore from Eqs. (4.33) and (4.34) it is evident that the loss of output regulation is due to irb not being
a constant in the face of variations in Vi.
The solution to this is as follows: It is required that irb is constant and independent of variations in
the input source voltage Vi. If this is achieved then ib1 and hence Vo will become independent of variations in the
input source voltage Vi. One possible method to make irb independent of variations in Vi is to replace the resis-
tor Rb by a constant-current source as illustrated in Figure 4.17(b). The constant-current source circuit is inte-
grated into the series regulator circuit and the resulting modified circuit is depicted in Figure 4.22.
The operation of the constant-current source portion is as discussed in the section “Current Regulation”.
The current irb is now a constant that is independent of variations in the input voltage Vi. Now the output
voltage Vo is given as
Vo = (ib1 + β1ib1 )RL = ( β1 + 1)ib1RL
⎡ V ⎛ R + R2 ⎞ ⎤
= ( β1 + 1)RL ⎢irb − β2 o + β2Vz ⎜ 1 ⎟⎥ (4.35)
⎢⎣ R1 ⎝ R1R2 ⎠ ⎥⎦
From Eq. (4.35), it can be observed that as irb is a constant, Vo is essentially independent of Vi and depends
only on Vz.
Chapter 04.indd 170 3/12/2009 3:50:09 PM

Constant-current source
Q1
Re
Vz1
R1
Q3
irb
Vi RL Vo
Rz1
Rz Q2
R2
Vz
Figure 4.22 Series regulator circuit: Rb is replaced by a constant-current source.
Effect of Current through R1

In the discussions till now, the output load current Io is considered to be ( β1 + 1)ib1 wherein the current
through R1 is neglected. If the current through R1 is also considered in the output voltage equation with
respect to the series regulator equivalent circuit of Figure 4.21, then Vo is given as
Vo = [(β1 + 1)ib1 − iR1 ]RL

⎡ ⎛ V − Vz ⎞ ⎤
( )
= ⎢ β1 + 1 ib1 − ⎜ o
⎝ R1 ⎠ ⎥⎦
⎟ ⎥ RL (4.36)
⎢⎣
Substituting for ib1 from Eq. (4.28) into Eq. (4.36), one obtains
⎡ V − Vo V ⎛ R + R2 ⎞ ⎤ ⎛ Vo − Vz ⎞
Vo = (β1 + 1)RL ⎢ i − β2 o + β2Vz ⎜ 1 ⎟⎥−⎜ ⎟ RL
⎢⎣ Rb R1 ⎝ R1R2 ⎠ ⎥⎦ ⎝ R1 ⎠
The above equation for Vo can be re-arranged in the following form:
Vo = k1Vi + k2Vz (4.37)

where
(β1 + 1)RL / Rb
k1 =
(β1 + 1)( RL / Rb ) + (β1 + 1)β 2 ( RL / R1 ) + ( RL / R1 ) + 1
and
(β1 + 1)β2 RL [( R1 + R2 ) / R1R2 ] + ( RL / R1 )
k2 =
(β1 + 1)( RL / Rb ) + (β1 + 1)β2 ( RL / R1 ) + ( RL / R1 ) + 1
Chapter 04.indd 171 3/12/2009 3:50:10 PM

The load current Io is given as
Vo ⎡ V − Vo V ⎛ R + R2 ⎞ ⎤ ⎛ Vo − Vz ⎞
Io = = ( β1 + 1) ⎢ i − β2 o + β2Vz ⎜ 1 ⎟⎥−⎜ ⎟
RL ⎢⎣ Rb R1 ⎝ R1R2 ⎠ ⎥⎦ ⎝ R1 ⎠
which can be re-arranged as

(β1 + 1) ⎛ (β + 1) (β1 + 1)β2 1 ⎞ ⎡ ⎛ R + R2 ⎞ 1 ⎤
Io = Vi − ⎜ 1 + − ⎟ Vo + ⎢( β1 + 1)β2 ⎜ 1 ⎟ + ⎥V z (4.38)
Rb ⎝ Rb R1 R1 ⎠ ⎢⎣ ⎝ R1R2 ⎠ R1 ⎥⎦
Regulator Parameters
The regulation parameters, viz., line-regulation coefficient, load-regulation coefficient and temperature-
regulation coefficient can be estimated based on the equations for Vo and Io as given in Eqs. (4.37) and (4.38).
Line-Regulation Coefficient
The line-regulation coefficient can be obtained from Eq. (4.37). It is the variation of the output voltage Vo
for variations of the input voltage Vi under the constraints of constant load and temperature. It is given as
∂Vo ( β1 + 1)RL / Rb
= (4.39)
∂Vi (β1 + 1)( RL / Rb ) + (β1 + 1)β2 ( RL / R1 ) + ( RL / R1 ) + 1
In the case of the modified series regulator circuit wherein the resistor Rb is replaced by a constant-current
source, irb is a constant. From Eq. (4.35), it can be observed that Vo is independent of Vi and therefore,
∂Vo
=0
∂Vi
Load-Regulation Coefficient
The load-regulation coefficient is a measure of the change in the output voltage Vo for changes in the load Io
under the constraints of constant temperature and input voltage Vi. The load-regulation coefficient can be
obtained from Eq. (4.38). Thus,
∂I o ⎛ β + 1 ( β1 + 1)β 2 1 ⎞
= − ⎜⎜ 1 + − ⎟⎟
∂Vo ⎝ Rb R1 R1 ⎠
The inverse of the above equation will give the load-regulation coefficient. This is given as
∂Vo Rb R1
=− (4.40)
∂I o (β1 + 1)R1 + (β1 + 1)β 2 Rb − Rb
Here ∂Vo / ∂I o is a measure of the output resistance Ro of the regulator circuit. The negative sign in Eq. (4.40)
does not mean that Ro is negative, but it implies that the nature of the change in Vo with respective to changes
in Io is opposite in phase, that is, if Io increases by DIo, Vo decreases by DVo and this is equal to ΔI o Ro.
From Eq. (4.40), it can be observed that the denominator contains a term that is a product of the bs, that
is, the current gains of the transistors Q1 and Q2. This implies that Ro value is a very small value such that
Ro 1. If Q1 and Q2 are chosen such that their current gains are high, then Ro can be negligibly small.
Chapter 04.indd 172 3/12/2009 3:50:10 PM

In the case of the modified series regulator circuit wherein Rb is replaced by a constant-current source
circuit, the load-regulation coefficient is given as
∂Vo R1
=− (4.41)
∂I o ( β1 + 1)β2 − 1
Temperature Effects
In the first-level modeling, as the transistors and the Zeners are considered as ideal. There is no contribution
from these elements to the output voltage variations due to temperature changes. However, there will be
changes in the output voltage for temperature changes due to variations in the value of the resistors with
temperature. The effect of temperature variations on the output voltage is discussed later while considering
non-ideal transistors and Zeners in the second-level model.
Problem 4.2
Consider a linear series regulation with R1 = 3.6 kΩ, R2 = 2 kΩ, Rb = 3.6 kΩ, Rz = 20 kΩ, Vz = 6.55 V,
RL = 1 kΩ, Vi = 20 V to 30 V, Vi-nominal = 25 V, b1 = b2 = 100. Calculate the output voltage a load change of
0.1 A to 0.2 A.
Solution
Applying the values to the variables in Eq. (4.37), one obtains
Vo = 9.89 × 10−3Vi + 2.77Vz = 9.89 × 10−3Vi + 18.14
1. At Vi = 25 V, Vo = 18.39 V
2. At Vi = 20 V, Vo = 18.3378 V
3. At Vi = 30 V, Vo = 18.4367 V
For a change of 10 V in Vi, the change in output voltage is 0.0989 V.
∂Vo 3600 × 3600

Ro = =− = 0.3529 Ω
∂I o 101× 3600 + 101× 100 × 3600 − 36600
For a load change of 0.1 A to 0.2 A,

DVo = 0.3529 × 0.1 = 0.03529 V
which is about 0.2% of Vo.
Range of Load Resistance RL

Referring to the output voltage in Eq. (4.37), both k1 and k2 factors are functions of RL. Therefore the
output voltage Vo is not independent of RL. Dividing the numerator and denominators of k1 and k2 by RL,
a term 1/RL is present in the denominator of both k1 and k2. Referring to the denominators of k1 and k2 of
the output voltage Eq. (4.37), if
Chapter 04.indd 173 3/12/2009 3:50:10 PM

⎧⎪ 1 1 1 ⎫⎪ 1
⎨( β1 + 1) + ( β1 + 1)β2 + ⎬
⎩⎪ Rb R1 R1⎪
⎭ RL
then the output voltage is not affected significantly by RL. As RL approaches infinity, that is, the series
regulator approaches open circuit condition, 1/RL anyway approaches zero and the above inequality
constraint is easily met thereby making Vo independent of RL. However, as RL decreases, that is, the
series regulator approaches the full-load or overload conditions, the dependency of Vo on RL is more
significant.
For Problem 4.2, wherein R 1 = 3.6 kΩ, R 2 = 2 kΩ, R b = 3.6 kΩ, R z = 20 kΩ, V z = 6.55 V,
RL = 1 kΩ, Vi = 20 V to 30 V, Vi-nominal = 25 V, b1 = b2 = 100, one can apply the inequality constraint
given above. Thus,
1 1 1
( β1 + 1) + ( β1 + 1)β2 + = 2.834
Rb R1 R1
1 1
= = 1 × 10−3
RL 1000
It can be observed that 1/RL is much less than 2.834 and hence the output voltage can be considered to
be reasonably independent of RL. However if RL approaches the short circuit condition, Vo will lose
regulation.
Second-Level Modeling
Till now the insights about the series regulator were obtained based on the idealized first-level models for the
transistors and the Zener diode. However, there are many non-idealities in the semiconductor devices. If the
dominant non-idealities are included in the device models, then the insights obtained and the conclusions
about the circuit will be closer to reality. Figure 4.23 depicts the second-level models for the transistors and
the Zener diode. For the NPN transistor shown in Figure 4.23(a), the second-level equivalent circuit is
shown in Figure 4.23(b). The base-spreading resistance and the base-to-emitter junction potential are
included in the equivalent circuit. For the Zener diode of Figure 4.23(c), the second-level equivalent circuit
is shown in Figure 4.23(d). The Zener resistance, rz in the breakdown region is also included in the equiva-
lent circuit.
In the series regulator circuit of Figure 4.18, the second-level equivalent circuits are used to replace the
transistors and the Zener diode. The equivalent circuit of the series regulator with the second-level models
introduced is shown in Figure 4.24. Like in the case of the first-level modeling, Thevenin’s theorem is
applied to the base circuit of transistor Q2. The reduced circuit after applying Thevenin’s theorem is shown
in Figure 4.25.
Referring to Figure 4.25, the base current of Q2 is given as
Vo [ R2 / ( R1 + R2 )] − Vbe2 − Vz − rz [(Vi − Vz ) / ( Rz + rz )]
ib2 = (4.42)
rb2 + R1 // R2 + ( β 2 + 1)rz
Chapter 04.indd 174 3/12/2009 3:50:10 PM

bib
ic ie
Vbe
rb
ib ib
(a) (b)
rz
Vz
Vz
(c) (d)
Figure 4.23 (a) NPN transistor; (b) NPN transistor second-level model;
(c) Zener diode; (d) second-level model of Zener diode.
Vbe1
Io
b1ib1 rb1
Rb
irb i b1
R1
Rz
b 2i b2
Vi RL Vo
r b2
Vbe2
R2
rz
vz
Figure 4.24 Equivalent circuit of series regulator with second-level models.
The current through Rb is given as

Vi − (Vo + Vbe1 + ib1rb1 )
irb = (4.43)
Rb
ib1 = irb − β 2ib2 (4.44)
Chapter 04.indd 175 3/12/2009 3:50:10 PM

Vbe1
Io
b1ib1 rb1
Rb
irb i b1
Rz
b 2i b2
Vi RL Vo
r b2 R1 //R2
Vbe2
Vo R2
(R1 + R2)
rz
vz
Figure 4.25 Equivalent circuit of series regulator after circuit reduction.
The output voltage Vo is given as

Vo = (ib1 + β1ib1 )RL = ( β1 + 1)ib1RL (4.45)
The output relationship can be expressed in the form
Vo = k1Vi + k2Vz + k3Vbe1 + k4Vbe2 (4.46)
where
1 ⎧⎪ ⎛ RL ⎞ ( β1 + 1)β 2 RL [rz / ( Rz + rz )]Rb ⎫⎪
k1 = ⎨( β1 + 1) ⎜ ⎟+ ⎬
D ⎩⎪ ⎝ Rb + rb1 ⎠ [rb 2 + R1 // R2 + (β 2 + 1)rz }( Rb + rb1 ) ⎭⎪
1 ⎧ ( β1 + 1)β 2 RL [ Rz / ( Rz + rz )]Rb ⎫
k2 = ⎨ ⎬
D ⎩[rb 2 + R1 // R2 + ( β 2 + 1)rz ]( Rb + rb1 ) ⎭
1 ⎪⎧ ( β1 + 1)RL ⎫⎪
k3 = ⎨− ⎬
D ⎪⎩ Rb + rb1 ⎪⎭
1⎧ (β1 + 1)β 2 RL Rb ⎫
k4 = ⎨ ⎬
D ⎩[rb2 + R1 // R2 + (β 2 + 1)rz ]( Rb + rb1 ) ⎭
and
⎧ ( β + 1)RL ( β1 + 1)β 2 RL [ R2 / ( R1 + R2 )]Rb ⎫
D = ⎨1 + 1 + ⎬
⎩ Rb + rb1 [rb2 + R1 // R2 + ( β 2 + 1)rz ]( Rb + rb1 ) ⎭
Chapter 04.indd 176 3/12/2009 3:50:11 PM

From Eq. (4.46), it is evident that Vo is dependent not only on Vi and Vz, but also on the base−emitter junc-
tion voltages of the transistors. It should be noted that when the non-idealities, that is, Vbe1, Vbe2, rz, rb1 and
rb2 are made equal to zero, the output voltage equation reverts back to the equation developed in the first-
level model of the previous sub-section.
Influence of Vi on Output Voltage Regulation

From Figure 4.25, it can be observed that there are two parameters that are dependent on Vi and have a
strong influence on the output voltage. They are the following:
1. The current irb which is dependent on Vi, as can be seen from Eq. (4.43).
2. The current through Rz that biases the Zener diode is also dependent on Vi. This current flows through
rz which causes the drop across rz to change with Vi.
However, in the first-level model, rz is zero and, therefore, only irb variations contributed to the output volt-
age de-regulation.
The solution to this is as follows: The first dependency with respect to irb can be solved by using a con-
stant-current source circuit as discussed in the first-level model. This is shown in Figure (4.26). It should be
noted that even though irb is a constant and does not vary with variations in Vi, both the Zener voltage Vz1
and Vbe3 are temperature-dependent. Therefore irb has a dependency with respect to temperature variations.
However, the Zener can be selected to have a positive temperature coefficient and Q3 base−emitter junction
has negative temperature coefficient. If these two devices are appropriately chosen, the temperature depen-
dency of irb can be significantly reduced.
The second dependency is with respect to the current through rz. The current through the Zener break-
down resistance rz is due to contributions from
1. ( β2 + 1)ib2
Vi − Vz
2.
Rz + rz
The second contribution, wherein the current flows from the input through Rz to bias the Zener diode, is
dependent on input voltage variations. To solve this dependency, the resistor Rz is re-located in such a
manner that the Zener bias current is drawn from the output side rather than the input side. As the output
is regulated, the Zener bias current becomes independent of the input voltage variations. Figure 4.26 illus-
trates the re-location of the Zener bias resistor Rz. After this relocation, the current through Rz is given as
(Vo − Vz ) / ( Rz + rz ) .
Effect of rz on Output Regulation

The current through rz has two components as discussed, viz., ( β2 + 1)ib2 and the bias current through Rz.
The variations in the drop across rz due to variations in Vi are solved by re-location of Rz. However, the
(β2 + 1)ib2 component of the current through rz also varies to de-regulate the output Vo. If this component
of current is made constant then the output voltage regulation action is lost. There are only two solutions,
viz., (a) make rz zero which is not practically possible or (b) do not allow the load-dependent component of
the current ( β2 + 1)ib2 to flow through rz. The latter solution is possible by use of a high impedance ampli-
fier like the op-amp. This solution is depicted in Figure 4.27. The op-amp has a very high input impedance
device. The current through the Zener diode is only the bias current component flowing through the resistor
Rz and no load-dependent component flows through it. Thus the reference is a constant and independent of
load variations.
Chapter 04.indd 177 3/12/2009 3:50:11 PM

Q1
Re
Vz1
Rz R1
Q3
Vo RL Vo
Q2
Rz1
R2
Vz
Figure 4.26 Eliminating dependency on input voltage by using

constant-current source and relocating Rz.
Q1
Re Rz
Vz1 R1
−
Q3
Vi + RL Vo
Rz1 R2
Vz
Figure 4.27 Series regulator with op-amp.
There is another advantage in using an op-amp in the series regulator instead of the transistor Q2. In the
first-level model, the range of RL depends on the inequality
⎧⎪ 1 1 1 ⎫⎪ 1
⎨( β1 + 1) + ( β1 + 1)β2 + ⎬
⎩⎪ Rb R1 R1⎪
⎭ RL
The gain of the op-amp is very high, of the order of 105 and greater. This makes β2 very large and the above
inequality is satisfied for even very low values of RL. As a consequence the output voltage regulation can be
achieved even at very high load, that is, very low RL. However, in this case the series pass should be rated for
high currents.
Chapter 04.indd 178 3/12/2009 3:50:12 PM

Influence of Temperature
There are four major parameters that are dependent on temperature. They are:
1. transistor current gains;
2. base−emitter junction potentials of transistors;
3. breakdown voltage of the Zener diode;
4. resistance variations.
From the output voltage in Eq. (4.46), it is evident that all these parameters will significantly affect the
output voltage regulation. The transistor current gain b increases with increase in temperature. One way to
minimize the effect of variations in b due to change in temperature is to make b as large as possible. The
large b will make the output Vo immune to changes in parameters. Another way is to use an op-amp as
shown in Figure 4.27. Due to its high gain, the dependency of Vo to changes in parameters due to tempe-
rature variations, is significantly reduced.
With regard to the base−emitter junction potential of transistors, the temperature coefficient is negative.
The Vbe of the transistor varies at the rate of −2.2 mV/°C rise in temperature. Zeners below 5.1 V generally
have negative temperature coefficient and Zeners above 5.1 V have positive temperature coefficients. Some
degree of temperature compensation can be achieved by choosing Zeners with positive temperature coeffi-
cient such that they negate the variations of Vbe with temperature changes.
Another solution to compensate for variations in Vbe is to use a diode D as shown in Figure 4.28(a). The
equivalent circuit for the base drive portion of Q2 by applying Thevenin’s theorem is shown in Figure 4.28(b).
The Thevenin voltage for the equivalent circuit of Figure 4.28(b) is given as
⎛ R2 ⎞ ⎛ R1 ⎞
Vth = Vo ⎜ ⎟ + VD ⎜ R + R ⎟
⎝ 1
R + R2⎠ ⎝ 1 2⎠
Q1
R1
Q2 R1 //R2
rb2
Vbe2
R2
Vz Vth
Vz
D
(a) (b)
Figure 4.28 Using diode to compensate for temperature effects.
Chapter 04.indd 179 3/12/2009 3:50:12 PM

Vo [ R2 / ( R1 + R2 )] + VD [ R1 / ( R1 + R2 )] − Vz − Vbe2
ib2 =
rb2 + R1 // R2
From the above equation for ib2, due to opposite signs for the diode voltage and the base–emitter voltage, a
certain degree of temperature compensation is achieved between the diode D and Vbe2 of the transistor Q2.
Alternately, yet another solution is to enclose Q1, Q2 and Vz in a constant temperature oven so that there
is no drift due to temperature. Though this solution is good for regulation performance, it is at the expense
of the efficiency of the series regulator. With regard to resistance variations with temperature, the only solu-
tion is to choose very tight tolerance resistors wherein the resistance value variations with temperature are
very small.
4.5 Analysis of a Current Regulator
T he modeling and analysis methodology followed in the previous section for the series regulator can be
applied for any other regulator too. In this section, the analysis is performed for a typical current regu-
lator. In the section “Current Regulation” the circuit and the operational principle of the current regulator
has been discussed. A current regulator based on PNP transistor is discussed. This current regulator is also
used in the series regulator to improve the output regulation performance as discussed in the previous sec-
tion. In this section, the analysis is performed on a current regulator with NPN transistor whose circuit
schematic is as shown in Figure 4.29.
The operation of this current regulator is exactly similar to that described in the section “Current
Regulation”. The voltage across Re is maintained constant and independent of the input voltage Vi. The
constant voltage across Re ensures a constant current through Re. This current flows as the output cur-
rent Io through RL. In a manner similar to that discussed for the series regulator, the first-level equivalent
circuit of the current regulator can be obtained by applying the first-level models for the transistor and
the Zener diode as illustrated in Figure 4.19. The first-level model for the current regulator is shown in
Figure 4.30.
From the model of Figure 4.30, the voltage across Re is Vz. The current through Re is given as
Vz
I Re = = ( β + 1)ib (4.47)
Re
Q Io
Re
Vi RL
Rz
Vz
Figure 4.29 Current regulator circuit.
Chapter 04.indd 180 3/12/2009 3:50:12 PM

Re Io
bib
ib
Vi RL
Rz
VZ
Figure 4.30 Equivalent circuit of the current regulator with idealized models.
From Eq. (4.47), it is evident that a constant-current source is established as Vz is a constant. Referring
again to Figure 4.30, it can be observed that
Vo = I o RL (4.48)
V i − V z − Vo
iRz = (4.49)
Rz
Io
ib = (4.50)
β +1
iz = iRz − ib (4.51)
The output current Io is given as
Vz
Io = +i (4.52)
Re z
Substituting Eqs. (4.48)−(4.51) into Eq. (4.52), one obtains
(Vz / Re ) − [ I Re / ( β + 1)] + [(Vi − Vz ) / Rz ]

Io = (4.53)
1 / ( RL / Rz )
The output current Io can be re-arranged as
I o = k1Vi + k2Vz (4.54)
where
1 / Rz
k1 =
1 + ( RL / Rz )
(1 / Re ) − [1 / Re ( β + 1)] − (1 / Rz )
k2 =
1 + ( RL / Rz )
Chapter 04.indd 181 3/12/2009 3:50:13 PM

It can be observed from the output current relationship of Eq. (4.54) that the output current Io is depen-
dent on Vi and the load resistance RL. It should be noted that Rz should be much larger in value than the
resistor Re to get effective current regulation. This can be achieved by using higher current gain transistors
such that the base side currents become very small.
Line Regulation
The term line regulation for current regulators is associated with the variations of the output current Io for
variations in the input voltage Vi. The line-regulation coefficient is expressed as
∂I o 1 / Rz 1
= = (4.55)
∂Vi 1 + ( RL / Rz ) RL + Rz
The variation of output current for variations in the input voltage can be significantly reduced by increasing
the value of Rz. This implies that high b transistors should be selected such that the base side currents are
small, thereby allowing choice of higher values for Rz.
Range of RL
In the case of the current regulators, RL can easily extend up to short circuit at the lower end. However, at
the higher end, if RL is infinite, that is, open circuited, then no Io flows. Therefore there is an upper limit on
the value of the load resistor up to which the current regulation is effective. The load regulation is ensured as
long as the following inequality is satisfied.
Rz RL (4.56)
Problem 4.3
Consider a current regulator wherein Vi = 20 V to 30 V, Vi-nominal = 25 V, Vz = 5.1 V, Re = 10 Ω, RL = 10 Ω,
Rz = 3.6 kΩ and b = 100. How much should be the output voltage for regulation to be effective?
Solution
From Eq. (4.54),
I o ≈ 0.5 A
The line-regulation coefficient is given as
∂I o 1
= = 2.77 × 10−4
∂Vi RL + Rz
If RL increases to say 100 Ω, then Io cannot be maintained at 0.5 A. If Io were to be 0.5 A, then Vo = IoRL =
50 V. This is not possible as the maximum input voltage itself is 30 V. Therefore the output current will
invariably lose regulation. For regulation to be effective, the output voltage should be less than the input
voltage.
Chapter 04.indd 182 3/12/2009 3:50:13 PM

Second-Level Modeling
The second-level models that include the dominant non-idealities of the semiconductor devices can be
introduced to obtain the second-level model of the current regulator circuit. The second-level models of the
transistor and the Zener diode are as discussed in the previous section and are illustrated in Figure 4.23.
Substituting these models in the current regulator circuit, the second-level model equivalent circuit is
obtained as shown in Figure 4.31.
Here the dominant transistor non-idealities like the base-spreading resistance and the base−emitter junc-
tion potential are introduced in the transistor model and the Zener breakdown resistance is introduced in
the Zener model.
Referring to the equivalent circuit of Figure 4.31, the current through Re is given as
Vz + iz rz − ibrb − Vbe
I Re = (4.57)
Re
The output current Io is given as
Vz + iz rz − ibrb − Vbe
I o = I Re + iz = + iz (4.58)
Re
From the output current given by Eq. (4.58), it can be noted that the term iz rz causes loss of regulation due
to the dependence of iz on Vi. The Zener current is given as
iz = iRz − ib
Here ib is constant for a constant Io. However, iRz depends on Vi and is given as
Vi − Vz − iz rz − Vo
iRz = (4.59)
Rz
If iRz is obtained from a constant-current source, then it becomes independent of Vi. This will make iz
independent of Vi which in turn will make the drop iz rz independent of Vi. As a consequence the output
ib Re Io
bib Vbe
rb
Vi RL
Rz iz rz
Vz
Figure 4.31 Equivalent circuit of the current regulator with dominant device non-idealities.
Chapter 04.indd 183 3/12/2009 3:50:13 PM

Re Io
Vi RL
iRz
Vz
Figure 4.32 Current regulator circuit with Rz replaced by a constant-current source.
current Io becomes independent of Vi. The current regulator with the Rz replaced by a constant-current
source is shown in Figure 4.32.
The effect of temperature on Vbe, Vz and b is similar to the discussion in the previous section on series
regulator. The solutions suggested are equally applicable here too.
The linear regulators are a class of DC−DC con- tion, the switched-mode converter topologies are
verters that are dissipative, have low efficiency, are the obvious choice.
bulky due to large heat sink requirements and also There are many commercial IC regulators that
have low power density. However, not withstand- incorporate all the features discussed in this chapter.
ing all the above disadvantages and more, when it Among the commercial IC regulators, the low drop-
comes down to the quality of the output voltage, out regulators are popular for applications demanding
there is no switched-mode converter topology that high performance and efficiency. The input−output
can even closely match the linear regulator differential voltage drop across the series-pass device is
performance. the cause for the loss of efficiency. The load current
Thus if an application demands high ripple rejec- multiplied by the input−output differential voltage
tion and low line-, load- and temperature-regulation drop across the series-pass device of the regulator is
coefficients, then linear regulator is way beyond any the dominant power dissipation of the regulator. In
other DC−DC converter and therefore must be the low dropout regulators this input−output differential
obvious choice. If compactness and high efficiency voltage is significantly lower around 0.5 V or 0.7 V.
are the governing constraints for a specific applica- As a consequence the efficiency is improved.
1. Consider the series regulator circuit shown in Mode of implementation: The above circuit
Figure 4.33. The input is a variable DC power can be studied by
supply that represents the unregulated DC a. Simulation in Spice
input voltage. b. Simulation in SciLAB
c. Hardware bread-boarding
Chapter 04.indd 184 3/12/2009 3:50:14 PM

Q1
Rb
R1
Rz
Vi Q2 Vo RL
R2
Vz
Figure 4.33 Series regulator circuit.
Tasks for study: (i) Instead of connecting one end of Rb to the

(a) Rig up the series regulator circuit as shown input voltage, connect Rb to another fixed
in Figure 4.33 with appropriately selected voltage source. Vary the input voltage
components values. source keeping load fixed and tabulate Vi,
(b) Connect an appropriate load resistor across Vo and Io. From the tabulated results, esti-
the output terminals. mate the line-regulation coefficient. How
(c) Keeping the output load resistor fixed, vary does it compare with the line-regulation
the input voltage and tabulate Vi, Vo and Io. estimate of Step (e)?
(d) Keeping the input voltage at a fixed value, (j) Instead of connecting one end of Rz to the
vary the load resistor and tabulate Vo and Io. input voltage, connect Rz to the output.
Evaluate line and load regulation and
(e) From the tabulated results, estimate the
compare with those evaluated in Step (e).
line- and load-regulation coefficients.
(k) Incorporate a current protection circuit
(f ) From the tabulated results, evaluate the effi-
with the help of an additional NPN tran-
ciency of the linear regulator at various loads
sistor and a current sense resistor.
and input voltages. Plot the efficiency versus
(l) Replace Q2 with a Darlington BJT (like
input voltage using load as a parameter.
TIP122). Measure the effect on the range
(g) Vary R1/R2 ratio and observe the effect on of the load resistor.
the output voltage.
(h) Measure the voltage across Rb for different 2. Consider the three-pin IC regulator circuit
values of the input voltage. Plot the cur- shown in Figure 4.34. The input is a variable
rent through Rb versus Vi. DC power supply that represents the unregu-
lated DC input voltage.
78××
IN OUT
IC
Vi Vo RL
78×× 7805−5 Vreg

7815−15 Vreg
Figure 4.34 Three-pin 78xx series IC regulator.
Chapter 04.indd 185 3/12/2009 3:50:14 PM

Mode of implementation: The above circuit (h) What is the minimum value of the input
can be studied by voltage for which the output is regulated?
3. Consider the 723 IC regulator circuit shown
b. Hardware bread-boarding
in Figure 4.35. The input is a variable DC
Tasks for study: power supply (Vcc) that represents the unreg-
(a) Rig up the circuit as shown in Figure 4.34 ulated DC input voltage. Rsc is the current
using either 7805 or 7815 regulator IC. sense resistor that is used for over-current
(b) Study the data sheet and connect an protection.
appropriate load resistor across the output
terminals.
can be studied by
(c) Keeping the output load resistor fixed,
vary the input voltage and tabulate Vi, Vo
and Io.
(d) Keeping the input voltage at a fixed Tasks for study:
value, vary the load resistor and tabulate (a) Study the data sheet of 723 regulator IC.
Vo and Io. (b) Rig up the IC regulator circuit as shown in
(e) From the tabulated results, estimate the Figure 4.35.
line- and load-regulation coefficients. (c) Connect an appropriate load resistor across
(f ) From the tabulated results, evaluate the the output terminals (start with 1 kΩ).
efficiency of the linear regulator at various (d) Measure the voltage at pin “Vref ”. What is
loads and input voltages. Plot the effi- its value? Does it agree with the data sheet
ciency versus input voltage using load as a value?
parameter. (e) Keeping the output load resistor fixed,
(g) From the tabulated results plot the effi- vary the input voltage and tabulate Vi, Vo
ciency versus input−output differential and Io.
voltage.
Vcc(15 V)
Vcc Vc
4.7 kΩ Vref
Vo
NON
INV 723
4.7 kΩ R2 IC 1 Ω Rsc
INV
CL
COMP
12 kΩ
CS
GND RL
R1
12 kΩ
Figure 4.35 Circuit schematic of 723 IC regulator.
Chapter 04.indd 186 3/12/2009 3:50:15 PM

(f ) Keeping the input voltage at a fixed Mode of implementation: The above circuit
value, vary the load resistor and tabulate can be studied by
Vo and Io. a. Simulation in Spice
(g) From the tabulated results, estimate the b. Simulation in SciLAB
line- and load-regulation coefficients. c. Hardware bread-boarding
(h) From the tabulated results, evaluate the Tasks for study:
efficiency of the linear regulator at various (a) Rig up the current regulator circuit as
loads and input voltages. Plot the efficiency shown in Figure 4.36 with appropriately
versus input voltage using load as a param- selected components values.
eter. (b) Connect an appropriate load resistor across
(i) Vary R1/R2 ratio and observe the effect on the output terminals.
the output voltage. (c) Keeping the output load resistor fixed,
( j) Change the reference voltage applied to vary the input voltage and tabulate Vi, Vo
the “NON INV” pin of the IC. What hap- and Io.
pens to the output? (d) Keeping the input voltage at a fixed
(k) Rsc is the current sense resistor for current value, vary the load resistor and tabulate
protection. Decrease the load resistor value Vo and Io.
and monitor the voltage across the resistor (e) From the tabulated results, calculate the
Rsc. At what value of the load resistor does variation of the output current with respect
current limit come into operation? to the input voltage variation and load
( l) Increase the load current capability by con- variation.
necting an external BJT. Connect the base (f ) From the tabulated results, evaluate the
of the external BJT to “Vo” pin of the IC, efficiency of the linear regulator at various
the collector to “Vc” pin of the IC and the loads and input voltages. Plot the effi-
emitter to the junction of the “CL” pin and ciency versus input voltage using load as a
one terminal of Rsc. By how much is the parameter.
load current capability increased? (g) Change Re and observe the effect on the
4. Consider the current regulator circuit shown in output current.
Figure 4.36. The input is a variable DC power (h) Increase Ro from zero and observe the effect
supply that represents the unregulated DC on the output current. At what value of Ro
input voltage. does the output current lose regulation?
Vcc
Re
Vz
Io
Rz
Ro
Figure 4.36 Current regulator circuit.
Chapter 04.indd 187 3/12/2009 3:50:15 PM


1. In linear regulators the power semiconductor 11. The no-load condition for the current regulator
devices are operated in the _______ region. is when the output resistor value is _______.
2. The output DC voltage value is always _______ 12. Efficiency primarily affects the volumetric
than the input unregulated DC voltage value. _______ of the linear regulator.
3. Even though the linear regulators have low 13. The power dissipated in the series-pass element
_______, the quality of the output voltage with determines the size of the _______ that needs
respect to the variations in _______, _______ to be employed.
and _______ are better than switched-mode
14. The first-level model for the transistor is a three-
regulator.
wire node with the collector current being a
4. In a simple shunt regulator, the shunt com- _______ current source of value _______.
ponent is a _______ that is operated in the
15. The first-level model of a Zener diode is a
_______ region.
_______ source of value _______.
5. In a shunt regulator, the output voltage is
16. For a series regulator, as the output resistor
regulated by virtue of the Zener diode being
value approaches infinity, the output voltage is
operated in the _______ region.
_______ of load.
6. The current absorbs the changes in the _______
17. For a series regulator, as the output resistor
and _______ currents.
value approaches zero, the output voltage
7. In a shunt regulator, under no-load condition, becomes more and more _______ on load.
the whole input current flows into the _______.
18. Th e second-level model for the transistor
8. A transistor or BJT operating in the linear includes the _______ and the base−emitter
region can be used as a _______ element in a _______.
series regulator.
19. The second-level model of a Zener diode
9. If the input voltage is a constant, then the current includes the _______.
through the bias resistor Rb of the series regulator
20. The load regulation of a current regulator is
can be considered as a _______ current.
ensured as long as _______.
10. Current regulators are built using _______
regulators.
1. How are voltage scaling and regulation achieved 4. Explain the operation of a Zener-based shunt
in a linear regulator? regulator.
2. What is the efficiency of a generic linear regu- 5. Discuss the operation of the series regulator.
lator?
6. Discuss the effect of input voltage variation on
3. What are the various linear regulator topologies? the output voltage.
Chapter 04.indd 188 3/12/2009 3:50:15 PM

7. Draw a circuit schematic of a negative voltage 19. What is the effect of the current flowing through
regulator and explain its operation. the output sense resistors R1 and R2 on the
output voltage?
8. What are the categories of the IC linear regula-
tors? Give examples for each category. 20. What is the range of the load for a series linear
regulator?
9. What is four-wire connection in linear regula-
tors and why is it needed? 21. What is the effect of the output resistor RL on
the output voltage regulation?
10. Discuss the protection features needed to be
incorporated in the linear regulator circuits. 22. Discuss the constraints on the range of the
output resistor, RL.
11. Give a current protection circuit for a series linear
regulator. If the load over-current limit is to be 23. What are the second-level models for BJT and
set at 5 A, then choose the components of the Zener diode?
protection circuit.
24. Using the second-level model of the series regu-
12. Explain the operation of the crow-bar voltage lator, discuss the effect of Vi on the output
protection circuit. regulation.
13. What are the quantitative measures of a linear 25. Does the Zener breakdown resistance rz have an
regulator that determine its quality? effect on the output voltage regulation? Explain.
14. Define line-regulation, load-regulation and 26. What are the parameters of the series regula-
temperature-regulation coefficients for a linear tor that are affected by temperature? What is
regulator. the effect of temperature on output voltage
regulation?
15. In the analysis of linear regulators, what is first-
level modeling? 27. How does one compensate for the temperature
variation of the BJT’s base−emitter junction
16. What are the first-level models for BJT and Zener
potential?
diode?
28. How does one compensate for the Zener diode’s
17. In a series regulator, using the first-level model,
breakdown voltage variation with temperature?
what is the relation between output and the
base current of the series-pass transistor?
18. In a series regulator, how is the current through
the bias resistor Rb made independent of the
input voltage?
| PROBLEMS
1. A linear regulator is used to regulate an un- source? What is the efficiency of the linear
regulated DC source with nominal voltage regulator?
of 15 V. The shunt resistor value is 100 Ω and
the series resistor value is 250 Ω. A load resis- 2. For Problem 1, if there is no shunt resistor, then
tor of 1 kΩ is connected across the output. what is the output voltage? What is the input
What is the output voltage? What is the input power drawn from the unregulated input source?
power drawn from the unregulated input What is the efficiency of the linear regulator?
Chapter 04.indd 189 3/12/2009 3:50:15 PM

3. Consider a Zener shunt regulator circuit wherein is the power rating of the series resistance used
the input voltage Vi varies between 10 V and in the regulator? What is the load resistor
20 V and the Zener diode used is a 5 V, 1 W range? If the Zener has a breakdown resis-
device. If the maximum allowable Zener current tance of 0.1 Ω, what is the output voltage
is 150 mA, then what is the value and power rat- variation?
ing of the series resistance used in the regulator?
7. For the current regulator circuit shown in
4. For Problem 3, what is the load resistor range? Figure 4.38, calculate the value of the load cur-
rent. What is the upper limit on the output
5. For the shunt regulator of Problem 3, if the
voltage?
Zener has a breakdown resistance of 0.1 Ω,
what is the output voltage variation? 8. For Problem 7, what is the range of the load
resistor?
6. Consider the shunt regulator circuit shown in
Figure 4.37. The Zener is a 1 W device with the 9. In a series regulator, the output voltage sens-
maximum allowable Zener current as 150 mA. ing resistors R1 and R2 are 5.6 kΩ each. The
The base−emitter voltage drop is 0.6 V. What output voltage is a regulated 15 V. The Zener
hFE = 100
Q
100 Ω
Io
10−20 V
5.1 V Vo
RL
Figure 4.37 Problem 6.
+
5Ω
5.6 V
− Q (b = 100)
12−18 V
Rz 1K
10 Ω Vo
Chapter 04.indd 190 3/12/2009 3:50:15 PM

diode value is 5.1 V. Using the fi rst-level 17. In a series regulator, the output voltage sensing
model, calculate the base current through the resistors R1 and R2 are 5.6 kΩ each. The output
transistor Q2. voltage is a regulated 15 V for a nominal input
voltage of 25 V. The Zener diode value is 5.1 V
10. The input voltage of a series regulator varies
and it is biased with a 20 kΩ resistor from the
between 20 V and 30 V. The output is a regu-
input. Using the second-level model, calculate
lated 15 V DC. If the bias resistor Rb has a
the base current through the transistor Q2.
value of 2 kΩ, then using the first-level model,
(Assume Vbe = 0.6, rb = 0.2 Ω for BJTs and rz =
estimate the nominal current through the bias
0.1 Ω for Zeners. The saturation current gains
resistor.
of Q1 and Q2 are 100.)
11. Consider a linear series regulator with R1 = 3.6 18. For Problem 17, the input voltage of a series
kΩ, R2 = 2.5 kΩ, Rb = 3.6 kΩ, Rz = 20 kΩ, regulator varies between 20 V and 30 V. The
Vz = 7 V, RL = 1 kΩ, Vi = 20 V to 30 V, output is a regulated 15 V DC. If the bias resis-
Vi-nominal = 25 V, b1 = b2 = 100. If the output tor Rb has a value of 2 kΩ, then using the
voltage is expressed as Vo = k1Vi + k2Vz , evalu- second-level model, estimate the nominal cur-
ate k1 and k2 using the first-level model. rent through the bias resistor for a load of 1 A.
12. In Problem 11, the current gain of Q1 is 19. Consider a linear series regulator with R1 =
increased to 1000. What is the effect on k1 and 3.6 kΩ, R2 = 2.5 kΩ, Rb = 3.6 kΩ, Rz = 20 kΩ,
k2? Explain the effect or lack of effect. Vz = 7 V, RL = 1 kΩ, Vi = 20 V to 30 V, Vi-nominal
13. In Problem 11, the current gain of Q2 is = 25 V, b1 = b2 = 100. If the output voltage is
increased to 1000. What is the effect on k1 and expressed as Vo = k1Vi + k2Vz + k3Vbe1 + k4Vbe2,
k2? Explain. evaluate k1, k2 k3 and k4 using the second-level
model. (Assume Vbe = 0.6, rb = 0.2 Ω for BJTs
14. Consider a linear series regulator with R1 = 3.6 and rz = 0.1 Ω for Zeners.)
kΩ, R2 = 2.5 kΩ, Rb = 3.6 kΩ, Rz = 20 kΩ,
Vz = 7 V, RL = 1 kΩ, Vi = 20 V to 30 V, 20. In Problem 19, the current gain of Q1 is
Vi-nominal = 25 V. Using the first-level model, increased to 1000. What is the effect on k1 and
find the line-regulation coefficient for (a) b1 = k2? Explain the effect or lack of effect.
b2 = 100, (b) b1 = 1000 and b2 = 100, (c) b1 = 21. In Problem 19, the current gain of Q2 is
100 and b2 = 1000. increased to 1000. What is the effect on k1 and
15. For Problem 14, using the first-level model, k2? Explain.
find the load-regulation coefficient for (a) b1 = 22. Consider a current regulator wherein Vi = 20 V
b2 = 100, (b) b1 = 1000 and b2 = 100, (c) b1 = to 30 V, Vz = 8.2 V, Re = 10 Ω, RL = 10 Ω,
100 and b2 = 1000. Rz = 3.6 kΩ and b = 100. What is the constant
output current using the first-level model?
16. In the series regulator of Problem 14, if Rb is
Apply and compare with the output current
replaced by a constant-current source regulator,
estimated with the second-level model.
then estimate the load-regulation coefficient
for (a) b1 = b2 = 100, (b) b1 = 1000 and b2 = 23. For Problem 22, find the line-regulation coeffi-
100, (c) b1 = 100 and b2 = 1000. cient.
Chapter 04.indd 191 3/12/2009 3:50:16 PM

| ANSWERS
Fill in the Blanks
1. linear 7. Zener 15. DC; Vz
2. lesser 8. series-pass 16. independent
3. efficiency; input voltage; tem- 9. constant 17. dependent
perature; load 10. voltage 18. base-spreading resistance; junc-
4. Zener diode; reverse break- 11. zero tion potential
down 12. power density 19. breakdown resistance
5. reverse breakdown 13. heat sink 20. Rz RL
6. input; load 14. dependent; bib
Chapter 04.indd 192 3/12/2009 3:50:16 PM

DC–DC Switched-Mode Converters
Learning Objectives
CHAPTER
5
analyze the steady-state operation of a DC–DC converter.
design non-isolated and isolated DC–DC converters.
understand the process of drawing and analyzing waveforms.
understand soft-switching converters.
I n the previous chapter DC–DC converters were discussed wherein the power devices were operated in the
linear region. As a consequence the power devices were acting as power dissipators. This results in reduced
conversion efficiency. If the power devices are operated either in the cut-off region or in the full-on or satura-
tion region, they act as switches. DC–DC converters that are based on this principle are called the switched-
mode converters. The switched-mode DC–DC converters would ideally have 100% efficiency. However,
due to switching and conduction losses this value is lower than 100%. After the voltage is switched with
a power switch, the voltage has a switched wave shape that needs to be filtered to obtain a DC output volt-
age. The filtering or averaging is done using non-dissipative components like the inductor and capacitor.
Therefore, any switched-mode converter will include at least a power switch that emulates a single pole
double throw (SPDT) mechanical switch, an inductor and a capacitor.
5.1 Single Pole Double Throw Switch
A single pole double throw (SPDT) switch is shown in Figure 5.1(a). The switch consists of a pole P and
two throws T1 and T2. The pole makes contact with either T1 or T2 at any given instant. Referring to
Chapter 1, it is seen that no single power device can behave as an SPDT switch. However, the power devices
can behave as a single pole single throw (SPST) switch. Therefore, from the point of view of implementation
using the power devices, the SPST switch is available as a single device. The SPDT switch can be constructed
using SPST switches as shown in Figure 5.1(b). Here the SPST switch S1 is linked to throw T1 and the
SPST switch S2 is linked to throw T2. At any given instant of time it should be noted that either S1 is ON
or S2 is ON. Such an operational topology will emulate an SPDT switch.
Few practical realizations of the SPDT switch are shown in Figure 5.2. These realizations are based on
the generic topology of Figure 5.1(b). In the circuits of Figure 5.2, the power switch represented as bipolar
transistor (BJT) can be replaced by either a metal oxide semiconductor field effect transistor (MOSFET) or
Chapter 05.indd 193 3/19/2009 1:30:15 PM

T1
T1
S1
P P
S2
T2
T2
(a) (b)
Figure 5.1 (a) SPDT switch; (b) two SPST switches used to emulate an SPDT switch.
T1 T1 T1
T1
S1 S1 S1 S1
P P P P
S2 S2 S2 S2
T2 T2 T2 T2
(a) (b) (c) (d)
Figure 5.2 Few realizations of the SPDT switch using power devices.
an insulated gate bipolar transistor (IGBT) without loss of generality. In Figure 5.2(d), the diodes are the
internal body diodes of the BJT or MOSFET or IGBT that is used. This is in fact one of the most popular
of all topologies and is extensively used in both DC–DC converters and DC–AC inverters.
5.2 The Chopper
T he chopper is a power converter wherein the input DC voltage is chopped and sent to the output with-
out incorporating any filters exclusively within the converter. The filtering of the voltage or current is the
responsibility of the applied load. The schematic of the chopper is illustrated in Figure 5.3(a). The simplest
Chapter 05.indd 194 3/19/2009 1:30:17 PM

DC–DC Switched-Mode Converters 195
v v
t
t
Vin Chopper Vo
(a)
P
T1
Vin T2 Vo
Chopper
(b)
T1
S1
Vin P
S2 Vo
T2
Chopper
(c)
Figure 5.3 Basic chopper: (a) Block schematic; (b) SPDT representation; (c) realization
using semiconductor switches.
chopper can be constructed using only one SPDT switch. As the output is an unfiltered chopped waveform,
there is no need for an inductor and capacitor. Figure 5.3(b) gives the chopper representation with an SPDT
switch and Figure 5.3(c) gives the practical realization of the chopper. The switches S1 and S2 are IGBT
switches with internal body diodes. The IGBT switches may be replaced with MOSFET or BJT switches.
The switch S1 is turned ON by appropriate application of the gate-to-emitter pulse. During this time, the
switch S2 is in the OFF-state. This duration is for a period Ton. For the period Toff , the switch S1 is turned
OFF and the switch S2 is turned ON. The pole voltage at P is a switching waveform as shown in Figure 5.4.
Chapter 05.indd 195 3/19/2009 1:30:17 PM

Vo
Vin
Ton
Toff
0 t
Ts
Figure 5.4 Chopper output waveform.
Referring to Figure 5.4,
Ts = Ton + Toff
where Ts is one switching period. A useful and important term that is widely used in the power electronic
literature is the duty ratio or the duty cycle. The duty ratio or the duty cycle (D) can be defined as the ratio
of the ON-time of switch S1 when the input is connected to the pole to the total switching period. This is
given as
Ton T
D= = on (5.1)
Ton + Toff Ts
The output voltage of the chopper is a switched waveform with an average value of
Ts Ton
1 1 V
Vo = ∫ vo dt = ∫ Vidt = Ti Ton = DVi
Ts 0 Ts 0 s
The chopper circuit is used in applications like DC motor drives wherein the motor’s armature inductance
filters the armature current flowing through it. Figure 5.5(a) shows the chopper connected to the armature
circuit of a separately excited DC motor. In Figure 5.5(b), the DC motor is replaced by an equivalent circuit
that acts as the load on the chopper. The DC motor is represented by the armature resistance Ra, the arma-
ture inductance La and the speed-dependent back emf source eb.
The DC motor can operate in any one of the four operating zones as shown in the v –i characteristic of
Figure 5.6, namely, (a) forward motoring mode or first quadrant operation, (b) reverse motoring mode or
the third quadrant operation, (c) forward generation mode or second quadrant operation, (d) reverse gen-
eration mode or fourth quadrant operation. The chopper drive shown in Figure 5.5 can work in only first
and second quadrants as it cannot supply negative average output voltage. Therefore, for this chopper drive
the motor can operate in only two modes: (a) forward motoring and (b) forward generation. These modes
are indicated in Figures 5.7 and 5.8.
In the forward motoring mode the power flows from the DC source Vi to the motor shaft. The armature
voltage and current are positive wherein the armature current ia flows into the motor +ve terminals. In this
mode the back emf source eb acts as a sink or a load. In one chopper switching period Ts, S1 is ON during
Chapter 05.indd 196 3/19/2009 1:30:17 PM

Load
Vi
DC
Chopper
(a)
ia
Vi
Ra La
DVi = Va eb = kw
Chopper DC motor equivalent
(b)
Figure 5.5 (a) Chopper connected to a separately excited DC motor; (b) circuit schematic
with electrical equivalent of DC motor.
va
Forward Forward
generating motoring
ia
Reverse Reverse
motoring generating
Figure 5.6 Four quadrant operation of DC motor.
Chapter 05.indd 197 3/19/2009 1:30:17 PM

S1
ia Acts as
La sink
Vi
Ra +
S2 eb = kw
−
−
Ton = DTs Toff = (1 − D)Ts

(a)
+ +
S1 S1
La La
Ra Ra
Vi Vi
ia + +
S2 eb S2 ia eb
− −
− −
(b) (c)
Figure 5.7 Forward motoring: (a) Forward motoring mode; (b) operative circuit during DTs;
(c) operative circuit during (1 – D)Ts.
DTs period and S2 is ON during (1 – D)Ts period. For DTs period, the operative portion of the circuit is
shown in Figure 5.7(b). As the armature current ia flows into the motor terminals as shown, the IGBT of
switch S1 is operative and is indicated in Figure 5.7(b). For (1 – D)Ts period, the operative portion of the
circuit is shown in Figure 5.7(c). During this time, only the diode of S2 is operative wherein the armature
current freewheels through the diode as shown in Figure 5.7(c).
In the forward-generation mode the power flows from the motor shaft to the DC source Vi. This occurs
during braking of the motor wherein the mechanical inertia translates to the electric domain and charges up
the DC source Vi. The armature voltage is positive as in forward motoring mode, but the armature current
is negative in a direction wherein the armature current ia flows out of the motor +ve terminals. In this mode
the back emf source eb acts as a generator. In one chopper switching period Ts, the switch S1 is ON during
DTs period and the switch S2 is ON during (1 – D)Ts period. For (1 – D)Ts period, the operative portion
of the circuit is shown in Figure 5.8(c). During this time, the IGBT of S2 is operative wherein the armature
current flows as shown in Figure 5.8(c) and charges up the energy in the armature inductance. For DTs
period, the operative portion of the circuit is shown in Figure 5.8(b) wherein the energy of the armature
inductance discharges through the diode and into the DC supply Vi as indicated in Figure 5.8(b).
A single basic chopper consisting of the SPDT switch can handle only two quadrants of operation
wherein the DC applied voltage being constant, the chopper output current (ia) in the DC motor example
Chapter 05.indd 198 3/19/2009 1:30:18 PM

Acts as
S1 source
ia
Vi
Ra La +
eb = kw
S2 −
−
DTs (1 − D)Ts
(a)
+ +
ia
S1 S1
Vi Vi
Ra La Ra La
+ +
S2 eb S2 eb
ia
− −
− −
(b) (c)
Figure 5.8 Forward regeneration: (a) Forward-regeneration mode; (b) operative circuit
during DTs; (c) operative circuit during (1 – D)Ts.
illustrated above can flow in either direction. However, if all four quadrant operation is needed wherein both
the chopper output voltage and current are bi-directional, then two basic choppers are required. The voltage
across the poles of the two SPDT switches, VP1P2 can handle bi-directional voltages as well. The schematic
of the dual–four quadrant chopper is shown in Figure 5.9. Figure 5.10 shows the application of the dual
chopper to the DC motor load for four quadrant operation.
P1 Load P2
Vi
VP1P2
Figure 5.9 Schematic of the dual chopper using two SPDT switches.
Chapter 05.indd 199 3/19/2009 1:30:18 PM

S3
S1
DC motor load
Vi P1 P2
Ra La eb
S4
S2
Figure 5.10 Circuit schematic of the dual chopper with DC motor load across the poles
P1 and P2.
5.3 DC Steady-State Principles
T he basic chopper is combined with the inductors and the capacitors to obtain the DC output. The fol-
lowing sections will discuss DC–DC converters using chopper, inductors and capacitors in various
topological combinations. The converters will be discussed with respect to their steady-state operation. The
dynamic modeling of the converters will be dealt with in Chapter 8.
The analyses presented in this chapter are based on the following two fundamental physical laws that
govern the steady-state operation of the switched-mode power converters:
1. Volt-Second Balance: The inductor in a circuit will not support an average DC voltage across it. This
implies that under steady-state conditions,
Ts
∫ vL dt = 0
0
where vL is the voltage across the inductor and Ts is the switching period. This means that the product
of voltage and time or, in other words, the area under the inductor voltage curve in one period should
be zero under equilibrium conditions. This is called the volt-second balance for inductors or windings.
2. Amp-Second Balance or Charge Balance: The capacitor in a circuit will not allow an average DC cur-
rent to pass through it. This implies that under steady-state conditions,
Ts
∫ iC dt = 0
0
where iC is the current through the capacitor and Ts is the switching period. This means that the
product of the current through the capacitor and time or, in other words, the area under the capacitor
current curve in one period should be zero under equilibrium conditions. This is called the amp-second
balance for capacitors. As amp-second is the same as the charge on the capacitor, it is also called
charge balance. This implies that there is neither decrease nor increase in the charge on the capacitor
under equilibrium conditions.
Chapter 05.indd 200 3/19/2009 1:30:18 PM

5.4 The Three Basic Converters
T his section discusses three converters synthesized using topological variations of a basic chopper, an
inductor and a capacitor. While synthesizing any converter the following two rules must be strictly
followed to obtain a practically realizable DC–DC converter:
1. The inductor stores energy by virtue of the current flowing through it [(1 / 2)Li 2]. This implies that the
current through the inductor cannot change instantaneously. Therefore, the inductor should be placed
such that the inductor current does not become discontinuous during any instant or mode of opera-
tion.
2. The capacitor stores energy by virtue of the voltage across it [(1 / 2)Cv 2]. This implies that the voltage
across the capacitor cannot change instantaneously. Therefore, the capacitor should be placed such that
the capacitor voltage does not become discontinuous during any instant or mode of operation.
Based on the application of the rules mentioned above, three converter variants are discussed in this section.
The three variants are (a) step-down or the buck converter, (b) step-up or the boost converter and (c) step-
up/down or buck–boost converter. These three converters are called the primary converters as other convert-
ers can be synthesized using these converters and so they form the basic building blocks.
Step-Down or Buck Converter

Topology
The step-down or the buck converter as the name suggests, gives an output voltage that is always smaller
than the input voltage. The buck converter topology is illustrated in Figure 5.11. Figure 5.11(a) shows the
buck converter consisting of an SPDT switch, an inductor that is connected to the pole of the SPDT switch
and a capacitor that is connected across the output as shown. The input DC voltage source Vi is connected
across the throws, T1 and T2, of the SPDT switches. Observe that the inductor is connected to the pole of
the SPDT switch where there is no discontinuity in the current. Figure 5.11(b) shows the buck converter
with the SPDT switch replaced by two SPST switches S1 and S2. Figure 5.11(c) shows the operative circuit
during the period DTs of a switching period. During this time, S1 is ON and S2 is OFF. The pole P is con-
nected to the input Vi. The inductor current flows from the input source to the output as shown.
Figure 5.11(d) shows the operative circuit during the period (1 – D)Ts of the switching period. During this
time S1 is OFF and S2 is ON. The inductor current cannot change instantaneously. The inductor current
starts to decrease. This negative di/dt of the inductor current develops sufficient voltage (Ldi/dt) with
a polarity such as to drive the inductor current through the switch S2. This is called the freewheeling action
of the inductor current.
Operation
The operation of the buck converter may be analyzed by visualizing the waveforms of the various signals of
the circuit. In general for any converter, the following signals may be considered important for analysis:
1. voltage across the inductor;
2. current through the inductor;
3. load current;
4. current through the capacitor;
5. input source current;
6. voltages across switches S1 and S2;
7. currents through switches S1 and S2.
Chapter 05.indd 201 3/19/2009 1:30:18 PM

T1
S1
T1
P
Vi Vi
L L
T2
C RL S2 C RL
T2
(a) (b)
T1 DTs period T1 (1 − D)Ts period

S1 iL S1
Io iL Io
Vi P Vi P
L ic ic
S2 C Vo S2 C Vo
T2 T2
(c) (d)
Figure 5.11 Step-down or buck converter: (a) Buck converter with a SPDT switch, an L and
a C; (b) SPDT switch is replaced with two switches S1 and S2; (c) operative
circuit during the period DTs; (d) operative circuit during the period (1 – D)Ts.
Figure 5.12 shows the plot of the various signals for the buck converter under steady-state operating
conditions.
DTs Duration The switch S1 is ON and S2 is OFF during the period DTs. The voltage vL across the
inductor during this time is Vi – Vo. Both Vi and Vo are DC quantities and hence Vi – Vo will be a DC
quantity during the period DTs. From the Faraday’s law, the inductor voltage and current are related as,
vL = L(diL / dt ). This implies that the current through the inductor is given as
1 1 V − Vo
L ∫ vL dt = ∫ (Vi − Vo )dt = i
iL =
L L ∫
dt
As Vi – Vo is a constant, the inductor current iL rises linearly with a slope of (Vi – Vo)/L as indicated in
Figure 5.12. During this time the current through S1 is the same as the inductor current. The voltage across
S2, which is OFF, is Vi as the pole P is connected to the input source Vi.
(1 – D)Ts Duration During the time when S1 is OFF and S2 is ON, the voltage vL across the inductor
during this time is –Vo. Again as Vo is a DC quantity, the inductor current iL falls linearly with a slope of
–Vo/L as indicated in Figure 5.12. During this time, the current through S2 is the same as the current
through the inductor. The voltage across S1, which is OFF, is Vi as the pole P is connected to the circuit
ground. The inductor current is composed of two parts: (a) the DC component (Io) and (b) the AC compo-
nent (ic). The DC component cannot flow through the capacitor, therefore the DC component flows into
the load RL. The AC component alone will flow through the capacitor. Therefore, the capacitor current ic
is the inductor current minus the load DC component.
Chapter 05.indd 202 3/19/2009 1:30:18 PM

S1 S2 S1 S2
ON ON ON ON
DTs (1 – D)Ts
Vi − Vo
vL
0
t
−Vo
(Vi − Vo)/L −Vo /L

Io
iL
ΔiL
0
t
iC
Ts /2
0
t
ΔiL Ts /2
(Vi − Vo)/L Io
Iin
iS1
0 −Vo /L Io t
iS2
0
t
Vi Vi
VS1
0
t
Vi Vi
VS2
0
t
Figure 5.12 Waveforms of various signals of a buck converter.
Chapter 05.indd 203 3/19/2009 1:30:19 PM

Semiconductor Switches
The buck converter schematic of Figure 5.11 shows only the conceptual SPDT and SPST switches.
The SPST switches may be replaced with power semiconductor switches as shown in Figure 5.13. Referring
to the operating modes given in Figures 5.11(c) and (d) and the OFF-state voltages across S1 and S2 given
in Figure 5.12, the switch S1 can be either a BJT or a MOSFET or an IGBT and the switch S2 is a diode
that is required for freewheeling the inductor current when S1 is OFF. This implementation is shown in
Figure 5.13(a). However, the power semiconductor switches like MOSFETs and IGBTs are mostly available
along with their body diodes in-built; therefore, in order to reduce the inventory, the switch configuration
of Figure 5.13(b) is also common. Here the IGBT part of S1 and the diode part of S2 are the operative
components for this topology.
Steady-State Analysis
Based on the application of the volt-second balance and charge balance rules, the input–output relationship
under steady-state conditions can be obtained.
With Ideality Constraints The analysis is performed by assuming the following:
1. The switches S1 and S2 have no ON-state drops.
2. The winding resistance of the inductor is zero.
3. The equivalent series resistance (ESR) of the capacitor is zero.
4. There are no losses in the converter implying that the efficiency is 100%.
Input–output relationship To obtain the input–output voltage relationship, apply the volt-second balance
rule to the inductor. This implies that the area under the inductor voltage curve in one period under steady-
state conditions should be zero. Referring to Figure 5.12,
(Vi − Vo )DTs + (−Vo )(1 − D )Ts = 0

Simplifying the above equation, one obtains
Vo = DVi (5.2)
S1
S1
Vi P
Vi L
P
L C RL Vo
S2
S2 C Vo RL
(a) (b)
Figure 5.13 Realization of buck converter with power semiconductor switches.
Chapter 05.indd 204 3/19/2009 1:30:19 PM

The duty ratio (D) as defined in Eq. (5.1) can take on values between the closed interval 0 and 1. This means
that the output voltage Vo is always less than the input voltage Vi. Hence this topology is called a step-down
converter or buck converter. To obtain the input–output current relationship one can use the 100% efficiency
constraint. This implies that
Vi I in = Vo I o (5.3)
Using Eq. (5.2) in Eq. (5.3), one obtains
I in = DI o (5.4)
Inductor value The value of the inductor L is calculated based on the amount of current ripple ΔiL that the
designer would like to allow for a given application. Referring to the waveforms given in Figure 5.12, the
slope of the inductor current waveform during the period DTs is given as
diL ΔiL ΔiL Vi − Vo
= = = (5.5)
dt Δt DTs L
Likewise the slope of the inductor current waveform during the period (1 – D)Ts is given as
diL ΔiL ΔiL −V
= = = o (5.6)
dt Δt (1 − D )Ts L
The voltages Vi and Vo are known from the converter specifications. D can be obtained from the input–
output relationship of Eq. (5.2). The switching frequency fs is a design choice. From this Ts = 1/fs is obtained.
ΔiL is also a design choice. Though this is chosen based on the application, the common choice in most cases
is ΔiL = 10% of Io. The only unknown L can be calculated from either Eq. (5.5) or Eq. (5.6). Between these
two equations the latter is preferred as it is independent of Vi which is an unregulated voltage. Taking the
absolute value of the current slope from Eq. (5.6), the value of inductor is given as
Vo (1 − D )
L= (5.7)
ΔiL f s
To regulate Vo with variations in Vi, the value of D should be changed. Thus if Vimax is the maximum input
voltage swing, then Dmin will be the corresponding minimum duty cycle to obtain a specified Vo. If Vimin is
the minimum input voltage swing, then Dmax will be the corresponding maximum duty cycle to obtain the
specified Vo. Thus, for a regulated Vo,
Vo = Vimax × Dmin = Vimin × Dmax (5.8)
The values Vo, Vimax and Vimin are known from specifications of the converter. Dmax is a design choice that is
dependent on the specific application. Then
Vo V × Dmax
Dmin = = imin
Vimax Vimax
The inductor current should have a current ripple that is less than or equal to the specified ΔiL. Therefore the
inductor value should be calculated at the duty ratio corresponding to Dmin. Thus,
Chapter 05.indd 205 3/19/2009 1:30:19 PM

Vo (1 − Dmin )
L= (5.9)
ΔiL f s
Capacitor value The capacitor value is calculated by applying the amp-second rule. Referring to the capaci-
tor current waveform given in Figure 5.12, if there is to be no charge build up or charge reduction in the
capacitor, then the area under the capacitor current curve in one period should be zero. The area under the
positive portion of the current curve implies charging of the capacitor and that under the negative portion
implies discharging of the capacitor. For charge balance both these areas should be equal. The change is
the capacitor charge ΔQ is given by the area under either the positive portion or the negative portion of the
capacitor current curve. Thus,
1 1 Ts ΔiL ΔiL
ΔQ = C ΔVo = × Base × Height = = (5.10)
2 2 2 2 8 fs
ΔiL
C= (5.11)
8ΔVo f s
Here ΔVo is known from the output ripple specification. ΔiL and fs are design choices as mentioned earlier.
The value of C should be greater than the value calculated by Eq. (5.11).
Effect of Non-Idealities Till now, the analysis assumed the switches, inductor and capacitor as ideal.
However, practical components are far from ideal. The switches have ON-state drops, the inductor has
winding resistance, the capacitor has the ESR that contributes to output voltage ripple and reduction in
efficiency, inductor core material has core losses that reduce the efficiency and so on. As a consequence, the
output of the practical buck converter will not exactly agree with the input–output relationship derived with
the ideality constraints. Figure 5.14 shows the buck converter schematic with few non-idealities included.
The ON-state drops of the switches are included as voltage sources in series with the SPDT switches. The
source Vcesat in series with the IGBT switch represents the ON-state drop of S1 when S1 is ON. The source
VD in series with the diode switch represents the ON-state drop of S2 when S2 is ON. The resistance rL
represents the inductor winding resistance and resr represents the capacitor ESR.
To derive a more exact input–output voltage relationship, one will have to apply the volt-second and
amp-second balance to the circuit of Figure 5.14 wherein the non-idealities are incorporated in the circuit
schematic. The design of the capacitor will also get affected due to the presence of the non-idealities espe-
cially the ESR non-ideality. The output voltage ripple will contain contributions from the voltage variation
due to ESR and that due to the change of charge as discussed earlier. Therefore for a specified output voltage
ripple, the voltage ripple contribution due to change of charge should be much lesser than when ESR is
absent. This implies that the capacitance value will be higher for a given output voltage ripple.
Step-Up or Boost Converter

Topology
The step-up or the boost converter, as the name suggests, gives an output voltage that is always higher
than the input voltage. The boost converter topology is illustrated in Figure 5.15. Figure 5.15(a) shows
the boost converter consisting of an SPDT switch, an inductor that is connected to the pole of the SPDT
Chapter 05.indd 206 3/19/2009 1:30:19 PM

T1
+
Vcesat
−
S1
L rL
Vi P
S2
resr
− RL Vo
VD
C
+
T2
Figure 5.14 Buck converter with non-idealities included.
T1 T1
S1
L P L
P
C RL C RL
Vi Vi S2
T2 T2
(a) (b)
DTs period (1 − D)Ts period
T1 T1
S1 Io S1 Io
ic
L iL L iL
P P
C RL Vo C RL Vo
Vi S2 Vi S2
ic
T2 T2
(c) (d)
Figure 5.15 Step-up or boost converter: (a) Boost converter with a SPDT switch, an L and
a C; (b) SPDT switch is replaced with two switches S1 and S2; (c) operative
circuit during the period DTs; (d) operative circuit during the period (1 – D)Ts.
Chapter 05.indd 207 3/19/2009 1:30:19 PM

switch and a capacitor that is connected across the throws of the SPDT switch as shown. Observe that
here also the inductor is connected to the pole of the SPDT switch where there is no discontinuity in the
current. However note that the inductor is connected between the pole and the input source in the boost
configuration whereas in the buck configuration the inductor is connected between the pole and the
output. Figure 5.15(b) shows the boost converter with the SPDT switch replaced by two SPST switches
S1 and S2. Figure 5.15(c) shows the operative circuit during the period DTs of a switching period.
During this time, S2 is ON and S1 is OFF. The pole P is connected to the circuit ground. The inductor
current flows from the input source to energize the inductor. Figure 5.15(d) shows the operative circuit
during the period (1 – D)Ts of the switching period. During this time S1 is ON and S2 is OFF. The
inductor current cannot change instantaneously. The inductor current starts to decrease. This negative di/dt
of the inductor current develops sufficient voltage (Ldi/dt) with a polarity such as to drive the inductor
current through S1 to charge the output capacitor.
Operation
The operation of the boost converter is analyzed by visualizing the waveforms of the various signals of the
circuit. The following signals are considered important for analysis:
3. load current;
Figure 5.16 shows the plot of the various signals of the boost converter under steady-state operating
conditions.
DTs Duration The switch S1 is OFF and S2 is ON during the period DTs. The voltage vL across the
inductor during this time is Vi which is a DC quantity. From the Faraday’s law, the inductor voltage and
current are related as vL = L(diL / dt ) . This implies that the current through the inductor is given as
1 1 V
iL =
L ∫ vL dt = ∫ Vi dt = i ∫ dt
L L
As Vi is a constant, the inductor current iL rises linearly with a slope of Vi/L as indicated in Figure 5.16.
During this time the current through S2 is the same as the inductor current. The voltage across S1, which is
OFF, is Vo as the pole P is connected to the circuit ground.
(1 – D)Ts Duration During this time when S1 is ON and S2 is OFF, the voltage vL across the inductor is
Vi – Vo. In the boost converter, Vo is always greater than Vi and therefore Vi – Vo is a negative quantity. The
inductor current iL falls linearly with a slope of (Vi – Vo)/L as indicated in Figure 5.16. During this time, the
current through S1 is the same as the current through the inductor. The voltage across S2, which is OFF, is
Vo as the pole P is connected to the output Vo.
The inductor current is composed of two parts: (a) the DC component (Iin) and (b) the AC component
(ΔiL). Here also the inductor current ripple is a design choice. The S1 switch current (iS1) also has two dis-
tinct components as shown in Figure 5.16: (a) the average or DC component (Io) and (b) the AC component
Chapter 05.indd 208 3/19/2009 1:30:19 PM

S2 S1 S2 S1
ON ON ON ON
DTs (1 – D)Ts
Vi
vL
0
t
Vi − Vo
(Vi − Vo)/L
Vi /L Iin
iL
ΔiL
0
(Vi − Vo) /L t
iC
0
t
Io Io
iS2 Vi /L Iin
0
(Vi − Vo) /L Iin Io t
iS1
0
Vo Vo t
VS2
0
t
Vo Vo
VS1
0
t
Figure 5.16 Waveforms of various signals of a boost converter.
Chapter 05.indd 209 3/19/2009 1:30:20 PM

(ic). The DC component cannot flow through the capacitor, therefore it flows into the load RL. The AC
component alone will flow through the capacitor. Therefore, the capacitor current ic is the S1 switch current
minus the load DC component.
The boost converter schematic of Figure 5.15 shows only the conceptual SPDT and SPST switches. The
SPST switches may be replaced with power semiconductor switches as illustrated in Figure 5.17. Referring
to the operating modes given in Figures 5.15(c) and (d) and the OFF-state voltages across S1 and S2 given
in Figure 5.16, the switch S2 can be either a BJT or a MOSFET or an IGBT and the switch S1 is a diode
that turns ON when S2 is switched OFF by virtue of the inductor voltage (Ldi/dt) build up. This imple-
mentation is shown in Figure 5.17(a). However, the power semiconductor switches like MOSFETs and
IGBTs are mostly available along with their body diodes in-built; therefore, the switch configuration of
Figure 5.17(b) is also common. Here the IGBT of S2 and the diode of S1 are the operative components of
each switch in this topology.
3. The ESR of the capacitor is zero.
(Vi )DTs + (Vi − Vo )(1 − D )Ts = 0
T1
T1
S1
S1 L
L P C RL Vo
P RL V o
C
Vi Vi S2
S2
T2 T2
(a) (b)
Figure 5.17 Realization of boost converter with power semiconductor switches.
Chapter 05.indd 210 3/19/2009 1:30:20 PM


Vi
Vo = (5.12)
1− D
The duty ratio D, as defined in Eq. (5.1), can take on values between the closed interval 0 and 1. This means
that the output voltage Vo is always greater than the input voltage Vi. Hence this topology is called a step-up
converter or boost converter. Note that Vo is infinite when D is unity. This however is not true. It should be
noted that the above input–output relationship is derived considering steady-state operating conditions.
This means that the switches S1 and S2 are switching and have reached an equilibrium condition in terms
of the ON and OFF times. If D is unity then S2 is always ON and S1 is always OFF. This further implies
that on the input side, the inductor current will rise till the inductor saturates and the input current will be
limited only by the winding resistance and the ON-state impedance of S2. The output capacitor is discon-
nected from the input side as S1 is OFF. This means that the capacitor will continually discharge into the
load RL till it losses all its charge and Vo becomes zero. Thus when D is unity, the output voltage Vo will in
actuality become zero as Eq. (5.12) is no longer valid. To obtain the input–output current relationship one
can use the 100% efficiency constraint. This implies that
Vi I in = Vo I o (5.13)
Io
I in = (5.14)
1− D
Inductor value The value of the inductor L is calculated on the basis of the amount of current ripple ΔiL
that the designer would like to allow for a given application. Referring to the waveforms given in Figure
5.16, the slope of the inductor current waveform during the period DTs is given as
diL ΔiL ΔiL Vi
= = = (5.15)
dt Δt DTs L
diL ΔiL ΔiL V − Vo
= = = i (5.16)
Here Vi and Vo are known from the converter specifications. D can be obtained from the input–output rela-
tionship of Eq. (5.12). The switching frequency fs is a design choice. From this Ts = 1/fs is obtained. ΔiL is
also a design choice. The common choice in most cases is ΔiL = 10% of Iin. The output power Po is known
from the converter specifications. From the knowledge of Po and Vo, Io is calculated. Using Eq. (5.14), Iin
can be estimated for a given load. The only unknown L can then be calculated from either Eq. (5.15) or
Eq. (5.16). From Eq. (5.15), the value of inductor is given as
Vi D
L= (5.17)
ΔiL f s
To regulate Vo with variations in Vi, the value of D should be varied. Thus if Vimax is the maximum input
Chapter 05.indd 211 3/19/2009 1:30:20 PM

Vimax V
Vo = = imin (5.18)
1 − Dmin 1 − Dmax
Here Vo, Vimax and Vimin are known from specifications of the converter. Dmax is a design choice that is
dependent on the specific application. From Eq.(5.18),
⎛V ⎞
Dmin = 1 − ⎜ imax ⎟ (1 − Dmax )
⎝ Vimin ⎠
The inductor value should be calculated with the appropriate duty ratio for a particular input voltage swing.
Thus,
V D
L = imax min (5.19)
ΔiL f s
Capacitor value It is calculated by applying the amp-second rule. Referring to the capacitor current wave-
form given in Figure 5.16, if there is to be no charge build up or charge reduction in the capacitor, the area
under the capacitor current curve in one period should be zero. The area under the positive portion of the
current curve implies charging of the capacitor and that under the negative portion implies discharging of
the capacitor. For charge balance both these areas should be equal. The change is the capacitor charge ΔQ is
given by the area under either the positive portion or the negative portion of the capacitor current curve. For
the boost converter, geometrically it is easier to find the area under the negative portion of the capacitor
current curve. Thus,
IoD
ΔQ = C ΔVo = I o DTs = (5.20)
fs
IoD
C= (5.21)
ΔVo f s
ΔVo is known from the output ripple specification. ΔiL and fs are design choices as mentioned earlier. The
value of C should be greater than the value calculated by Eq. (5.21). Compare this with the capacitor value
estimate for the buck converter as given by Eq. (5.11). In the case of the buck converter, the capacitor cur-
rent is only the inductor ripple current which is only a fraction of the load current whereas in the case of the
boost converter, the capacitor has to handle a larger swing in the current through it. Therefore, the capacitor
size is bigger in the case of the boost converter as compared to the buck converter.
Effect of Non-Idealities The analysis assumed the switches, inductor and capacitor as ideal. However,
practical components are far from ideal. The switches have ON-state drops, the inductor has winding
resistance, the capacitor has the ESR that contributes to output voltage ripple and reduction in efficiency.
As a consequence, the output of the practical boost converter will not exactly agree with the input–output
relationship derived with the ideality constraints. Figure 5.18 shows a boost converter schematic with
few non-idealities included. The ON-state drops of the switches are included as voltage sources in series
with the SPDT switches. The source Vcesat in series with the IGBT switch represents the ON-state drop of
S2 when S2 is ON. The source VD in series with the diode switch represents the ON-state drop of S1
when S1 is ON. The resistance rL represents the inductor winding resistance and resr represents the
capacitor ESR.
Chapter 05.indd 212 3/19/2009 1:30:20 PM

T1
−
VD
+
S1 resr
L rL
P
RL Vo
S2
Vi C
+
Vcesat
−
T2
Figure 5.18 Boost converter with non-idealities included.
amp-second balance to the circuit of Figure 5.18 wherein the non-idealities are included in the schematic.
The design of the capacitor will also get affected due to the presence of the non-idealities especially the ESR
non-ideality. As discussed in the buck converter case, the output voltage ripple will contain contributions
from the voltage variation due ESR and that due to the change of charge. Therefore for a specified output
voltage ripple, the voltage ripple contribution due to change of charge should be much lesser than when
ESR is absent. This would result in a higher value of capacitance than that resulting from Eq. (5.21).
Step-Up/Down or Buck–Boost Converter

Topology
The step-up/down or the buck–boost converter gives an output voltage that can be either higher or lower
than the input voltage depending on the operating duty ratio. The buck–boost converter topology is illus-
trated in Figure 5.19. Figure 5.19(a) shows the buck–boost converter consisting of an SPDT switch, an
inductor that is connected to the pole of the SPDT switch and a capacitor that is connected across the
output. In all three primary converters, the capacitor is always connected across the output. However,
observe that in the case of the buck converter, the inductor is connected between the pole of the SPDT
switch and the output; in the case of the boost converter, the inductor is connected between the pole of the
SPDT switch and the input; and now in this case of the buck–boost converter, the inductor is connected
between the pole of the SPDT switch and the circuit ground. Figure 5.19(b) shows the buck–boost con-
verter with the SPDT switch replaced by two SPST switches S1 and S2. Figure 5.19(c) shows the operative
circuit during the period DTs of a switching period. During this time, S1 is ON and S2 is OFF. The pole P
is connected to the input source Vi. The inductor current flows from the input source to energize the induc-
tor. Figure 5.19(d) shows the operative circuit during the period (1 – D)Ts of the switching period. During
this time, S1 is OFF and S2 is ON. The inductor current starts to decrease. This negative di/dt of the induc-
tor current develops sufficient voltage (Ldi/dt) with a polarity such as to drive the inductor current through
the switch S1 to charge the output capacitor with the polarity as shown. It should be observed that the
output voltage polarity is negative with respect to the circuit ground as shown in Figure 5.19.
Chapter 05.indd 213 3/19/2009 1:30:20 PM

T1 T2 S1 S2 T2
T1
P
P
− −
Vi C RL Vo Vi L C RL Vo
L + +
(a) (b)
DTs period (1 − D )Ts period

T1 S1 P S2 T2 T1 S1 P S2 T2
− −
Vi L C RL Vo Vi L C RL Vo
+ +
iL ic iL ic
Io
(c) (d)
Figure 5.19 Step-up/down or buck–boost converter: (a) Buck–boost converter with a SPDT switch, an
L and a C; (b) SPDT switch is replaced with two switches S1 and S2; (c) operative circuit
during the period DTs; (d) operative circuit during the period (1 – D)Ts.
Operation
The operation of the buck–boost converter is analyzed by visualizing the waveforms of the various signals of
the circuit. The following signals are considered for analysis:
3. load current;
Figure 5.20 shows the plot of the various signals of the buck-boost converter under steady-state operating
conditions.
Chapter 05.indd 214 3/19/2009 1:30:20 PM

S1 S2 S1 S2
ON ON ON ON
DTs (1 − D)Ts
Vi Vi
vL
0
t
Vo
Vo /L
Vi /L Iin + Io
iL
ΔiL
0
Vo /L t
ic
0
t
Io Io
Vi /L Iin Iin + Io
iS1
0
Vo /L Iin + Io t
Io
iS2
0
Vi + Vo Vi + Vo t
VS1
0
t
Vi + Vo Vi + Vo
VS2
0
t
Figure 5.20 Waveforms of various signals of a buck–boost converter.
Chapter 05.indd 215 3/19/2009 1:30:21 PM

DTs Duration The switch S1 is ON and S2 is OFF during the period DTs. The voltage vL across the induc-
tor during this time is Vi which is a DC quantity. From the Faraday’s law, the inductor voltage and current
are related as vL = L(diL /dt ) . This implies that the current through the inductor is given as
1 1 V
L ∫ vL dt = ∫ Vi dt = i ∫ dt
iL =
L L
As Vi is a constant, the inductor current iL rises linearly with a slope of Vi/L as indicated in Figure 5.20.
During this time the current through S1 is the same as the inductor current. The voltage across S2, which is
OFF, is Vi + Vo by applying the Kirchhoff ’s voltage law.
(1 – D)Ts Duration During the time when S1 is OFF and S2 is ON, the voltage vL across the inductor
during this time is Vo. However, Vo is a negative quantity with respect to the circuit ground. The inductor
current iL falls linearly with a slope of Vo/L as indicated in Figure 5.20. During this time, the current through
S2 is the same as the current through the inductor. The voltage across S1, which is OFF, is Vi + Vo. The S1
switch current (iS1) is shown in Figure 5.20. The average or DC component of the S1 switch current is Iin.
The S2 switch current (iS2) is also shown in Figure 5.20. The average or DC component of the S2 switch cur-
rent is Io. The iS2 has two distinct components as shown in Figure 5.20: (a) the average or DC component (Io)
and (b) the AC component (ic). The DC component cannot flow through the capacitor; therefore it flows
into the load RL. The AC component alone will flow through the capacitor. Therefore, the capacitor current
ic is the S2 switch current minus the load DC component. The inductor current is composed of two parts,
viz., (a) the average or the DC component that is the sum of the DC components of S1 and S2 switch cur-
rents (i.e., Iin + Io) and (b) the AC component ΔiL. Here also the inductor current ripple is a design choice.
The boost converter schematic of Figure 5.19 shows the conceptual SPDT and SPST switches. The SPST
switches may be replaced with power semiconductor switches as illustrated in Figure 5.21. Referring to the
operating modes given in Figures 5.19 (c) and (d) and the OFF-state voltages across S1 and S2 given in
Figure 5.20, S1 can be either a BJT or a MOSFET or an IGBT and S2 is a diode that turns ON when S1 is
switched OFF by virtue of the inductor voltage (Ldi/dt) build up and allows the inductor current to free-
wheel by charging the capacitor. This implementation is shown in Figure 5.21(a). However, the power semi-
conductor switches like MOSFETs and IGBTs are available along with their body diodes built-in; therefore,
the switch configuration of Figure 5.21(b) is also common. Here the IGBT of S1 and the diode of S2 are the
operative components of each switch in this topology.
S1 S2
T1 P T2 T1 P T2
S1 S2
− −
Vi C RL Vo Vi C RL Vo
L + L +
(a) (b)
Figure 5.21 Realization of buck–boost converter with power semiconductor switches.
Chapter 05.indd 216 3/19/2009 1:30:21 PM

3. The ESR of the capacitor is zero.
Vi DTs + Vo (1 − D )Ts = 0
−Vi D
Vo = (5.22)
1− D
Note the negative sign in Eq. (5.22). This accounts for Vo being negative with respect to the circuit ground.
The duty ratio D can take on values between the closed interval 0 and 1. If D is between 0 and 0.5, then
the output Vo is less than Vi and results in buck operation. If it is between 0.5 and 1, then Vo is greater than
Vi and results in boost operation. Hence this topology is called a step-up/down converter or buck–boost
converter.
Note that Vo is infinite when D is unity. The same argument as discussed for the boost converter is valid
here also. When D = 0.5, the converter acts as a unit pass circuit wherein Vo = Vi.
To obtain the input–output current relationship one can use the 100% efficiency constraint. This
implies that
Vi I in = Vo I o (5.23)
⎛ D ⎞
I in = I o ⎜ ⎟ (5.24)
⎝ 1− D ⎠
Inductor value The value of the inductor L is calculated based on the amount of current ripple ΔiL that
the designer would like to allow for a given application. Referring to the waveforms given in Figure 5.20, the
slope of the inductor current waveform during the period DTs is given as
diL ΔiL ΔiL Vi
= = = (5.25)
dt Δt DTs L
diL ΔiL ΔiL V

= = = o (5.26)
Chapter 05.indd 217 3/19/2009 1:30:21 PM

Vi and Vo are known from the converter specifications. D can be obtained from the input–output relationship
of Eq. (5.22). The switching frequency fs is a design choice. From this Ts = 1/fs is obtained. ΔiL is also a design
choice. Though this is chosen based on the application, the default choice is ΔiL = 10% of Iin + Io for the
buck–boost converter. The output power Po is known from the converter specifications. From the knowledge
of Po and Vo, Io is calculated. Using Eq. (5.24), Iin can be estimated for a given load. The only unknown L can
then be calculated from either Eq. (5.25) or Eq. (5.26). From Eq. (5.26), the value of inductor is given as
Vo (1 − D )
L= (5.27)
ΔiL f s
To regulate Vo with variations in Vi, the value of D will be varied. Thus if Vimax is the maximum input
Vimax Dmin Vimin Dmax

Vo = = (5.28)
1 − Dmin 1 − Dmax
Vo, Vimax and Vimin are known from specifications of the converter. Dmax is a design choice that is dependent
on the specific application. Then for the buck–boost converter,
Vmin Dmax
Dmin =
Vmin Dmax + Vmax (1 − Dmax )
The inductor value should be calculated for Dmin to ensure that the inductor current ripple is within
specified limits. Thus,
Vo (1 − Dmin )
L= (5.29)
ΔiL f s
Capacitor value The capacitor value is calculated by applying the amp-second rule. Referring to the
capacitor current waveform given in Figure 5.20, if there is to be no charge build up or charge reduction in
the capacitor, the area under the capacitor curve in one period should be zero. The area under the positive
portion of the current curve implies charging of the capacitor and that under the negative portion implies
discharging of the capacitor. For charge balance the area under both should be the same. The change is the
capacitor charge ΔQ is given by the area under either the positive portion or negative portion of the capaci-
tor current curve. For the buck–boost converter as in the case of the boost converter, it is easier to find the
area under the negative portion of the capacitor current curve. Thus,
IoD
ΔQ = C ΔVo = I o DTs = (5.30)
fs
IoD
C= (5.31)
ΔVo f s
Chapter 05.indd 218 3/19/2009 1:30:21 PM

S1 S2
T1 P T2
+ − − +
Vcesat VD
resr
L
Vi RL Vo
−
C
rL +
Figure 5.22 Buck–boost converter with non-idealities included.
ΔVo can be obtained from the output ripple specification. ΔiL and fs are design choices as mentioned earlier.
The value of C should be greater than the value calculated by Eq. (5.31).
Effect of Non-Idealities The analysis assumed that the switches, inductor and capacitor as ideal. How-
ever, practical components are far from ideal. As discussed for the buck and the boost converters, the switches
have ON-state drops, the inductor has winding resistance, the capacitor has the ESR that contributes to
output voltage ripple and reduction in efficiency. As a consequence, the output of the practical buck–boost
converter will not exactly agree with the input–output relationship derived with the ideality constraints.
Figure 5.22 shows a buck–boost converter schematic with few non-idealities included. The ON-state drops
of the switches are included as voltage sources in series with the SPDT switches. The source Vcesat in series
with the IGBT switch represents the ON-state drop of S1 when S1 is ON. The source VD in series with the
diode switch represents the ON-state drop of S2 when S2 is ON. The resistance rL represents the inductor
winding resistance and resr represents the capacitor ESR.
amp-second balance to the circuit of Figure 5.22 wherein the non-idealities are included in the schematic.
The design of the capacitor will also get affected due to the presence of the non-idealities especially the ESR
non-ideality as discussed for the buck and boost converters.
5.5 Selection of Power Switches

Device Selection
The switch element in any converter is selected based on the ideal i–v < static characteristic requirement.
The static i–v characteristic requirement is obtained from the operative converter circuits during the periods
when the particular switch is ON and OFF. During the period when the switch is ON, the current flow
requirement is obtained. During the period when the switch is OFF, the voltage withstanding capa-
bility requirement is obtained. The required idealized i–v characteristics is compared with the idealized i–v
characteristics of the various power switching devices discussed in Chapter 1 and the most appropriate
device type is chosen for the particular SPST switch position in the converter. The following example illus-
trates the selection of S1 and S2 switches.
Chapter 05.indd 219 3/19/2009 1:30:21 PM

EXAMPLE 5.1 Buck Converter S1 and S2 Switch Selection
i T1P ΔiL i T2P ΔiL

I + Io +
o 2 2
S1 ON
−Vi S2 ON
0 Vi VT1P 0 VT2P
S1 OFF S2 OFF
(a) (b)
Figure 5.23 (a) S1 switch i–v requirement; (b) S2 switch i–v requirement
for the buck converter.
Based on the operation of the buck converter and the waveforms shown in Figure
5.12, i–v requirements for S1 and S2 are drawn as shown in Figure 5.23. To be
consistent for all switches and converters, i–v requirements should be plotted for
switch currents flowing from the corresponding throw to the pole when the switch is
ON. For voltage across the switch, the corresponding throw potential with respect
to the pole should be considered. Figure 5.23(a) shows the i–v requirements for the
S1 switch. When S1 is ON, a maximum current of Io + (ΔiL/2) will flow through it
at zero drop across the device. When S1 is OFF, a maximum voltage drop of Vi
occurs across it at zero current through it. If one compares these S1 i–v require-
ments with the idealized static i–v characteristics of devices discussed in Chapter 1,
the BJT, MOSFET and IGBT with or without the internal body diode can handle
the S1 requirement.
Likewise the S2 i–v requirement is shown in Figure 5.23(b). When S2 is ON, a
maximum current of Io + (ΔiL/2) will flow through it at zero drop across the device.
When S2 is OFF, a maximum voltage drop of –Vi occurs at throw T2 with respect
to the pole P. This i–v characteristic can be handled by a diode. Thus a diode is
chosen for the S2 switch of the buck converter. However, in order to reduce the
inventory of the components, the internal body diode of the MOSFET or IGBT
itself may be used to perform the function of the S2 switch as shown in Figure
5.13(b). In such cases, both S1 and S2 will be the same type of power semiconduc-
tor devices; however, S1 will utilize the MOSFET or IGBT portion only and S2
will utilize the internal body diode portion only by giving zero gate drive to the
MOSFET or IGBT portion. The selection of the power device for the boost, buck–
boost or any converter can be done using the above approach.
Chapter 05.indd 220 3/19/2009 1:30:21 PM

Electrical Stress Rating

The electrical stresses that occur in the device are the voltage across the device and the current through the
device. When ON, the current capability of the device should be higher than the maximum current that will
pass through the device by a factor of 1.5 to 2. When OFF, the voltage withstanding capability of the device
should be higher than the maximum voltage that occurs across the device by a factor of 1.5 to 2. The wave-
forms of voltage across the device and the current through the device will decide the voltage and current
rating selection of the device.
Apart from the voltage and current ratings, the device will also encounter stressing during switching.
The voltage across the device will swing from zero to maximum and vice-versa. Likewise the current through
the device will also swing from maximum to zero and vice-versa. The rate of change of the voltage across
switch and/or the current through the device has a bearing on the failure rate of the device. These rates
should be less than the dv/dt and the di/dt ratings of the device, respectively. The switching frequency and
the rise time and fall time required for the voltage across the device and the current through the device will
decide the dv/dt and the di/dt ratings for the device.
Thermal Stress Rating

The operating junction temperature Tj of the power switching device is an important stress on the device
that affects the failure rate of the device. It is directly dependent on the power dissipated within the
device and the rate at which this dissipated power is conducted away to the ambient by means of heat
sinks. The power dissipation in the device is due to (a) ON-state or conduction losses, (b) OFF-state losses
and (c) switching losses. From the device voltage and current waveforms for the converter, the ON-state, the
OFF-state and the switching losses are calculated as discussed for the various devices in Chapter 1. The sum
of all these losses gives the total power dissipation in the device. This dissipated power is conducted away as
heat to the ambient through the junction to case, case to heat sink and sink to ambient thermal resistances
so as to maintain the required operating junction temperature below the maximum allowable junction tem-
perature. The dissipated power, the ambient temperature and the required operating junction temperature
will decide the thermal resistance of heat sink that needs to be selected. Heat sink sizing and selection are
discussed in Chapter 13. The removal of heat from the junction and the thermal dynamics for the design of
heat sinks is also discussed in detail later in Chapter 13.
One should note that all electronic components like the resistor, inductor and capacitor should be
chosen such that they are within the electrical stress limits and also the thermal stress limits.
5.6 Continuous and Discontinuous Conduction Modes
C ontinuous conduction mode (CCM) and discontinuous conduction mode (DCM) are defined with
respect to the inductor current. If the current through the inductor is continuous without becoming
zero at any instant in a switching period, then the converter operation is referred to as the CCM. If the cur-
rent through the inductor becomes zero for some portion of the switching period then the converter opera-
tion is referred to as the DCM. Referring to the inductor current waveforms of the three primary converters,
it can be observed that the ripple portion of the current is a triangular waveform. As the switching period Ts
is very small compared to the time constant associated with L and the load, the ripple slopes are considered
constant. If the ripple slopes are constant, the ΔiL swing of the inductor current during either DTs or (1 –
D)Ts periods is independent of the load Io. However, the average value is dependent on the load Io. Thus for
a converter with input voltage, output voltage and frequency specified, the duty ratio D is determined. In such
Chapter 05.indd 221 3/19/2009 1:30:22 PM

a case ΔiL is of constant amplitude. This constant amplitude ΔiL will, however, swing up or down depending
on the load Io. However, it should be noted that, in a practical converter the output impedance Ro is not zero
but finite. Therefore with change in Io, there will be a change in Vo by an amount equivalent to the drop
across the output impedance Ro. This will cause a change in the slope of the inductor current ripple as it is
dependent on the converter output voltage.
Consider the case of the inductor current of a buck converter. Figure 5.24 shows the inductor currents
for various load conditions. As the load current Io is decreased, observe that the inductor current ripple
amplitude ΔiL does not change; however, the average decreases. At a specific load, Iomin, the bottom of the
inductor current ripple just touches the zero line. This is the boundary of CCM and DCM. If the load is
decreased any further, there will be portions of the switching period when the inductor current will become
zero and the converter is said to be operating in DCM.
Referring to Figure 5.24, at the boundary of CCM/DCM,
ΔiL
I o = I omin = (5.32)
2
The Eq. (5.32) implies that for the converter to operate in CCM, the condition ΔiL / 2 I o < 1 should be
satisfied. Referring to Eq. (5.5), for a buck converter, the inductor current ripple is given as
Vo (1 − D )Ts
ΔiL = (5.32a)
L
Dividing Eq. (5.32a) by 2Io, one obtains
ΔiL ⎛ Vo ⎞ (1 − D )Ts 1− D 1− D
=⎜ ⎟ = =
2I o ⎝ I o ⎠ 2L 2 L / RLTs K
Thus,
ΔiL 1 − D
= (5.32b)
2I o K
DTs (1 − D)Ts
iL
Io1
ΔiL Io2
Iomin
Boundary of CCM / DCM
Figure 5.24 Inductor current waveform for the buck converter with different loads.
Chapter 05.indd 222 3/19/2009 1:30:22 PM

iL
(Vi − Vo) / L −Vo / L

ILm
Io
0 t
D1Ts D2Ts (1 − D1 − D2) Ts
Ts
Figure 5.25 Inductor current waveform in DCM operation for buck converter.
where K 2 L / RLTs and is called the conduction parameter. From Eq. (5.32b), it can be seen that if
K > 1 – D then the buck converter is in CCM operation; if K < 1 – D then the buck converter is in DCM
operation and if K = 1 – D then the buck converter is operating at the boundary of CCM and DCM. The
value of K that results in the buck converter operating in the CCM/DCM boundary is called Kbound. Thus,
Kbound = 1 – D for the buck converter.
To maintain the converter operation in CCM, the load should not be decreased any lower than that
given by Eq. (5.32). If the converter is designed for CCM operation, it should not be operated in DCM and
vice-versa because the input–output relations and design equations will not be the same for both the modes
of operations. Considering the buck converter, if the converter is operating in DCM, then in one switching
period, there are three parts as indicated in Figure 5.25, viz., (a) duration D1Ts when switch S1 is ON
and the inductor current is rising, (b) duration D2Ts when the switch S2 is ON and the inductor current is
freewheeling and (c) duration (1 – D1 – D2)Ts wherein the inductor current is zero. To obtain the input–
output relationship, the volt-second balance rule should be applied to the voltage across the inductor. Thus,
(Vi − Vo )D1Ts + ( −Vo )( D2Ts ) + 0(1 − D1 − D 2 )Ts = 0
Solving the above equation, one obtains

D1
Vo = V i (5.33)
D1 + D2
The form of the Eq. (5.33) obtained above is not very useful for design purposes as there is one equation and
two unknowns, that is, D1 and D2, Vo and Vi being known from the specifications of the converter. Therefore,
the above equation should be further simplified such that there remains only one unknown parameter,
either D1 or D2.
Referring to Figure 5.25, the peak inductor current ILm is used to relate D1 and D2. Thus using the rising
and falling slope information of the inductor current waveform, one has
⎛ V − Vo ⎞ ⎛ Vo ⎞
I Lm = ⎜ i ⎟ D1Ts = ⎜ ⎟ D2Ts (5.34)
⎝ L ⎠ ⎝ L ⎠
Chapter 05.indd 223 3/19/2009 1:30:22 PM

The average value of the inductor current Io is given by

T
1 s Area under triangle 1
Ts ∫0 L
Io = i dt = = I Lm ( D1 + D2 ) (5.35)
Ts 2
From Eqs. (5.34) and (5.35), one obtains
Vo 1 ⎛ V − Vo ⎞
Io = = I Lm ( D1 + D2 ) = ⎜⎜ i ⎟⎟ D1( D1 + D2 )Ts (5.36)
RL 2 ⎝ 2L ⎠
Equation (5.36) can be re-arranged to obtain
⎛ 2 L ⎞ ⎛ Vo ⎞ ⎛ 1 ⎞
D2 = ⎜ ⎟⎜ ⎟ ⎜ ⎟ − D1
⎝ RLTs ⎠ ⎝ Vi − Vo ⎠ ⎝ D1 ⎠
From the definition of the conduction parameter K given in Eq. (5.32b), D2 is now simplified as
⎛ Vo ⎞⎛ 1 ⎞
D2 = K ⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟ − D1 (5.37)
⎝ V i − Vo ⎠ ⎝ D1 ⎠
In Eq. (5.37), Vi and Vo are known from the buck converter specification, the conduction parameter K is a
design choice which gives a measure of the depth of DCM operation. Smaller the value of K, larger is the zero
inductor current portion and vice-versa. Therefore for a given load, Io, larger value of K would imply smaller
peak inductor current ILm. However, one must note that compared to the CCM operation wherein only a
part of the energy in the core of the inductor is transferred to the output, in the DCM operation, the com-
plete inductor energy is transferred to the output. This will reduce the inductor core sizes for a specified
amount of energy transfer every switching cycle. Herein lies the classic trade between inductor peak current
and the inductor core size which dictates the choice of K for a particular application. Considering Eqs. (5.33)
and (5.37), there are two equations and two unknowns. Thereby D1 and D2 can be determined.
Kbound
K < Kbound K > Kbound
1 ∴ DCM ∴ CCM
K = 2L /RLTs
Kbound = (1 − D)
0 Dmin 1 D
Figure 5.26 Illustration of the choice of conduction parameter K.
Chapter 05.indd 224 3/19/2009 1:30:23 PM

As discussed earlier, Kbound is the value of the conduction parameter wherein the converter is at the
boundary of CCM/DCM. The Kbound for the buck converter is shown to be 1 – D. Figure 5.26 shows Kbound
versus D. For a given load, the value of the conduction parameter K is also shown in Figure 5.26. For the
range of D where K > Kbound, the converter operates in CCM and for the range of D where K < Kbound, the
converter operates in DCM. The choice of K determines the range of duty ratio D for which the converter
will operate in CCM and/or DCM. At the boundary of CCM/DCM, K = Kbound and the corresponding
duty cycle is the minimum duty ratio Dmin for CCM operation of the converter.
Similar analysis can be performed for the other converters also. For the boost converter it can be shown
that Kbound is D(1 – D)2 and for the buck–boost converter the Kbound can be shown to be (1 – D)2.
5.7 Isolated Converters

T he three primary converters discussed till now do not provide isolation between the input source and
the output load. However, in practical applications the power supplies and power converters are
designed and built to incorporate isolation between the input source side circuit and the output load side
circuit. Providing isolation has many benefits. Few of the important benefits are the following:
1. Galvanic isolation between source side and load side, that is, no physical connection between source
side and load side. This suppresses all common mode electromagnetic interference (EMI).
2. The turns ratio can be used as a fixed gain for achieving output voltage amplification or attenuation.
This will give the designer more flexibility to choose the range of duty ratio.
3. With isolation, the converter can be designed to have multiple outputs each with different voltages and
grounds.
The isolated converters are topological extensions of the basic converters. Of the three primary converters,
the buck and the buck–boost converters are extended to isolated topologies. The boost converter, however,
does not lend itself well to isolated extensions. Thus the following two categories of isolated converters are
popular:
1. buck-based isolated converters;
2. buck–boost-based isolated converter or flyback converter.
In the buck-based isolated converters there are many topological variants existing in the literature for
different applications. However, the buck–boost-based isolated converter, or the flyback converter as it
is known in the literature, is a direct extension of the buck–boost topology by including an isolating
secondary winding on the same inductor core. The primary winding is connected to the source side cir-
cuit and the secondary winding is connected to the output side circuit. This will be discussed in detail
later.
As there are many variants of the buck-based isolated converters, the schematic shown in Figure 5.27
gives a generic topology of the buck-based isolated converters. The buck-based isolated topologies can be
apportioned into three parts as shown in Figure 5.27: (a) the output buck converter, (b) the input side
chopper and (c) the isolation transformer.
The DC input source Vi cannot be directly fed to the isolation transformer as the transformer will func-
tion only if its primary is an AC voltage. Therefore, a chopper circuit is needed to convert the DC input
source to an AC pulse voltage Vp before feeding to the transformer. Depending on the chopper topology, Vp
may have a peak value of Vi or Vi/2. The secondary of the transformer will produce a voltage of nVp as
shown in the Figure 5.27. The value of the turns ratio n will determine whether nVp will be amplified
Chapter 05.indd 225 3/19/2009 1:30:23 PM

1:n
T1 P
Vi VP nVp C Vo
T2
Chopper HF transformer Buck
Figure 5.27 Generic schematic of buck-based isolated converter.
or attenuated. The voltage nVp is fed as input to the conventional buck converter. If the buck converter is
switching with a duty ratio of D, then the output is given as
Vo = nVp D (5.38)
Based on the input chopper topology, the buck-based isolated converters are classified as follows:
1. Forward Converter: The chopper is a single unidirectional power flow SPDT switch.
2. Push–Pull Converter: The chopper consists of two back-to-back forward converters with the trans-
formers wound on the same core.
3. Half-Bridge Converter: The chopper is a single bi-directional power flow SPDT switch.
4. Full-Bridge Converter: The chopper consists of dual bi-directional power flow SPDT switches.
Both the buck-based converter variants and the buck–boost-based flyback converter will be discussed in the
following sections.
Forward Converter
Topology
The topology of the forward converter is shown in Figure 5.28. As the power flow from the input source to
the output through the transformer is only in one direction (i.e., the forward direction from source to load),
the converter is called the forward converter. It consists of a high-frequency transformer with primary-side
chopper having an SPDT switch with pole Pp. The secondary side of the high-frequency transformer is the
classical buck converter topology having the SPDT switch with pole Ps. The primary-side SPDT switch pole
Pp can either connect to the negative terminal of the input source Vi or to the resistor Rf .
1:n
Rf
Ts1 Ps
Vi
TP2
VP nVp Vo
Pp Ts2
TP1
Chopper HF transformer Buck
Figure 5.28 Forward converter schematic.
Chapter 05.indd 226 3/19/2009 1:30:23 PM

Ss1 Ps
1:n
L
Vp Vs C RL Vo
Vi
Pp Tx
Sp1
(a)
1:n Ps
L
Ss2
Rf Vp Vs RL Vo
Vi C
Sp2 Pp
Tx
(b)
Figure 5.29 (a) Forward converter operative circuit during DTs; (b) operative circuit
during (1 – D)Ts.
The dot polarities of the transformer Tx are as shown in Figure 5.29. The operation of the forward
converter is depicted in Figure 5.29. Figure 5.29(a) gives the operative circuit during the DTs period and
Figure 5.29(b) shows the operative circuit during the (1 – D)Ts period.
DTs Period
During this period the pole Pp of the primary-side SPDT switch is connected to the negative terminal of the
input source Vi. The pole Ps of the secondary side SPDT switch is connected to the secondary winding
through switch Ss1. The voltage across the primary is Vi during this time. The secondary voltage vs is nVi
which is applied to the input of the secondary side buck converter. The power flows from the input source
Vi through the transformer and into the buck converter. During this period, a constant voltage Vi is applied
across the transformer primary. The flux in the core is governed by the equation,
dim dφ
v p = Lm= Np
dt dt
where Lm is the magnetizing inductance of the transformer, im the magnetizing current, Np the primary
number of turns and f the flux in the core. This implies that the flux in the core is given by
Vi
Np ∫
φ= dt
As Vi and Np are constants, f increases linearly with time in this period.
Chapter 05.indd 227 3/19/2009 1:30:24 PM

Vi
Ss1
1:n Ps iL Io
ip is L
Rf
isp2 Ss2 C RL Vo
vp vs = nvp
Sp2
f
Pp Tx
Sp1 isp1
Figure 5.30 Schematic of forward converter implementation.
(1 – D)Ts Period
During this period, the pole Pp is connected to the resistor Rf and the pole Ps is connected to the secondary
side ground through switch Ss2 allowing the inductor L to freewheel. As one of the secondary windings is
open, there is no current flowing through the secondary of the transformer. When the primary side switch
Sp1 is switched OFF, the primary current is interrupted. The flux within the core is a continuous function of
time and it will not become discontinuous. The sudden cut-off of the primary current implies a negative
d im /dt and therefore the voltage across the primary will reverse. This reverse voltage vp that appears across
the primary winding will tend towards infinity as im is cut-off instantaneously with the non-dot side of the
transformer being more positive than the dot side. This high voltage will damage either the transformer
insulation or the switch Sp1. Therefore, the pole Pp is connected to the resistor Rf through the switch Sp2. The
negative voltage across the primary will cause a current to flow through Rf in the direction as shown in
Figure 5.29(b). A negative voltage across the primary implies that dφ / dt is negative. This causes the flux in
the core f to decrease or decay with the time constant as decided by Lm/Rf .
The SPDT switches can be replaced by the power semiconductor switches by knowing the required i–v
characteristics of each of the switches. This is shown in Figure 5.30. When switch Sp1 is ON, the current
through the switch flows from the pole to the throw. The voltage across the switch pole with respect to the
throw is positive when Sp1 is OFF. Therefore, a BJT, a MOSFET or an IGBT can be used for Sp1. When Sp2
is ON, the current through the switch flows from the pole to the throw. The voltage across the switch pole
with respect to the throw is negative when Sp2 is OFF as Tp2 is connected through Rf to Vi and Tp1 is con-
nected to ground. Thus a diode is used for Sp2. By similar arguments Ss1 and Ss2 are also replaced by diodes
as shown based on their i–v requirements.
Chapter 05.indd 228 3/19/2009 1:30:24 PM

Waveforms
The waveforms of the various signals of the forward converter are shown in Figure 5.31. Each switching
period is divided into two parts: (a) the DTs period where the switches Sp1 and Ss1 are ON and the switches
Sp2 and Ss2 are OFF and (b) the (1 – D)Ts period where the switches Sp2 and Ss2 are ON and the switches Sp1
and Ss1 are OFF.
DTs (1 − D)Ts
Sp1 ON Sp2 ON Sp1 ON Sp2 ON

Ss1 ON Ss2 ON Ss1 ON Ss2 ON
VP
Vi Vi
0
t
ImRf
f
Vi /Np τ = Lm /Rf
0
nIo t
isp1 nΔi L
0
t
isp2 Im
τ = Lm /Rf
0
t
Vsp1 Vi Vi
0
t
Vi Vi
Vsp2
0
t
Figure 5.31 Forward converter waveforms.
Chapter 05.indd 229 3/19/2009 1:30:25 PM

DTs Duration
In this period the primary voltage vp is equal to Vi the input source voltage. The secondary voltage vs is equal
to nVi. As vp is a constant, the flux in the transformer core f is a linearly increasing function with slope Vi/Np.
As the pole Pp is connected to ground through switch Sp1, the voltage across Sp2, which is OFF, is Vi. The
primary switch current isp1 is the reflected secondary current. This is same as the reflected inductor current
for this period. However, it should be noted that isp1 also carries the magnetizing current im in addition to
the load reflected component. The switch current isp1 is shown enlarged in Figure 5.32. However, as Im
is generally very small compared to the load, it is usually neglected in design calculations.
(1 – D)Ts Duration
During this period Ss1 is OFF, therefore there is no secondary reflected current in the primary. The primary
consists of only the decaying magnetizing current. The primary voltage vp is decided by the voltage across Rf
which is imRf . Note that the dot of the Tx is negative with respect to the non-dot winding end as the magne-
tizing current freewheels through Rf . The magnetizing current im decays with the time constant of Lm/Rf .
The flux waveform f also decays with the same time constant. The primary switch Sp1 is now OFF and
will have to withstand a maximum voltage of Vi + ImRf by applying Kirchhoff ’s voltage law to the input
source side loop comprising Vi–Rf –Sp1.
Governing Equations
The operation of the output side buck converter has already been discussed earlier. The input–output voltage
relationship is similar to the buck converter input–output relationship except for a factor of n which is the
turns ratio of the transformer. The input–output relationship is given as
Vo = nVi D
where n = Ns/Np. The design of the transformer magnetics will be discussed in detail in Chapter 7. Here the
design equations for the primary side chopper components will only be discussed. The primary side chopper
consists of three components: (a) switch Sp1, (b) switch Sp2 and (c) the freewheeling resistor Rf .
Switch Sp1 Selection

Referring to Figure 5.32, the peak current through switch Sp1, Ipm, is given as
nΔiL
I pm = nI o + + Im (5.39a)
2
where Ipm is the peak primary current when switch Sp1 is about to turn OFF. The peak current rating of the
device, Icm, should satisfy the following inequality:
I cm > I pm (5.39b)
Referring to Figure 5.31, the device peak OFF-state voltage-handling capacity is given as
Vceo > Vimax + I m Rf + VD (5.40a)
where Vimax is the maximum input voltage, VD is the diode drop of switch Sp2 and Im is the maximum magne-
tizing current.
However, it should be noted that Eq. (5.40a) is valid for an ideal transformer that has no leakage induc-
tance reflected at the primary. In the case of a practical transformer, the coupling between the primary and
Chapter 05.indd 230 3/19/2009 1:30:25 PM

DTs (1 – D )Ts
nIo nΔiL
Load
reflected
component
0
t
Im
im
0
t
isp1 Im
nΔiL
0
t
Figure 5.32 Primary switch current isp1 showing the magnetizing and load
reflected components.
the secondary windings is not 100%. There will be a finite leakage inductance as seen from the primary. As a
2
consequence, when Sp1 is ON, the leakage inductance Ls stores an energy of (1 / 2)Lσ I pm when switch Sp1
is about to turn OFF. This energy will also get dissipated in the resistor Rf. Thus, at turn-ON of switch Sp1,
the voltage drop across Rf is I pm Rf rather than I m Rf as in the ideal transformer case. Thus for a practical
transformer with leakage inductance, the IGBT device peak OFF voltage-handling capacity should be
Vceo > Vimax + I pm Rf + VD (5.40b)
where Ipm is as given in Eq. (5.39a). The power dissipation of the switch Sp1 should be calculated by eva-
luating the ON-state and switching losses as discussed in Chapter 1. This calculated power dissipation
should be used to select the heat sink for the device to maintain the device junction temperature below the
maximum specified value.
Switch Sp2 Selection

The switch Sp2 is a diode. When the diode is conducting, the peak current that it will encounter is Im as can be
observed from the waveforms of Figure 5.31. For the diode selection, it is important to estimate the peak, root
mean square (rms) and the average currents flowing through it. The rms and the average currents are required to
estimate the power dissipation in the diode which will facilitate the selection of the heat sinks if required. Thus,
I dm > I pm (5.41)
where Idm is the peak diode current rating. The load reflected portion of the primary current that is stored in
the leakage inductance will decay quickly as the energy stored in the leakage inductance is small. Therefore,
Chapter 05.indd 231 3/19/2009 1:30:25 PM

for calculating the diode average and rms currents only Im can be used. The diode conducts during the
period (1 – D)Ts. The current waveform is an exponential decaying waveform. The average current rating of
the diode should be greater than the average value of the exponentially decaying waveform as shown in
Figure 5.31. However, it is difficult to find the average of the exponentially decaying waveform. Without
loss of generality, the exponentially decaying waveform can be approximated to a triangular waveform that
would enclose the exponentially decaying waveform entirely. For a triangular waveform approximation,
1
I d-avg >
I (1 − D ) (5.42)
2 m
where Id-avg is the average current rating of the diode and
1− D
I d-rms > I m (5.43)
3
where Id-rms is the rms current rating of the diode. When the diode is in the OFF condition, the peak inverse
voltage (PIV) across the diode is given as
PIV > Vi (5.44)
Selection of Resistor Rf
The function of the resistor Rf is to dissipate the magnetic energy so that the flux in the core gets reset to
zero. The energy stored in the inductor at the time of turn-OFF of switch Sp1 is given as
1
ε m = Lm I m2 (5.45)
2
This energy as well as the energy stored in the leakage inductance has to get dissipated in the resistor Rf
before the next switching cycle begins. Thus, in every switching period Ts, an amount of energy em as given
by Eq. (5.45) will be dissipated in resistor Rf . Thus, the power dissipated in the resistor is given as
(1 / 2)Lm I m2 + (1 / 2)Lσ I pm
2
1
PRf = = ( Lm I m2 + Lσ I pm
2
) fs (5.46)
Ts 2
It may be noted from Eq. (5.46) that the power dissipated in the resistor Rf is independent of the value of the
resistor. The resistor value will therefore only affect the time constant of the magnetizing current decay. The
flux and the magnetizing current should decay to zero in the (1 – D)Ts period before Sp1 switches ON again.
As the decay is exponential in nature, one can expect the flux to have decayed to nearly zero in about five time
constants. Therefore, to ensure that the flux decays to zero during the time Sp2 is ON, one should have
Lm
5 < (1 − Dmax )Ts as Lm Lσ
Rf
or
5Lm f s
Rf > (5.47)
1 − Dmax
Another constraint for Rf comes from the voltage withstanding capability of the IGBT switch (i.e., switch
Sp1). From Eq. (5.40b), one obtains
Chapter 05.indd 232 3/19/2009 1:30:25 PM

Vceo − Vimax − VD
Rf < (5.48)
I pm
From Eqs. (5.47) and (5.48) the bounding values for the selection of the resistor value are obtained as
5Lm f s V − Vimax − VD
< Rf < ceo (5.49)
1 − Dmax I pm
In the case of transformer with negligible leakage inductance the above inequality reduces to
5Lm f s V − Vimax − VD
< Rf < ceo
1 − Dmax Im
If the inequality of Eq. (5.49) is not satisfied, then the maximum duty cycle for the converter should be reduced
and/or the Vceo rating of the primary side switch should be increased till the inequality Eq. (5.49) is satisfied.
Forward Converter with Demagnetizing Winding

The forward converter topology of Figure 5.30 has two main drawbacks:
1. The flux resetting is lossy as the magnetizing energy is lost in the resistor Rf .
2. As the flux decays asymptotically to zero and does not become truly zero at the start of the next switch-
ing cycle, there exists a finite average flux which steadily builds up and saturates the core. This has an
upper limit on the operational duty ratio of the converter to avoid flux build up.
To address the drawbacks mentioned above, the lossy flux resetting part of the forward converter of Figure 5.30
is modified by introducing the demagnetizing or the flux resetting winding as shown in Figure 5.33(a). The
primary side chopper that includes the SPDT switch is re-positioned with the pole Pp at the primary side
ground. One of the throws (switch Sp1) is connected to the primary winding and the other throw (switch Sp2)
is connected to the demagnetizing winding as shown in Figure 5.33(a). Observe the dot polarity of the demagne-
tizing winding with respect to the primary winding as shown. It is an important part of the operation of
demagnetizing circuit.
Circuit Operation
The operation of the forward converter is shown in Figures 5.33(b) and (c). During the time DTs, the opera-
tion of the forward converter is exactly same as that of the forward converter of Figure 5.30. During this time
when the switch Sp1 is ON, the voltage across the primary is Vi with the dot end positive. The demagnetizing
winding will have a voltage of Vi ( N d /N p ) with the dot end positive. As Sp2 is OFF, there is no current
flowing in the demagnetizing winding and it is not operative during this period. The difference is in its opera-
tion during the time (1 – D)Ts. During this time, there is no change in the operation of output buck portion
of the forward converter. The switch Sp1 is OFF and the switch Sp2 is ON as shown in Figure 5.33(c). Also,
the primary current is cut-off suddenly when Sp1 is switched OFF. This causes a negative di/dt that will reverse
the primary voltage polarity making the non-dot end positive. Likewise, by induction, the non-dot end of the
demagnetizing winding will be positive. This will cause a current to flow through the demagnetizing winding
against the source voltage Vi as shown in the Figure 5.33(c). The voltage across the demagnetizing
winding will be Vi which will cause the flux in the core to decay linearly with a slope of Vi/Nd.
Chapter 05.indd 233 3/19/2009 1:30:25 PM

To reduce the flux leakage and keep the leakage inductance seen from the primary to a minimum, Nd
is made equal to Np so that both the primary and the demagnetizing windings can be wound together
in a bifilar fashion. However, if Nd = Np, then for the flux that is built up to a maximum of fm during DTs
will decay to zero in exactly the same time when Sp1 is switched OFF and Sp2 is ON. Thus if Nd = Np, then
(1 – D)Ts > DTs to ensure that flux will decay to zero. From this inequality, it can be seen that D < 0.5, that
is, the maximum duty cycle is limited to 50% if Nd = Np.
1:1:n Ps L
Vi Np Nd Ns C RL Vo
Pp
(a)
Vi C RL Vo
Sp1
(b)
Vi C RL Vo
Sp2
(c)
Figure 5.33 (a) Forward converter schematic with flux reset winding or demagnetizing winding;
(b) operative circuit during DTs; (c) operative circuit during (1 – D)Ts.
Chapter 05.indd 234 3/19/2009 1:30:25 PM

Vi
Ss1 Ps
1:1:n
L
Ss2 C RL Vo
isp1
Sp2
Sp1
isp2
Pp
Figure 5.34 Forward converter implementation circuit with demagnetizing winding.
Figure 5.34 shows the circuit schematic of the forward converter with demagnetizing winding. The
SPDT switch is replaced with power semiconductor switches. Sp1 is implemented with IGBT and Sp2 is
implemented with a diode. The governing equations for selecting the two switches are as follows.
Switch Sp1 Selection There is no change in the device peak current rating Icm. It is same as Eq. (5.40a)
and is repeated here for easy reference:
nΔiL
I cm > nI o +
+ Im
2
The device peak OFF-state voltage-handling capacity is given as
Np
Vceo > Vimax + (Vimax + VD ) (5.50)
Nd
where Vimax is the maximum input voltage, VD the diode drop of switch Sp2 and Im the maximum magnetiz-
ing current. Observe that if VD is considered negligible and Np = Nd, then Vceo should be greater than
2Vimax. The power dissipation of the switch Sp1 should be calculated by evaluating the ON-state and switch-
ing losses as discussed in Chapter 1. This calculated power dissipation should be used to select the heat sink
for the device to maintain the device junction temperature below the maximum specified value.
Switch Sp2 Selection The switch Sp2 is a diode. When the diode is conducting, the peak magnetizing cur-
rent that is reflected from the primary winding is I m ( N p / N d ) . The peak, rms and average currents flowing
through the diode are
Np
I dm > I m (5.51)
Nd
Chapter 05.indd 235 3/19/2009 1:30:26 PM

where Idm is the peak diode current rating. The diode conducts during the period (1 – D)Ts. The current
waveform is a triangular waveform. Thus,
1 Np
I d-avg > I (1 − D ) (5.52)
2 Nd m
where Id-avg is the average current rating of the diode and
Np 1− D
I d-rms > Im (5.53)
Nd 3
where Id-rms is the rms current rating of the diode.
When the diode is in the OFF condition, the PIV across the diode is given as
Nd
PIV > Vimax + Vimax (5.54)
Np
Figure 5.35 shows the typical waveforms for the forward converter with demagnetizing winding. The wave-
forms are drawn for Np = Nd. Observe that the flux decays at the rate of Vi/Nd. If the duty ratio D is less
than 0.5, then the flux decays to zero much before (1 – D)Ts time is completed. The voltage across the Sp1
switch is 2Vi till the demagnetizing winding conducts. Once the flux has completely decayed to zero, the
voltage across the primary winding is zero and only Vi appears across Sp1 which is in OFF-state. The diode
current (switch Sp2) is a triangular waveform wherein the magnetizing current decreases at the rate of Vi/Lm.
When the switch Sp2 is OFF during the period DTs, it has to withstand a PIV of 2Vi as indicated.
Dual Switch Forward Converter

The forward converter with demagnetizing winding addressed both the issues of lossy flux reset and true
zero flux reset value. However, there is one problem in this topology that can unduly stress the primary
side switch Sp1 which is the finite leakage inductance as seen from the primary. This issue is illustrated in
Figure 5.36(a) where the leakage inductance between the primary and the demagnetizing winding as seen
from the primary is shown as Ls . The energy stored in the Ls is not coupled to the demagnetizing side.
Therefore, when Sp1 is switched OFF, the energy in Ls will cause a large reverse Lσ (dim /dt ) across the leakage
inductor. As im is cut-off instantaneously, the dim /dt and hence the reverse voltage drop across the leakage
inductor will be very large. This drop will appear across Sp1 when it is being turned OFF and appear as the
electrical stress on the device. This stress may damage the IGBT (switch Sp1). Even if Np is made equal to Nd
and both the windings are wound in bifilar fashion, there will be a finite leakage inductance as seen from the
primary.
Figure 5.36(b) provides a solution to this problem. Instead of using a separate demagnetizing winding
to do the function of flux resetting, the same primary winding is used to do the function of flux resetting. In
this way there is no issue of magnetic coupling. The primary winding consists of two SPDT switches, one at
each end of the winding. The SPDT switches comprise IGBTs and diodes as shown in Figure 5.36(b).
During the time DTs, switches S1 and S2 are ON allowing power flow to the output buck converter
module. At this time, the voltage across the primary is Vi with the dot end positive. During this time, the
primary current flows from the input source through the path Vi–S1–A–B–S2–ground. During the time
(1 – D)Ts, the switches S1 and S2 are OFF and the diodes D1 and D2 turn ON by virtue of the voltage
reversal across the primary winding and leakage inductance Ls in order to drive the magnetizing current
through the diodes against the input voltage source Vi. At this time, the magnetizing current freewheels
Chapter 05.indd 236 3/19/2009 1:30:26 PM

DTs (1 − D)Ts
S p1 S p2 ON S p1 ON S p2 ON
ON
S s1 ON S s2 ON S s1 ON S s2 ON
vp
Vi Vi
0
t
−Vi −Vi
f
Vi / Nd
Vi / Np
0
t
isp1 nΔiL
nIo
0
Vi / Lm t
isp2 Im
0
t
Vsp1 2Vi 2Vi
Vi Vi
0
t
2Vi 2Vi
Vsp2
0
t
Figure 5.35 Forward converter waveforms with demagnetizing winding.
through the path ground–D1–A–B–D2–Vi. Thus, during both the periods the current direction in the pri-
mary winding is maintained in the direction A to B. Observe that even the energy stored in the leakage
inductance, as seen from the primary, freewheels through the diodes D1 and D2. As a consequence, the volt-
age spikes stress on the primary switches S1 and S2 is absent. During the time when S1 and S2 are OFF, D1
and D2 are conducting resulting in voltage across the OFF-state switches to be Vi. Likewise when S1 and S2
are ON, D1 and D2 will have to withstand voltage of Vi. Thus the voltage rating of the switches can be lower
than that for the single switch forward converter topologies. All other governing equations for the switches
are the same as discussed for the forward converter with demagnetizing winding.
Chapter 05.indd 237 3/19/2009 1:30:26 PM

Vi
Vi
Buck Vo
converter D2
S1
Tx
Lσ A
Buck Vo
VP converter
B
Lσ
D1 S2
(a) (b)
Figure 5.36 (a) Leakage inductance problem; (b) dual switch forward converter.
Push–Pull Converter
The forward converter utilizes only the positive half of the core magnetization as the magnetizing current
and the core flux are unidirectional. As a consequence the core is under-utilized and the size of the core
for a given output power is larger than if core magnetization had been done in both directions. The push–
pull converter magnetizes the core in both directions to better utilize the core. It is so named because
it consists of two forward converters operating back to back. During the period DTs, one forward con-
verter is operative in transferring power to the load and during the period (1 – D)Ts the other forward
converter is operative in transferring power to the load. The schematic of the push–pull converter is
shown in Figure 5.37.
It consists of a primary side single pole triple pole switch with pole Pp as shown. The transformer is
center-tapped both in the primary and the secondary. The secondary side consists of the buck converter. The
operating modes of the circuit are shown in Figure 5.38. Each part of the center-tapped winding behaves as
a simple forward converter that comes into operation on alternate cycles. This is illustrated in Figure 5.38.
Figure 5.38(a) is the operative circuit when the pole Pp is connected to Tp1, that is, switch Sp1 is ON as
shown. The section Vi–C1–B1–Sp1 acts as the primary portion of the forward converter. During this time
the dot end is positive. Due to magnetic coupling, all the dot ends are positive with respect to their non-dot
ends. As a consequence diode D1 is forward-biased and diode D2 is reverse-biased. The A2–C2 winding is the
active winding during this period. It supplies the power to the buck converter module.
Figure 5.38(b) is the operative circuit when the pole Pp is connected to Tp2, that is, switch Sp2 is ON
as shown. The section Vi–C1–A1–Sp2 acts as the primary portion of the forward converter now. During this
time the non-dot ends are positive with respect to their dot ends. As a consequence, D2 is forward-biased
and D1 is reverse-biased. The B2–C2 winding is the active winding which delivers power to the output
during this time.
Chapter 05.indd 238 3/19/2009 1:30:26 PM

A1 A2
D1 L
Vp Np Ns Vs C RL Vo
T p2
T p3 C1 C2
Pp
T p1 Vi
Vp Np Ns Vs
Tx D2
B1 B2
Np : Ns :: 1 : n
Figure 5.37 The push–pull converter schematic.
Figure 5.38 shows the third operative mode of the push–pull circuit. Here the pole Pp is connected to
the primary circuit ground as shown. Both switches Sp1 and Sp2 are OFF. During this time the inductor
current freewheels by sharing the current between D1 and D2. Thus in the secondary, there is a current leav-
ing the dot at A2 and an equal current leaving the non-dot end at B2. These two equal and opposite flux
producing components will make df/dt in the core zero and thus the voltages across all windings are zero.
The waveforms for the various parameters of the push–pull circuit are shown in Figure 5.39. During the
first DTs period, Sp1 is ON and Sp2 is OFF. The voltage across the primary winding A1 to B1 is 2Vi and that
across the secondary winding between A2 and B2 is 2nVi with the dot ends being positive in all windings.
The flux in the core increases linearly with a rate of Vi/Np from –fm to +fm. The diode D1 carries the induc-
tor current during this time. During this time the voltage across the OFF switch Sp2 is 2Vi.
During the (1 – D)Ts period, both the primary switches Sp1 and Sp2 are OFF. As explained before, the
df/dt is zero during this time. Thus f is a constant in this time period. The voltage across the primary side
OFF switches Sp1 and Sp2 is Vi as there are zero voltages across the windings. Observe also that the inductor
current freewheels by equally sharing its current between the secondary diodes D1 and D2.
During the next DTs period, Sp2 is switched ON and Sp1 is OFF. This makes the dot ends of the wind-
ings negative with respect to their non-dot ends, thereby reversing the flux to swing from +fm to –fm with
a rate of –Vi/Np. Every alternate cycle Sp2 is switched ON to reverse the flux. It can be observed that the duty
ratio D is defined with respect to the waveforms of the inductor current ripple. The charging up of the
inductor energy is defined as the DTs period and the freewheeling portion of the inductor current is defined
as the (1 – D)Ts period. However, the switches Sp1 and Sp2 are switched every alternate cycles and therefore,
the switching frequency of the primary side switches is half the frequency of the output buck converter as
can be seen from the waveforms of Figure 5.39.
The circuit schematic of the push–pull converter is shown in Figure 5.40 wherein the SPTT switch is
replaced by two IGBTs with internal body diodes. The governing equations are:
Vo = nVi D (5.55)
Chapter 05.indd 239 3/19/2009 1:30:27 PM

A1 f A2 D1 D1
A1 f A2
+ + L L
− −
C C
− − + +
C1 C2 Sp2
C1 C2
Sp1 Vi + +
Vi − −
− − + +
B1Tx B2 D2 B1 TxB2 D2
(a) (b)
D1
A1 f A 2
L
0V 0 C
C1 C2
Pp V
i
0V 0
B1 Tx B2
D2
(c)
Figure 5.38 (a) Operative circuit during DTs (Sp1 ON); (b) operative circuit during DTs (Sp2
ON) of alternate cycle; (c) operative circuit during (1 – D)Ts when both primary
switches are OFF (i.e., Sp1 and Sp2 are OFF).
where n is the secondary to primary turns ratio and D is the duty ratio as defined for the buck converter
inductor current waveform.
Primary Side Switch Selection

Referring to the waveforms of Figure 5.39, the current rating for the primary side switches Sp1 and Sp2 is
nΔiL
I cm > nI o + (5.56)
2
where Icm is the peak current requirement of the IGBT. Referring to Figure 5.39 again, the peak OFF-state
voltage that the primary side switches should withstand is
Vceo > 2Vimax (5.57)
Chapter 05.indd 240 3/19/2009 1:30:27 PM

DTs (1 − D)Ts DTs (1 − D)Ts DTs (1 − D)Ts

Sp1 ON Sp1 & Sp2 Sp1 OFF Sp1 & Sp2 Sp1 ON Sp1 & Sp2
Sp2 OFF OFF Sp2 ON OFF Sp2 OFF OFF
2Vi
vA1B1
0
t
−2Vi
Vi /Np fm
f
−Vi / Np
0
t
Io −fm
iL
0 Io Io /2 t
iD1
0
Io Io/2 t
iD2
0
nIo t
isp1
0
nIo t
isp2
0
2Vi 2Vi t
Vsp1 Vi Vi Vi
0
t
2Vi 2Vi
Vi Vi Vi
Vsp2
0
t
Figure 5.39 Waveforms for the push–pull circuit.
Chapter 05.indd 241 3/19/2009 1:30:27 PM

S p2
D1
L
C RL Vo
Np Ns
Vi Np Ns
1:n
D2
S p1 N
n= s
Np
Figure 5.40 Circuit schematic of the push–pull converter with power semi-
conductor devices.
Secondary Side Switch Selection

The peak current requirement for the secondary side diodes, D1 and D2, is
ΔiL
I dm > I o + (5.58)
2
The average current requirement for D1 and D2 is
⎛ DTs ⎞ I o ⎛ (1 − D )Ts ⎞ Io
I d-av = I o ⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ = (5.59)
⎝ 2Ts ⎠ 2 ⎝ Ts ⎠ 2
The rms current requirement for D1 and D2 is
⎛ DTs ⎞ I o2 ⎛ (1 − D )Ts ⎞ D 1− D
I d-rms = I o2 ⎜⎜ ⎟⎟ + ⎜ ⎟⎟ = I o + (5.60)
⎝ 2Ts ⎠ 4 ⎜⎝ Ts ⎠ 2 4
The PIV requirements for the diodes D1 and D2 is

PIV > 2nVimax (5.61)
One should note that the push–pull circuit is not a robust topology in terms of circuit operation. The flux
in the core swings between –fm and +fm and remains in this band only if the voltage across the primary has
an average value of zero. However, in a practical circuit Sp1 and Sp2 do not have identical ON-state drops.
Let Vcesat1 and Vcesat2 be the ON-state drops of Sp1 and Sp2, respectively. Then the voltage across the primary
when Sp1 is ON is given by Vi – Vcesat1 and the voltage across the primary when Sp2 is ON is determined by
Vi – Vcesat2. As the pulse duration is DTs for both the Sp1 and Sp2 ON periods, there will be a finite non-zero
average value across the primary voltage as the positive pulse and the negative pulse amplitudes are different.
This finite average voltage value across the primary will cause the flux in the core to drift towards +∞ or −∞
due to the relation
Chapter 05.indd 242 3/19/2009 1:30:27 PM

1
Np ∫ p
φ= v dt
This drifting flux will ultimately saturate the core. This phenomenon is called flux walking. To solve this
problem, additional circuitry is required to sense the flux walking and appropriately adjust the duty ratio so
as to reset the primary average voltage to zero. This will arrest the flux drift accordingly. However, this will
increase the cost and lower the reliability of the circuit.
There is also another problem with the push–pull circuit. The coupling between the primary and the
secondary is not 100%. Therefore, there will be a finite leakage inductance component as seen from the
primary. The energy stored in the leakage inductance will cause large voltage spikes to occur across the OFF-
state switch. The energy stored in the leakage inductances can be removed by providing alternate paths for
them through circuits called the snubber circuits as discussed in Chapter 2. Due to the above drawbacks,
this topology is not a very popular one.
Half-Bridge Converter
For a given power rating, the push–pull circuit requires devices that should have OFF-state voltage with-
standing capability that is greater than twice the input voltage Vi. The half-bridge converter reduces the
OFF-state voltage requirement of the primary side switches to Vi apart from maintaining the bi-directional
flux swing in the core. Thus the voltage stress and cost of the power switches is significantly reduced as com-
pared to the push–pull topology. The schematic of the half-bridge topology is shown in Figure 5.41.
There is no change in the secondary side circuit. It is exactly same as the secondary side circuit of the
push–pull converter. On the primary side, one end of the primary winding is connected to a SPDT switch
with pole Pp as shown. The other end is connected to the center of a capacitor divider as shown. The voltage
at this capacitor divider point is at Vi/2.
The operation of the half-bridge converter circuit is illustrated in Figure 5.42. Here again, there are three
operative modes like in the case of the push–pull converter circuit. The primary switches Sp1 and Sp2 switch
during alternate cycles and their switching frequency is half the output inductor current ripple frequency.
Figure 5.42(a) shows the operative circuit when Sp1 is ON. This occurs during the first DTs period.
The input source voltage Vi is connected to one end of the primary winding through switch Sp1. The other
end is at Vi/2. The voltage across the primary is Vi/2 with the dot end being positive with respect to the
Vi
D1
Pp f
L
Ns RL Vo
C
+
Vi / 2 Vp Np
−
Ns
+
Vi / 2
−
D2
Figure 5.41 Half-bridge converter schematic.
Chapter 05.indd 243 3/19/2009 1:30:28 PM

Vi
D1
f
Sp1 L
C RL Vo
Vi / 2
D2
(a)
Vi
D1
f
L
Sp2 C RL Vo
Vi / 2
D2
(b)
Vi
D1
f
L
0 C RL Vo
0V
Vi / 2
D2
(c)
Figure 5.42 Half-bridge topology: (a) Operative circuit during DTs (switch Sp1 ON); (b) operative
circuit during DTs (switch Sp2 ON); (c) operative circuit during (1 – D)Ts (switches Sp1
and Sp2 OFF).
Chapter 05.indd 244 3/19/2009 1:30:28 PM

non-dot end. The diode D1 is ON and D2 is OFF. The energy is supplied to charge up the inductor
through diode D1.
Figure 5.42(b) shows the operative circuit when Sp2 is ON. This occurs during the second DTs period.
Sp1 and Sp2 will be ON during alternate DTs periods with D and Ts being defined with respect to the induc-
tor current ripple. During this time, Sp2 is ON and the voltage across the primary winding is Vi/2 but the
dot end is negative with respect to the non-dot end. This will facilitate the core flux to swing in the negative
direction. In the secondary, D2 is ON and D1 is reverse-biased. The inductor energy is replenished through
diode D2.
Figure 5.42(c) shows the operative circuit when both Sp1 and Sp2 are OFF. During this time, the induc-
tor current freewheels through both D1 and D2. The currents flowing in the center-tapped secondary are in
directions that will cancel the flux in the core due to each other. This will result in df/dt being zero and as a
consequence the voltages across all the windings of the transformer being zero.
The circuit schematic of the half-bridge circuit with IGBT switches is shown in Figure 5.43. Switches
Sp1 and Sp2 are replaced with IGBT switches with internal body diodes. C1 and C2 are two equal capaci-
tances that form the capacitive divider. The resistances R1 and R2 are charge equalization resistors which will
balance the charges on the capacitors C1 and C2 such that the center point is at Vi/2. One more component,
the flux walking capacitor (Cfw) is included in series with the winding here to prevent the flux walking
phenomenon. Here too, Sp1 and Sp2 may have non-identical ON-state drops. This will result in a finite non-
zero average value across the primary winding causing the flux to walk away and saturate the core. If a
capacitor Cfw is included in series with the winding, it will block the DC or average voltage that results due
to non-identical ON-state drops. This will ensure that the primary winding will always have zero average
voltage across it and hence flux walking phenomenon is absent.
Another advantage over the push–pull topology is that the body diodes are active here in removing energy
stored in the leakage inductances. If Sp1 is being turned OFF, the energy stored in the primary leakage induc-
tance flows through the winding–C1–Vi–diode of Sp2–winding and winding–C2–diode of Sp2–winding paths. If
Sp2 is being turned OFF, the energy stored in the primary leakage inductance flows through the winding–
diode of Sp1–Vi–C2–winding and winding–diode of Sp1–C1–winding paths.
Vi
D1
Cfw 1:n
Sp1 C1 R1 f L
C RL Vo
Ns
Vp Np
Sp2 Ns
C2 R2 Tx
Ns D2
n=
Np
Figure 5.43 Half-bridge circuit implementation.
Chapter 05.indd 245 3/19/2009 1:30:28 PM

The waveforms for the various signals of the half-bridge circuit are shown in Figure 5.44. The voltage
across the primary is Vi/2 during the period when Sp1 is ON and it is –Vi/2 during the time when Sp2 is ON.
The slopes for the core flux are Vi/2Npand –Vi/2Np as shown in Figure 5.44. All other waveforms are similar
to that discussed for the push–pull topology.
The governing equations for the half-bridge converter are
Vi
Vo = n D (5.62)
2
where n is the secondary center tap to primary turns ratio and D is the duty ratio as defined for the buck
converter inductor current waveform.
Primary-Side Switch Selection

Referring to the waveforms of Figure 5.44, the current rating for the primary side switches Sp1 and Sp2 is
nΔiL
I cm > nI o + (5.63)
2
Vceo > Vimax (5.64)
Secondary-Side Switch Selection

ΔiL
I dm > I o + (5.65)
2
⎛ DTs ⎞ I o ⎛ (1 − D )Ts ⎞ I o
I d-av = I o ⎜ ⎟+ ⎜ ⎟= (5.66)
⎝ 2Ts ⎠ 2 ⎝ Ts ⎠ 2
⎛ DTs ⎞ I o2 ⎛ (1 − D )Ts ⎞ D 1− D
I d-rms = I o2 ⎜ ⎟+ ⎜ ⎟ = Io + (5.67)
⎝ 2Ts ⎠ 4 ⎝ Ts ⎠ 2 4
The PIV requirements for the diodes D1 and D2 is
PIV > nVimax (5.68)
The capacitance dividers C1 and C2 are actually the output capacitors of the diode–capacitor filter circuit.
The design and selection of C1 and C2 is discussed in Chapter 3. R1 and R2 are charge equalization resistors.
An empirical rule is to choose R1 and R2 such that they allow around 2% of the load current to flow through
them. Thus,
Vi
R1 = R2 =
2(0.02 × I o )
Chapter 05.indd 246 3/19/2009 1:30:29 PM

DTS (1 − D)TS DTS (1 − D)TS DTS (1 − D)TS

Sp1 ON Sp1 & Sp2 Sp2 ON Sp1 & Sp2 Sp1 ON Sp1 & Sp2
VP Sp2 OFF OFF Sp1 OFF OFF Sp2 OFF OFF
Vi / 2
0 t
f Vi / 2 fm − Vi / 2
− Vi / 2
NP
NP
0
t
Io
IL − fm
0
Io t
Io / 2
iD1
0
Io t
Io / 2
iD2
0
isp1 t
nIo
0
isp2 nIo t
0
Vsp1 t
Vi Vi
Vi / 2 Vi / 2 Vi / 2
0
t
Vi Vi
Vsp2
Vi / 2 Vi / 2 Vi/ 2
0
t
Figure 5.44 Waveforms of the various signals of the half-bridge circuit.
Chapter 05.indd 247 3/19/2009 1:30:29 PM

Flux Walking Capacitor

Referring to the Figure 5.44, the charge added to the capacitor during the time DTs when the switch Sp1 is
ON is
ΔQ = C fw ΔVfw = nI o DTs
This gives
nI o D
C fw > (5.69)
ΔVfw f s
where D and fs are the duty ratio and the switching frequency respectively as defined for the output side
inductor current ripple. ΔVfw is a design choice and it is chosen as 10% of Vi. There is another constraint
that the flux walking capacitor should satisfy. The resonant frequency of the reflected secondary side induc-
tor and the flux walking capacitor should be very much lower than the switching frequency otherwise the
waveforms of the various signals will be oscillatory within a switching cycle. Thus,
2
⎛ Np ⎞
Lp = L ⎜ ⎟
⎜ Ns ⎟
⎝ ⎠
where Lp is the secondary inductance as seen from the primary and L is the secondary side buck converter filter
inductance. The resonant frequency of the LpCfw should be less than fs by a factor of 10. Thus,
1 fs
<
2π LpC fw 10
This gives the following inequality:

100
C fw > (5.70)
4π 2 f s2 Lp
Both the inequalities [Eqs. (5.69) and (5.70)] should be satisfied while selecting the capacitor value. The rms
value of the current through the flux walking capacitor is given as
I Cfw-rms = nI o D (5.71)
Full-Bridge Converter
The half-bridge circuit applies a voltage of Vi/2 at the primary winding whereas a full-bridge circuit applies
a voltage of Vi at the primary winding. Thus for a given power, the average current flowing through the
primary of the full-bridge converter is half of that flowing through the half-bridge converter. Thus
the current-carrying capacity for the full-bridge converter switches is less than that of the switches of the
half-bridge converter of the same output power rating. All other operational features of the full-bridge
converter are same as those of the half-bridge converter. The schematic of the full-bridge topology is shown
in Figure 5.45.
There is no change in the secondary side circuit. It is exactly same as the secondary side circuit of the
half-bridge converter. On the primary side, one end of the primary winding is connected to an SPDT
switch with pole Pp1 as shown. The other end is connected to another SPDT switch with pole Pp2 as
shown.
Chapter 05.indd 248 3/19/2009 1:30:29 PM

Vi
D1
Pp1 f
Ns L
C RL Vo
Vp Np
Ns
Pp2
D2
Figure 5.45 Full-bridge converter topology.
The operation of the full-bridge converter circuit is illustrated in Figure 5.46. Here again, there are three
operative modes like in the case of the push–pull and the half-bridge converter circuits. The primary switches
Sp1 and Sp4 are linked and they switch ON and OFF simultaneously. Likewise the primary switches Sp2 and
Sp3 are linked and they switch ON and OFF simultaneously. The pairs Sp1,4 and Sp2,3 switch ON during
alternate cycles and their switching frequency is half the output inductor current ripple frequency as dis-
cussed for the push–pull and half-bridge converter circuits.
Figure 5.46(a) shows the operative circuit when Sp1,4 are ON. This occurs during the first DTs period.
The input source voltage Vi is connected to one end of the primary winding through switch Sp1. The other
end is connected to ground through Sp4. The voltage across the primary is Vi with the dot end being positive
with respect to the non-dot end. The diode D1 is ON and D2 is OFF. The energy is supplied to charge up
the inductor through diode D1.
Figure 5.46(b) shows the operative circuit when Sp2,3 are ON. This occurs during the second DTs
period. During this time, Sp2 and Sp3 are ON and the voltage across the primary winding is Vi but the dot
end is negative with respect to the non-dot end. This will facilitate the core flux to swing in the negative
direction. In the secondary, D2 is ON and D1 is reverse-biased. The inductor energy is replenished through
diode D2.
Figure 5.46(c) shows the operative circuit when Sp1,4 and Sp2,3 are OFF. During this time, the inductor
current freewheels through both D1 and D2. The currents flowing in the center-tapped secondary are in
directions that will cancel the flux in the core due to each other. This will result in df/dt being zero and as a
consequence the voltages across all the windings of the transformer are zero.
The circuit schematic of the full-bridge circuit with IGBT switches is shown in Figure 5.47. The switches
Sp1 to Sp4 are replaced with IGBT switches with internal body diodes. Cfw is included in series with the
winding here. This is called the flux walking capacitor that will prevent the flux walking phenomenon in a
manner similar to that discussed for the half-bridge circuit.
Here the body diodes are active in removing energy stored in the leakage inductances as seen from the
primary. If Sp1,4 is being turned OFF, the energy stored in the primary leakage inductance flows through the
winding–Sp3 diode–Vi–Sp2 diode–winding path. If Sp2,3 is being turned OFF, the energy stored in the primary
leakage inductance flows through the winding–Sp1 diode–Vi–Sp4 diode–winding path.
The waveforms for the various signals of the full-bridge circuit are shown in Figure 5.48. The voltage
across the primary is Vi during the period when Sp1,4 are ON and it is –Vi during the time when Sp2,3
Chapter 05.indd 249 3/19/2009 1:30:29 PM

Vi
D1
Sp1 L
C
Sp4
D2
(a)
Vi
D1 L
Sp2
C
Sp3
D2
(b)
Vi
D1 L
0 C RL Vo
0V
D2
(c)
Figure 5.46 Full-bridge topology: (a) Operative circuit during DTs (switch Sp1,4 ON); (b) operative circuit
during DTs (switch Sp2,3 ON); (c) operative circuit during (1 – D)Ts (switches Sp1,4 and Sp2,3
are OFF).
Chapter 05.indd 250 3/19/2009 1:30:29 PM

Vi
D1
Cfw L
S p1 S p3 f
C RL Vo
Vp
S p2 S p4
Tx
D2
Figure 5.47 Full-bridge circuit implementation.
are ON. The slopes for the core flux are Vi/Np and –Vi/Np as shown in Figure 5.48. All other waveforms are
similar to that discussed for the half-bridge topology.
The governing equations for the full-bridge converter are
Vo = nVi D (5.72)
where n is the secondary center tap to primary turns ratio and D is the duty ratio as defined for the buck
converter inductor current waveform.
Primary Side Switch Selection

Referring to the waveforms of Figure 5.48, the current rating for the primary side switches Sp1 to Sp4 is
nΔiL
I cm > nI o + (5.73)
2
Vceo > Vimax (5.74)
Secondary Side Switch Selection

ΔiL
I dm > I o + (5.75)
2
⎛ DTs ⎞ I o ⎛ (1 − D )Ts ⎞ I o
I d-av = I o ⎜ ⎟+ ⎜ ⎟= (5.76)
⎝ 2Ts ⎠ 2 ⎝ Ts ⎠ 2
Chapter 05.indd 251 3/19/2009 1:30:29 PM

DTs (1 − D)Ts DTs (1 − D)Ts DTs (1 − D)Ts

Sp1,4 ON Sp1,4 / Sp2,3 Sp2,3 ON Sp1,4 / Sp2,3 Sp1,4 ON Sp1,4 / Sp2,3
Sp2,3 OFF OFF Sp1,4 OFF OFF Sp2,3 OFF OFF
Vi
Vp
0
t
−Vi
fm
Vi /NP
−Vi /NP
f
0
t
−fm
iL Io
0
Io t
Io /2
iD1
0
Io Io /2 t
iD2
0
nIo t
isp1,4
0
nIo t
isp2,3
0
Vi Vi t
Vsp1,4 Vi /2 Vi /2 Vi /2
0
Vi Vi t
Vi/2 Vi /2 Vi /2
Vsp2,3
0
t
Figure 5.48 Waveforms for the various signals of the full-bridge converter.
Chapter 05.indd 252 3/19/2009 1:30:30 PM

⎛ DTs ⎞ I o2 ⎛ (1 − D )Ts ⎞ D 1− D
I d-rms = I o2 ⎜ ⎟+ ⎜ ⎟ = Io + (5.77)
2T
⎝ s ⎠ 4 ⎝ Ts ⎠ 2 4
The PIV requirement for the diodes D1 and D2 is

PIV > 2nVimax (5.78)
Flux Walking Capacitor

The design of the flux walking capacitor is same as that discussed for the half-bridge converter circuit. The
capacitance value of the flux walking capacitor should be chosen based on
nI o D
C fw > (5.79)
ΔVfw f s
where D and fs are the duty ratio and the switching frequency, respectively, as defined for the output side
inductor current ripple. ΔVfw is a design choice and it is chosen as 10% of Vi.
100
C fw > (5.80)
4π 2 f s2 Lp
where Lp is the secondary inductance as seen from the primary given by
2
⎛ Np ⎞
Lp = L ⎜ ⎟
⎜ Ns ⎟
⎝ ⎠
Both the inequalities [i.e., Eqs. (5.79) and (5.80)] should be satisfied while selecting the capacitor value.
The rms value of the current through the flux walking capacitor is given by
I Cfw-rms = nI o D (5.81)
Flyback Converter
Flyback converter is an isolated converter that is based on the primary buck–boost converter topology. This
converter has the least component count and hence is a very popular topology. Figure 5.49 shows the develop-
ment of the flyback converter topology. Figure 5.49(a) shows the primary buck–boost topology that is dis-
cussed earlier section. Figure 5.49(b) shows the same converter with the SPDT switch shifted to the ground
side. The inductor is designed like in the non-isolated buck–boost converter case. However, one more wind-
ing called the secondary is wound on top of the existing inductor winding to provide galvanic isolation. The
SPDT switch is split into two SPST switches, each being connected to one winding of the two winding
inductor. This is shown in Figure 5.49(c). Figure 5.49(d) shows the secondary side modified so that the
output voltage is measured in the conventional positive sense. Observe the dot polarities on the windings
implying the out of phase switching of Sp1 and Ss1.
Figure 5.50 illustrates the operation of the flyback converter. There are two operating states for the con-
verter. One is the period when the inductor energy builds up or the DTs period and the other is the period
when the inductor energy discharges to the load or the (1 – D)Ts period. Figure 5.50(a) shows the operative
circuit during the DTs period. During this time, the switch Sp1 is ON and the switch Ss1 in the secondary
side is OFF. The dot ends of the windings are positive. The inductor energy builds up as it is now directly
Chapter 05.indd 253 3/19/2009 1:30:30 PM

Vi Vo Vi L C Vo
L C
(a) (b)
Ss1
1:n
1:n
+
L C Vo
−
L − Vi
Vi C Vo
+
Sp1
Ss1
Sp1
(c) (d)
Figure 5.49 Development of the flyback converter topology from the primary buck–boost converter.
connected across the input source Vi. At this same time, the capacitor discharges into the load. During this
period there is transfer of energy from the primary to the secondary.
Figure 5.50(b) shows the operative circuit during the (1 – D)Ts period. During this time, the switch Sp1
is OFF and the switch Ss1 is ON. The winding voltages reverse polarity and the non-dot ends become
positive with respect to the dot ends. The primary winding of the inductor is out of action. The secondary
winding now discharges the energy stored in the core to the capacitor and the load.
The SPST switches are replaced by power semiconductor switches in Figure 5.51. The switch Sp1 is replaced
by an IGBT and the switch Ss1 is replaced by the diode depending on the i–v characteristic requirements of the
switches. The waveforms at various points on the flyback converter circuit are shown in Figure 5.52.
1:n 1:n
C C
Vi vp Vo Vi vp Vo
L
L
Sp1 Sp1
(a) (b)
Figure 5.50 Flyback converter: (a) Operative circuit during DTs period; (b) operative circuit during
(1 – D)Ts period.
Chapter 05.indd 254 3/19/2009 1:30:30 PM

Vi Ss1
1:n
Vp Np Ns C RL Vo
Sp1
Figure 5.51 Schematic of the flyback topology implementation.
DTs (1 − D)Ts DTs (1 − D)Ts

Sp1 ON Ss1 ON Sp1 ON Ss1 ON
Ss1 OFF Sp1 OFF Ss1 OFF Sp1 OFF
Vp Vi Vi
0
t
−Vo /n −Vo /n
Vi /Lp
isp1 Ip+
CCM
Ip− DCM
0
Is+ Vo /Ls t
CCM Io
iss1
DC
M Is−
0
Vi + Vo /n Vi + Vo /n t
Vsp1
0
t
nVi + Vo
nVi + Vo
Vss1
Figure 5.52 Waveforms of various signals of the flyback converter.
Chapter 05.indd 255 3/19/2009 1:30:30 PM

During the DTs period, the switch Sp1 is ON. The voltage across the primary is Vi with the dot end
positive. The current through the primary inductor and the switch isp1 increases with a rate of Vi/Lp where
Lp is the inductance as seen from the primary. The voltage across the secondary is nVi with the dot end
positive. During this time the diode switch Ss1 is OFF. The voltage across Ss1 is nVi + Vo.
During the (1 – D)Ts period, the switch Ss1 is ON. The voltage across the secondary is Vo and the volt-
age across the primary is Vo/n with the dot end negative. The stored inductor energy now flows through the
secondary diode and into the capacitor and the load. The rate of this current is Vo/Ls where Ls is the induc-
tance as seen from the secondary. During this time the primary side IGBT switch Sp1 is OFF. The voltage
across Sp1 is Vi + (Vo / n ).
Governing Equations
The design equations for selecting the switches of the flyback converter are as follows:
⎛ D ⎞
Vo = nVi ⎜ ⎟ (5.82)
⎝ 1− D ⎠
where n is the secondary to primary turns ratio and D is the duty ratio. Referring to the waveforms of
Figure 5.52,
⎛I +I ⎞
I o = ⎜⎜ s − s + ⎟⎟ (1 − D ) (5.83)
⎝ 2 ⎠
The stored inductor energy transferred to the capacitor during (1 – D)Ts period has to supply power Po to
the load during (1 – D)Ts period and also to replenish the energy lost by the capacitor during the previous
DTs period. Thus, PoTs amount of energy has to supplied from the stored inductor energy to the load,
during the (1 – D)Ts period. Thus,
1 1
PoTs = Ls ( I s2+ − I s2− ) = Ls ( I s + + I s − )( I s + − I s − ) (5.84)
2 2
Vo (1 − D )
I s+ − I s− = (5.85a)
Ls f s
From Eqs. (5.83) and (5.85a), one obtains
Io V (1 − D )
I s+ = + o (5.85b)
1− D 2 Ls f s
Io V (1 − D )
I s− = − o (5.85c)
1− D 2 Ls f s
Similarly on the primary side,
Pi ⎛ I p + + I p − ⎞
=⎜
I in = ⎟D (5.86)
Vi ⎜⎝ 2 ⎟
⎠
Assuming 100% efficiency, the energy that is put into the inductor during DTs period is
Chapter 05.indd 256 3/19/2009 1:30:30 PM

1 1
Po ⋅Ts = L ( I 2 − I 2 ) = L ( I + I )( I − I ) (5.87)
2 p p+ p− 2 p p+ p− p+ p−
Vi D
I p+ − I p− = (5.88)
Lp f s
I in Vi D (5.89)
I p+ = +
D 2 Lp f s
I in Vi D
I p− = −
D 2 Lp f s (5.90)
Iin and Vi can be obtained from the specified Io and Vo values using the input–output relationship for the fly-
back converter topology as given in Eq. (5.82).
For DCM operation, Ip– and Is– will be zero. In DCM operation, the switching period will be divided
into three parts: (a) D1Ts duration when the primary switch (IGBT) is ON, (b) D2Ts duration when the
secondary diode is ON and (c) the remaining duration when both the switches are OFF. The current require-
ment for the Sp1 IGBT switch is given as
I cm > I p + (5.91)
where Icm is the peak current requirement of the IGBT. The current requirement for the Ss1 diode switch is
given as
I dm > I s + (5.92)
The average current requirement for the Ss1 diode switch is
I d-avg = I o (5.93)
The rms current requirement for the Ss1 diode switch is

⎛I +I ⎞
I d-rms = ⎜⎜ s − s + ⎟⎟ 1 − D (5.94)
⎝ 2 ⎠
Referring to Figure 5.52 again, the peak OFF-state voltage that the primary side switch Sp1 should withstand is
Vo
Vceo > Vimax + (5.95)
n
The peak OFF-state voltage that the secondary side switch Ss1 should withstand is
PIV > nVi + Vo (5.96)
Chapter 05.indd 257 3/19/2009 1:30:31 PM

5.8 Practical Issues
I n the previous sections the three primary converters and the isolated converters were discussed. However,
there are a few practical issues that are not directly evident from the steady-state analysis. One should be
wary of the following problems when actually implementing the converter circuits:
1. effects due to leakage inductance, track or wire inductance and stray series inductance;
2. effects due to parasitic capacitances across the switching device;
3. current shoot-through during device turn-ON.
All practical transformers and magnetic couplings will always have a finite non-zero leakage inductance as
seen from a given winding. The voltage drops across these inductances that result due to the switch OFF of
the current through it will invariably result in large voltage spikes across the switch being switched OFF. If
either the leakage inductance is high due to poor magnetic coupling between windings or the di/dt for the
current through the switch which is being cut-off is high, then the resulting voltage spike that occurs across
the switch will lead to deteriorated operation of the switch due to increased voltage stress if not damaging
the device. Therefore, the issue of leakage inductance is a very important issue. There are, in general, two
solutions that are commonly adopted: (a) reduce the leakage inductance by using tightly wound windings
and (b) divert the energy stored in the leakage inductance. In the former solution, close coupling between
windings will increase the inter-winding capacitance and thereby reduce the differential voltage capacity
between the windings as high differential can lead to arcing or break over of the insulation. The latter solu-
tion is addressed by means of snubber circuits. These circuits are discussed in Chapter 2.
The power semiconductor switches have significant capacitance across their collector–emitter or drain–
source terminals. When the switch is OFF, the capacitance across the device will charge up to the OFF-state
voltage. Again when the switch is turned ON, the charged up capacitance sees a very low impedance path
through the ON-state device and discharges through it limited only by the ON-state resistance. Thus a very
large current spike will flow through the device at turn-ON that may damage the device. This is solved by
using series snubber circuits. Alternately, this can be addressed by using zero-current switching (ZCS) or
zero-voltage switching (ZVS) topologies. This will be discussed in a later section.
In many topologies, for example the half-bridge and full-bridge converters, there is a serious problem of
current shoot-through. In the case of a bridge arm, when the top device of a bridge arm is ON, the
corresponding bottom device is OFF. On switching OFF the top device and turning ON the bottom device,
the top device would not have recovered completely to the OFF-state as the turn-OFF process is slower than
the turn-ON process. Thus, there is a very small but finite period of time during the switching transition
when the input voltage source Vi is directly connected to the ground which will result in a very large current
spike. This will eventually damage the devices. There are again two solutions that are commonly adopted:
(a) providing dead time between the top and bottom devices to give enough time for the switching OFF
device to turn OFF and (b) using snubber circuits that are discussed in Chapter 2. It should be noted that
the above problem is not specific to only half-bridge or full-bridge circuits. It occurs in many topologies of
the converters discussed previously, in both primary converters and isolated converters.
5.9 Other Topologies
T he topologies discussed till now are the more common and standard topologies that are very popular.
However, there are numerous topologies that have been adapted from the primary converters for vari-
ous applications. In fact using the rules stated at the start of this chapter, one may construct different
Chapter 05.indd 258 3/19/2009 1:30:31 PM

topologies such that the volt-second balance, charge balance, flux reset and i–v characteristic requirements
for the switches are ensured. All the topologies can be analyzed in a similar manner as presented in the previ-
ous sections. The steady-state input–output relationship, the inductor and capacitor values, the switch rat-
ings and the type of semiconductor switches can be deduced from the waveforms of the various signals of
the converter. In this section, few topologies like the Cuk converter, the high-boost converter, zero-current
switching (ZCS ) and zero-voltage switching (ZVS) converters will be discussed briefly to give a flavor of their
operation and application.
Cuk Converter
The Cuk converter is named after the inventor of the converter. It consists of two inductors, one placed at the
input side like the boost topology and the other placed at the output side like the buck topology. The topol-
ogy also has two capacitors C1 and C2 as shown in Figure 5.53. There is an SPDT switch placed such that the
pole is connected to the circuit ground and the two throws T1 and T2 are connected, as shown, across the
capacitor C1. The output Vo that is obtained from this topology is negative with respect to the circuit ground
as will become clear from the circuit operation. It is interesting to note that both the input side and the
output side contain an inductor each. This implies that the input and output currents of this converter are not
switching or pulsed currents. Figure 5.54 shows the operative modes of the Cuk converter topology.
Referring to Figure 5.54, there are two operative modes for the Cuk converter circuit. During the DTs
period as shown in Figure 5.54(a), the SPDT switch is connected to throw T1 (i.e., S1 is ON and S2 is
OFF). The energy in the inductor L1 builds up by the current flowing from the input source Vi. The energy
in the inductor L2 also builds up due to transfer of the stored energy from capacitor C1 to the inductor and
capacitor C2. Observe the direction of flow of the L2 inductor current. It charges the output capacitor C2 in
such a direction that Vo is negative with respect to the circuit ground.
L1 C1 L2
−
Vi T1 T2 C2 RL Vo
P +
Figure 5.53 Schematic of the Cuk converter topology.
L1 C1 L2 L1 C1 L2
+ − + −
−
Vi C2 RL V Vi C2 R L Vo
+ o
S1 S2
(a) (b)
Figure 5.54 Cuk converter: (a) Operative circuit during DTs period; (b) operative circuit during
(1 – D)Ts period.
Chapter 05.indd 259 3/19/2009 1:30:31 PM

During the (1 – D)Ts period, the operative circuit is as shown in Figure 5.54(b). During this time, the
pole of the SPDT switch is connected to the throw T2 (i.e., S2 is ON and S1 is OFF). The energy that was
stored in the inductor L1 during the DTs period is now transferred to the capacitor C1. The inductor current
in L2 freewheels through the S2 switch just like in the buck converter case. The energy of inductor L2 is now
transferred to the capacitor C2 and the load RL.
Input–Output Relationship
To obtain the input–output voltage relationship, apply the volt-second balance rule to both the inductors.
This implies that the area under the inductor voltage waveforms in one period under steady-state conditions
should be zero.
Volt-Second Balance for L1 During DTs period the voltage across the inductor L1 is Vi and during
(1 – D)Ts period, the voltage across the inductor is Vi – VC1. Thus,
(Vi )DTs + (Vi − VC1 )(1 − D )Ts = 0 (5.97)
where VC1 is the voltage across capacitor C1. Re-arranging Eq. (5.97), one obtains
Vi
VC1 = (5.98)
1− D
Volt-Second Balance for L2 During DTs period the voltage across the inductor L2 is VC1 – Vo and during
(1 – D)Ts period, the voltage across the inductor is –Vo. Thus,
(VC1 − Vo )DTs + ( −Vo )(1 − D )Ts = 0 (5.99)
Re-arranging Eq. (5.99), one obtains
Vo
VC1 = (5.100)
D
From Eqs. (5.98) and (5.100), the voltage input–output relationship is given as
Vi D
Vo = (5.101)
1− D
Observe that the input–output voltage relationship is similar to the buck–boost input–output voltage
relationship. Figure 5.55 shows the circuit implementation schematic. Switch S1 is replaced by an IGBT
and switch S2 is replaced by a diode after considering the i–v characteristic requirements of the two
switches.
+ −
L1 L2
C1
−
Vi S1 S2 C2 RL Vo
+
Figure 5.55 Schematic of Cuk converter circuit implementation.
Chapter 05.indd 260 3/19/2009 1:30:31 PM

Zero-Current Ripple
The Cuk converter can be configured to give additional benefits by coupling the inductors L1 and L2. The
input side and output side inductors carry non-switching DC currents. However these currents have a ripple
current riding over them. By coupling the two inductors, one may obtain either the input side or the output
side inductor current with zero ripples. For example, if the output side inductor current ripple is made zero,
the size of the capacitor C2 which is dependent on the inductor ripple current will be drastically reduced if
not eliminated.
The inductive coupling may be done for any converter having multiple inductors provided the instanta-
neous voltages across the two inductors are the same or have the same scale factors. In the case of the Cuk
converter, consider the voltage across the two inductors at every instant. First, an estimate of the voltage
across the capacitor C1 in the steady state can be obtained from Eqs. (5.98) and (5.100) as
VC1 = Vi + Vo (5.102)
This is also evident from applying the Kirchhoff ’s voltage law to the outer loop of the Cuk converter in the
steady state when the drops across the inductors can be considered zero. Let VL1 and VL2 be the voltages
across the inductors L1 and L2, respectively.
During DTs period
VL1 = Vi
VL2 = VC1 − Vo = Vi
During (1 – D)Ts period
VL1 = Vi − VC1 = Vi − Vi − Vo = −Vo
VL2 = −Vo
M N2
M V
V L1σ + M N1 L
L1 L2 L1σ + M L
L1σ L2σ
N1:N2
M
+ −
Ideal
C1 VL Tx VL
+ −
−
Vi RL Vo C1
S1 S2 C2 −
+ S1 S2 C2 RL Vo
+
(a) (b)
Figure 5.56 (a) Cuk converter with inductive coupling; (b) non-idealities in the inductive coupling.
Chapter 05.indd 261 3/19/2009 1:30:31 PM

From the above it is evident that the voltage across L1 and L2 are same at every instant. Thus there is a case
for coupling the two. Figure 5.56(a) gives the schematic of the Cuk converter with the two inductors
coupled with the specified dot polarity. If the output current ripple is to be reduced to zero, the equivalent
circuit of the coupled inductor portion is as shown in Figure 5.56(b). The non-idealities like the leakage
inductances are put to good use in bringing down the current ripple to zero. The leakage inductance L2s is
the component in focus. If the voltage across the leakage inductance L2s is made zero, then as L2s is non-
zero, di/dt through the inductor L2s has to be zero. This implies that the ripple in the current flowing
through L2s is zero. As the current flowing through L2s is the same as the current flowing through L2, the
L2 current ripple is eliminated and shifted to the L1 side. This means that C2 can be very small or
eliminated.
Referring to Figure 5.56(b), the voltage that gets translated from the L1 side to the L2 side is
⎛ M ⎞ ⎛ N2 ⎞
⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟ VL1
⎝ L1σ + M ⎠ ⎝ N1 ⎠
The voltage across L2s should be zero. The condition for that is
⎛ M ⎞ ⎛ N2 ⎞
⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟ VL1 − VL2 = 0
⎝ L1σ + M ⎠ ⎝ N1 ⎠
As VL1 = VL2 at every instant,
⎛ M ⎞ ⎛ N2 ⎞
⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟ = 1
⎝ L1σ + M ⎠ ⎝ N1 ⎠
M N
= 1 (5.103)
L1σ + M N 2
The constraint of Eq. (5.103) should be satisfied to obtain zero ripple in the L2 inductor current. Similarly,
analysis can be done to obtain zero ripple current through L1 inductor at the input side by making the volt-
age across the L1s leakage inductance zero. It should be noted that zero ripple current can be obtained either
in the L1 inductor current or the L2 inductor current and not both together.
Isolated Cuk Converter

Isolation can be incorporated in the Cuk converter also. Figure 5.57 shows the schematic of the isolated Cuk
converter. The capacitor C1 is split into two equal parts as C1p and C1s and the transformer is placed in
between them as shown. The input–output relationship can be derived as discussed before for other convert-
ers by using the volt-second balance. Thus,
nVi D
Vo = (5.104)
1− D
where n is the secondary to primary turns ratio.
In the case of the isolated Cuk converter shown in Figure 5.57, consider the voltage across the two
inductors and the two windings of the transformer at every instant. Applying the Kirchhoff ’s voltage law
to the primary loop and the secondary loop of the Cuk converter in the steady state, it is evident that the
voltage across C1p is Vi and the voltage across C1s is Vo as indicated in the Figure 5.57. Let VL1 and VL2 be
Chapter 05.indd 262 3/19/2009 1:30:32 PM

Vi Vo
+ − 1:n + −
L1 C1p C1s L2
−
Vi Vp Vs C2 RL Vo
+
Tx
Figure 5.57 Isolated Cuk converter.
the voltages across the inductors L1 and L2, respectively, and Vp and Vs be the voltages across the primary
and secondary windings of the transformer, respectively.
During DTs period
VL1 = Vi
Vp = Vi
Vs = nVi
VL2 = VC1s + Vs − Vo = Vo + nVi − Vo = nVi
During (1 – D)Ts period

Vo
VL1 = Vi − VC1p + Vp = −
n
Vo
Vp = −
n
Vs = −Vo
VL2 = −Vo
Thus from above one can observe that if n = 1, then all the windings, that is, the L1 inductor, the L2 induc-
tor, the primary winding and the secondary winding, all have the same voltage across them at every instant.
For n other than 1, the voltages across the primary side windings, that is L1 and primary winding, is differ-
ent from the voltages across the secondary side windings, that is L2 and secondary winding, by a constant
factor of n. Thus, there is a case to couple all the windings onto the same magnetic core. This coupled mag-
netic Cuk converter wherein the L1 inductor, the L2 inductor and the transformer windings are all wound
on the same common core is shown in Figure 5.58. Here both the input side and the output side current
ripples can be made zero simultaneously by transferring the ripples to the transformer. Analysis for this can
be performed along similar lines as discussed for the non-isolated Cuk converter with coupled inductors.
High-Boost Converter
The boost converter discussed as one of the primary converters cannot give very high gain as explained ear-
lier. As the duty ratio D becomes unity, the output does not become infinite but zero. In fact, a practical
boost converter can give a gain of only about 4 or 5. To get a very high boost, the circuit of Figure 5.59 is a
very useful and interesting topology.
Chapter 05.indd 263 3/19/2009 1:30:32 PM

L1 L2
C1p C1s
Vi S1 S2 C2 RL Vo
Figure 5.58 Coupled magnetic Cuk converter.
LI D1 L2
S3
S1
C1
−
Vi − + D2 C2 RL Vo
+
S2 S4
Figure 5.59 Schematic of a high-boost converter.
This topology also consists of two inductors L1 and L2 as shown and four switches in a bridge configura-
tion. During the time when S1 and S4 are OFF, the capacitor C1 charges up to Vi with the polarity, as
shown, by charging from the input source Vi through the diodes S3 and S2. When switches S1 and S4 are
turned ON, the voltage across C1 appears across the diodes S3 and S2 to reverse bias them thereby turning
OFF the diodes S3 and S2. The output portion of the circuit is similar to the buck converter topology. The
operation of the converter is described as follows,
During the Time DTs

Here S1 and S4 are ON and S3 and S2 are OFF. The voltage across L1 is given as
VL1 = Vi + VC1 (5.105)
The inductor energy builds up during this time. During this time, the capacitor C1 also appears across
the output circuit in such a manner as to reverse bias diode D2 and forward bias D1. The capacitor C1
energy is transferred to the output inductor wherein there is build up of energy. The voltage across L2 is
given as
VL2 = VC1 − Vo (5.106)
Chapter 05.indd 264 3/19/2009 1:30:32 PM

During the Time (1 – D)Ts

In this case S1 and S4 are turned OFF and S3 and S2 are ON. The energy in the inductor L1 charges up the
capacitor C1 through the diodes S3 and S2. The voltage across L1 is given as
VL1 = Vi − VC1 (5.107)
During this time, the current in L2 will freewheel through D2 transferring energy from L2 to C2. The diode
D1 is reverse-biased during this time and is therefore OFF. The voltage across L2 is given as
VL2 = −Vo (5.108)
Input–Output Relationship
The input–output relationship can be found out as before by applying the volt-second balance rule for the
inductors during the steady-state operating conditions. Using Eqs. (5.105) and (5.107) and applying the
volt-second balance rule for L1, one obtains
(Vi + VC1 )DTs + (Vi − VC1 )(1 − D )Ts = 0
Vi
VC1 = (5.109)
1 − 2D
Using Eqs. (5.106) and (5.108) and applying the volt-second balance rule for L2, one obtains
(VC1 − Vo )DTs + ( −Vo )(1 − D )Ts = 0
Vo = VC1D (5.110)
From Eqs. (5.109) and (5.110), the following input–output relationship results:
Vi D
Vo = (5.111)
1 − 2D
From the above input–output relationship, it can be observed that when D = 0.5, Vo tends towards infinity.
It can be noted that when D = 0.5, the converter switches are switching with 50% ON and 50% OFF dura-
tions and the steady-state input–output relationship of Eq. (5.110) is still valid. Thus one can obtain high
boost or gains from this topology.
From Eqs. (5.105)–(5.109) and Eq. (5.111), it can be deduced that during the period DTs
2Vo (1 − D ) ; V (1 − D )
VL1 = VL2 = o
D D
Likewise, during the period (1 – D)Ts,
VL1 = −2Vo ; VL2 = −Vo
Thus it can be observed that VL1 and VL2 differ from each other by a constant factor of 2 at every instant of
time. Therefore, the two inductors can be coupled just as was discussed for the Cuk converter to make either
the input current or the output current ripple zero.
Soft Switching in Converters

The switched-mode converter topologies consist of power semiconductor switches. They are in no way ideal
switches as can be seen from the discussion on the various semiconductor switch characteristics in Chapter 1.
Chapter 05.indd 265 3/19/2009 1:30:32 PM

In general, most of the switches in the switched-mode power converters are not switched in excess of
100 kHz. Switching at higher frequencies in excess of 1 MHz is desirable as the isolating transformer
can become smaller and the sizes of the output L and C can also become smaller, thereby improving or
bringing down the volumetric power density (power per unit volume) of the converter. However, one of the
most important factors which prevent switching at higher frequencies is the switching losses of the power
devices. As discussed in Chapter 1, every switch dissipates power during switching transitions. Thus as the
switching frequencies increases, there are more switching transitions in a given time and thereby more power
loss. Therefore, it becomes important to tackle the issue of switching losses in the switched-mode
converters.
The switching losses occur due to the increase of one variable (voltage or current) and the decrease of the
other variable during the switching transition. The product of these two variables is non-zero during the
switching transition which results in the switching loss. If one of the two variables, either the voltage across
the switch or the current through the switch, is maintained at zero during the switching transition, then the
product of the voltage and current is zero resulting in zero switching loss. Such zero switching loss convert-
ers are called soft switching converters. If the current is maintained at zero during the switching transition,
then such soft switching converters come under the category of ZCS converters. On the other hand, if the
voltage is maintained at zero during the switching transition, then such soft switching converters come
under the category of ZVS converters.
Any of the converters discussed till now can be converted into ZCS or ZVS converter. The change is
actually in the structure of the controlled switch. To realize ZCS, a primary requirement is that the con-
trolled switch should have an inductor in series with it so that the current does not have discontinuities.
Figure 5.60(a) shows the realization of a zero-current switch. The internal body diode of the controlled
switch is needed for the operation of the switch.
To realize ZVS, a primary requirement is that the controlled switch should have a capacitor across it so
that the voltage across the device does not have discontinuities. Figure 5.60(b) shows the realization of a zero-
voltage switch. A diode D is included in series with the switch S. Most of the commercially available switches
D
D
S D
S C S C
L L L
(a) (b) (c)
Figure 5.60 (a) Zero-current switch realization; (b) zero-voltage switch realization; (c) zero-voltage
switch realization if internal body diode is present in the controlled switch.
Chapter 05.indd 266 3/19/2009 1:30:33 PM

like the MOSFETs and IGBTs will have internal body diodes. If these internal body diodes are present, they
will automatically be blocked by the diode D as shown in Figure 5.60(c) to realize zero-voltage switch.
The controlled switches of any converter topology can be replaced with the zero-current switch or the
zero-voltage switch as indicated in Figure 5.60 to obtain ZCS or ZVS operation of the converter. This prin-
ciple can be demonstrated by replacing the controlled switch of the traditional buck converter with the
zero-current switch and the zero-voltage switch to achieve ZCS and ZVS operations.
Zero-Current Switching Applied to Buck Converter

Figure 5.61 shows the circuit schematic of a buck converter with the controlled switch replaced with a zero-
current switch. The switch S1 is an IGBT with an internal body diode D1. Ls and Cs are components associ-
ated with the zero-current switch. Ls is the inductor that is in series with the switch S1 as per the zero-current
switch requirements. Cs is a capacitor placed in parallel with the switch and inductor together as shown in
Figure 5.61. The rest of the circuit is same as the buck converter circuit.
The following are the four operative modes of the ZCS converter:
1. load current freewheeling with controlled switch in OFF-state;
2. energy build up of ZCS inductor Ls at constant voltage;
3. Ls–Cs resonance operation;
4. capacitor Cs charging at constant current.
Figure 5.62 shows the four operative circuits in the four modes of operation of the ZCS converter. As the
output inductor is very large compared to the soft-switching inductor Ls, Io is considered to be constant
during the switching period. The operation of the ZCS converter is described as follows after having reached
the steady-state operating condition,
Mode 1 – S1 OFF and Load Current Freewheeling During this time, the output inductor current with
average Io will freewheel through the diode D2 as shown. As D2 is ON, one end of the capacitor Cs is at
circuit ground potential and the other end is at Vi. Therefore, Cs will have a voltage of Vi across it with the
polarity as shown in Figure 5.62(a).
D1
vcs
Iin Cs
S1
Vi
Ls iLS
Io
L
D2 Id C RL Vo
Figure 5.61 Buck converter with zero-current switch.
Chapter 05.indd 267 3/19/2009 1:30:33 PM

Iin
S1
+ Vcs
Cs
Vi −
Ls
Io
D2 RL Vo
C
(a)
Iin
S1
Vcs
Cs
Vi
Ls iLs
Io
L
D2 RL V o
Id C
(b)
Iin
S1
Vcs
Cs
Vi
Ls iLs
Io
a
L
D2 RL Vo
C
(c)
Figure 5.62 ZCS buck converter: (a) Load current freewheeling and S1 OFF; (b) energy build up of
inductor Ls; (c) resonance operation; (d) capacitor Cs charging at constant current.
Chapter 05.indd 268 3/19/2009 1:30:33 PM

Iin
S1
Vcs
Cs
Vi
Ls
Io
L
D2 RL Vo
C
(d)
Mode 2 – S1 ON and Inductor Energy Build Up At time t1, the switch S1 is turned ON. The operative
circuit is shown in Figure 5.62(b). As D2 is still conducting, the voltage across the inductor Ls is Vi. The
inductor current iLs rises linearly with a rate of Vi/Ls. Thus,
Vi
iLs = (t − t1 ) (5.112)
Ls
The current iLs rises linearly till it reaches the value of Io when D2 becomes reverse-biased and switches
OFF.
Mode 3 – S1 ON and Resonance Operation The operative circuit is shown in Figure 5.62(c). After iLs
reaches the value of Io, D2 is OFF. iLs will reach Io at time t2 that can be deduced from Eq. (5.112) and is
given as
I o Ls
t 2 − t1 = (5.113)
Vi
Cs and Ls form a resonant tank. Cs discharges in a resonant manner through S1 and Ls. The current through
the inductor Ls is the sum of the currents at the node “a” of Figure 5.62(c). Considering the time reference
to start at the beginning of this mode, the inductor current is given as
⎛V ⎞
iLs = I o + ⎜ i ⎟ sin ω t (5.114)
⎝Z ⎠
where Z = Ls / C s is the impedance of the resonant tank and ω = LsC s is the natural frequency of reso-
nance of the Ls–Cs tank. The voltage across the capacitor Cs is given as
vCs = Vi cos ω t (5.115)
Chapter 05.indd 269 3/19/2009 1:30:33 PM

Observing the iLs waveform shown in Figure 5.63, it can be seen that the current waveform will resonate and
cut the zero axis only if
Vi
> Io (5.116)
Z
If t3 is the time at which the inductor current iLs cuts the zero axis, then from Eq. (5.114), one obtains
⎛V ⎞
0 = I o + ⎜ i ⎟ sin ω(t 3 − t 2 )
⎝Z ⎠
⎛ 1 ⎞ ⎧⎪ ⎛I Z ⎞⎫⎪
t 3 − t 2 = ⎜ ⎟ ⎨π + sin −1 ⎜⎜ o ⎟⎟⎬ (5.117)
ω
⎝ ⎠ ⎩⎪ ⎝ Vi ⎠⎪⎭
After the inductor current iLs reverses at t3, the current flows through the internal body diode of the switch
S1. During this time S1 can be turned OFF as the current through it is 0. Thus during both switch-ON and
switch-OFF of S1, the current through it is 0 ensuring ZCS operation.
Mode 4 – S1 OFF and Capacitor Charging After the inductor current iLs reverses and flows through
the body diode of the switch S1, the S1 is turned OFF. S1 is turned OFF anytime in the time band
when iLs is negative and flowing through the body diode. The resonant current flow through the body
diode of switch S1 continues till the inductor current iLs reaches zero again. Then the whole of the
output current Io will flow through the capacitor Cs. The capacitor Cs will now be charged to Vi at con-
stant current Io.
If t4 is the moment when the current through the body diode and Ls becomes zero again, then from
Eq. (5.114), one obtains
⎛ 1 ⎞ ⎧⎪ ⎛ I Z ⎞⎪⎫
t 4 − t 2 = ⎜ ⎟ ⎨2π − sin −1 ⎜ o ⎟⎬ (5.118)
ω
⎝ ⎠ ⎩⎪ ⎝ Vi ⎠⎪⎭
From t4 onwards up to t5 the capacitor charges linearly with constant current Io. It reaches the voltage Vi at
time t5. The time t5 can be calculated as
C s (Vi − vCs-t4 )
t5 − t 4 = (5.119)
Io
where vCs-t4 is the voltage across Cs at time t4 when the charging of the capacitor at constant current begins.
It can be obtained from Eq. (5.115).
After the capacitor is charged to Vi, the current through the capacitor is zero and the Io now flows
through the freewheeling diode D2. This is the same state as Mode 1. The whole cycle repeats.
Referring to the waveforms of Figure 5.63, it can be observed that the output voltage is the average of
the VD2 waveform. Thus the input–output relationship is not a simple linear function of the duty ratio as in
the case of the traditional buck converter, but it is a non-linear function of the switching frequency and the
resonant frequency. It should also be noted that the time duration of the ON-state of the switch S1 is
Chapter 05.indd 270 3/19/2009 1:30:33 PM

Turn ON S1 Turn OFF S1 Turn ON S1

Mode
Mode 1 Mode 2 Mode 3 4
t1 t2 t3 t4 t5 t
iLS Io
Vi /Ls Vi /Z
0
Io /Cs t
Vi
VCS
0
t
VD2 = Vi − VCS
0
t
Io
iin
0
t
Figure 5.63 Waveforms of important signals of ZCS buck converter.
decided by the resonant circuit of the zero-current switch. But the time duration of the OFF-state of the
switch S1 can be controlled to any length.
One of the main drawbacks of the ZCS topology is that at very high frequency the parasitic capacitance
across the switch S1 becomes significant and its discharge current through S1 cannot be limited on any
account. Therefore, ZCS topology has an upper frequency bound beyond which it does not serve the pur-
pose of lossless switching.
Zero-Voltage Switching Applied to Buck Converter

Figure 5.64 shows the circuit schematic of a buck converter with the controller switch replaced with a zero-
voltage switch. The switch S1 is an IGBT. Ls and Cs are components associated with the zero-voltage switch.
Cs is the capacitor that is in parallel across the switch S1 as per the zero-voltage switch requirements. Ls is an
inductor placed in series with the switch and capacitor together as shown in Figure 5.64. The rest of the cir-
cuit is same as the buck converter circuit.
Chapter 05.indd 271 3/19/2009 1:30:34 PM

Iin
D
vcs
CS
S1
Vi
Ls iLS
Io
L
D2 Id C RL Vo
Figure 5.64 Schematic of the ZVS buck converter.
The following are four operative modes of the ZVS converter:

1. load current through switch S1;
2. capacitor Cs charging at constant current;
3. Ls–Cs resonance operation;
4. energy build up of ZVS inductor Ls at constant voltage.
Figure 5.65 shows the four operative circuits in the four modes of operation of the ZVS converter. As the
output inductor is very large compared to the soft-switching inductor Ls, Io is considered to be constant
during the switching period. The operation of the ZVS converter is described as follows after having reached
the steady-state operating condition.
Mode 1 – S1 ON and Load Current through S1 During this time, the output inductor current with
average Io will flow through S1 as shown. As S1 is ON, the voltage across Cs is zero. The current through Ls
is same as the current flowing through S1, that is, Io.
Mode 2 – S1 OFF and Capacitor Charging At time t1, the switch S1 is turned OFF. Due to the presence
of Cs the voltage across the switch S1 is zero and will not change suddenly. The operative circuit is shown in
Figure 5.65(b). Once the switch S1 turns OFF, the inductor current now shifts to Cs which starts charging
at constant current Io. The capacitor voltage vCs rises linearly with a rate of Io/Cs. Thus,
Io
vCs = (t − t1 ) (5.120)
Cs
The voltage vCs rises linearly till it reaches the value of Vi when D2 becomes forward-biased and switches ON.
Mode 3 – S1 OFF and Resonance Operation The operative circuit is shown in Figure 5.65(c). After vCs
reaches the value of Vi, D2 is ON. The value of vCs will reach that of Vi at time t2 that can be deduced from
Eq. (5.120) and is given as
VC
t 2 − t1 = i s (5.121)
Io
Chapter 05.indd 272 3/19/2009 1:30:34 PM

D1 D1
Iin Iin
0V 0V
S1 Cs S1 Cs
Vi Vi
Ls iLS Ls iLS
L L
Io Io
D2 C RL Vo D2 C RL Vo
(a) (b)
D1 D1
Iin Iin
Vcs Vcs
S1 Cs S1 Cs
Vi Vi
Ls iLS Ls iLS
Io Io
L L
D2 C RL V o D2 C RL Vo
(c) (d)
Figure 5.65 ZVS buck converter: (a) Load current through S1; (b) capacitor Cs charging at constant
current; (c) resonance operation; (d) energy build up of inductor Ls.
Cs and Ls form a resonant circuit. Ls current falls in a resonant manner through D2 and Cs. Applying
Kirchhoff’s voltage law in the resonant loop shown in Figure 5.65(c) and considering the time reference to
start at the beginning of this mode, the capacitor voltage is given as
vCs = Vi + ( I o Z )sin ω t (5.122)
where Z = Ls / C s is the impedance of the resonant tank and ω = LsC s is the natural frequency of reso-
nance of the Ls–Cs tank.
Chapter 05.indd 273 3/19/2009 1:30:34 PM

The current through the inductor Ls is given as

iLs = I o cos ωt (5.123)
Observing the vCs waveform shown in Figure 5.66, it can be seen that the voltage waveform will resonate
and cut the zero axis only if
I o Z > Vi (5.124)
If t3 is the time at which the capacitor voltage vCs cuts the zero axis, then from Eq. (5.122), one obtains
0 = Vi + ( I o Z )sin ω (t 3 − t 2 )
⎛ 1 ⎞ ⎧⎪ ⎛ V ⎞⎫⎪
t 3 − t 2 = ⎜ ⎟ ⎨π + sin −1 ⎜⎜ i ⎟⎟⎬ (5.125)
⎝ ω ⎠ ⎩⎪ ⎝ I o Z ⎠⎪⎭
After the capacitor voltage vCs reverses at t3, the switch S1 can be turned ON again.
Mode 4 – S1 ON and Inductor Energy Build Up After the capacitor voltage vCs reverses the switch S1 is
turned ON. S1 is turned ON anytime in the time band when vCs is negative. The negative voltage will be
supported by the diode D. The resonant current flows through D2, Ls and Cs till the capacitor voltage vCs
Turn OFF S1 Turn ON S1 Turn OFF S1

Mode
Mode 1 Mode 2 Mode 3 4
t1 t2 t3 t4 t5 t
VCS
Io Z
Vi Io /Cs
0
Vi /Ls t
Io
iLS
iin
0
t
VD2
Vi
0
t
Figure 5.66 Waveforms of important signals of ZVS buck converter.
Chapter 05.indd 274 3/19/2009 1:30:34 PM

reaches zero. Then the whole input voltage Vi will appear across Ls. The energy in Ls will now build up to Io
at constant voltage Vi across it.
If t4 is the moment when the voltage across the capacitor reaches zero again, then from Eq. (5.122), one
obtains
⎛ 1 ⎞ ⎪⎧ ⎛ V ⎞⎪⎫
t 4 − t 2 = ⎜ ⎟ ⎨2π − sin −1 ⎜⎜ i ⎟⎟⎬ (5.126)
⎝ ω ⎠ ⎩⎪ ⎝ I o Z ⎠⎭⎪
From t4 onwards up to t5 the inductor energy builds up linearly with constant voltage Vi across it. It reaches
the current Io at time t5. The time t5 can be calculated as
L (I − i )
t 5 − t 4 = s o Ls-t4 (5.127)
Vi
where iLs-t4 is the inductor current through Ls at time t4 when the inductor energy build up at constant volt-
age begins. The value iLs-t4 can be found from Eq. (5.123).
After the inductor current has build up to Io, the converter is in Mode 1. The whole cycle repeats.
Referring to the waveforms of Figure 5.66, it can be observed that the output voltage is the average of
the VD2 waveform. It should also be noted that the time duration of the OFF-state of the switch S1 is
decided by the resonant circuit of the zero-voltage switch. However, the time duration of the ON-state of
the switch S1 can be controlled to any length. It should also be noted that as the device parasitic capacitance
can be taken as part of the resonating capacitor Cs; the ZVS topology can be operated at much higher fre-
quencies as compared to the ZCS topology.
The ZCS and ZVS converters are also called quasi-resonant converters as Ls and Cs resonate during a
fraction of the switching period.
The DC–DC converters are one of the most widely semiconductor device that may be used as a switch for
used circuits in electronic equipments. Any elec- a given converter topology. Among the isolated power
tronic equipment requires one or more power sup- converters, the flyback converter is by far the most
plies for its operation. Therefore, whether it is popular. This is because the flyback converter has the
low-power applications like the mobile phones, digi- least component count and simple to implement.
tal camera, battery chargers or high-power applica- With MOSFETs as the power switch, switching
tions like motor drive controller, uninterruptible frequencies of 100 kHz are easily achievable. How-
power system, etc., DC–DC converters are used to ever, to reduce the size of the magnetics (inductor
act as power interface between various DC buses. and transformer cores) one may need to switch to
The converters discussed in this chapter are the higher frequencies that tend towards 1 MHz. Also,
more popular ones. However, there are numerous the MOSFET switching losses and the core losses
topologies in the literature for specific applications. could become prohibitively high. Towards this end,
The key point to be stressed here is that the steady- soft switching may be introduced to decrease the
state analysis gives a framework for designing the switching losses.
open-loop converter system. The static i–v character- The power supplies are in general multiple-
istics of the devices and the SPDT switch require- output winding converters that supply multiple iso-
ments should be matched to arrive at the best power lated outputs. The regulation aspects and the
Chapter 05.indd 275 3/19/2009 1:30:34 PM

controller issues with respect to control of multiple meant to improve the insights gained in this chapter
outputs are addressed in Chapter 10. Finally, the and form the first steps towards practical implemen-
laboratory exercises given in the next section are tation of full-fledged DC–DC power converters.
1. A pulse-width modulated (PWM) waveform is (b) View the waveform at pin 5. What is the
required to operate the switches of the DC– nature of the waveform? Is it a sawtooth
DC converter. There are numerous ways in waveform? Measure the extremities on the
which a PWM waveform may be generated. It y-axis and also measure the frequency of
can be generated either by discretely building the waveform. Check from the nomo-
circuits for generating a triangular carrier and graph in the datasheet of the IC. Observe
comparing a DC waveform with the carrier or whether or not the frequency obtained is
by using a digital port of the microcontroller. correct.
There are also many commercial ICs available (c) Change RT and CT and observe the change
that are dedicated to generate PWM wave- in the frequency.
forms. A sample PWM generation circuit using (d) Measure the voltage at pin 1. Adjust the
TL494 IC is shown in Figure 5.67. potentiometer till the voltage at pin 1 is at
a value in between the y-axis extremities of
12/15 V sawtooth voltage at pin 5.
Vcc
(e) Monitor the voltage at pins 11 and 8.
(f) What happens when the voltage at pin 1 is
14 16 12 1.5 K
changed? Why?
15 11 PWM1 (g) What is the voltage at pin 14?
4 (h) What is the function of pin 13? Check with
10
7 the datasheet and discuss its application.
13 TL 494 1.5 K (i) Why are pins 15 and 16 connected in the
2 manner shown in Figure 5.67? What other
8 PWM2
3 ways can pins 15 and 16 be used?
9 (j) Resistors of value 1.5 kΩ are connected
1K 1
6 5 from pins 8 and 11 to Vcc. What is the
CT lowest value to which these resistors may
5K
pot 5.6 K RT 10 KpF
be reduced? Can they directly drive BJTs
and MOSFETs? Discuss.
2. The circuit shown in Figure 5.68 is a buck
Figure 5.67 Pulse-width modulator circuit. converter circuit or a step-down DC–DC
converter. The transistor or BJT is operated as
Mode of implementation: The above circuit is a switch, that is, it is either in ON-state or OFF-
studied by state. Instead of a BJT, one may use a MOSFET
a. Hardware bread-boarding or IGBT.
Tasks for study: Mode of implementation: The above circuit is
(a) Rig up the circuit as shown in Figure 5.67. studied by
Chapter 05.indd 276 3/19/2009 1:30:34 PM

iin a iL c Io
Q
id L ic
Vi
C RL Vo
PWM
generator
Figure 5.68 Buck converter circuit.
a. Simulation in SciLab the x-axis and Vo/Vi on the y-axis. Based on

b. Simulation in SPICE the plot, specify the range of valid duty
c. Hardware bread-boarding ratio for the converter.
(j) What is the value of the conduction param-
Tasks for study:
eter? Plot Kbound versus d, 1 – d, 1 – d )2 and
(a) Rig up the above circuit with appropriate
d(1 – d )2 for the converter. From the plots,
values for the components.
comment on the value of Kbound for the
(b) Observe the voltage waveforms across the
converter.
diode, inductor and output.
(c) What is the average voltage waveform 3. The circuit shown in Figure 5.69 is a buck con-
across the inductor during steady state? verter circuit. The specifications for this circuit
(d) Observe the inductor current and the are: Vin = 15 V; Vo = 5 V; Io = 50 mA – 500
output voltage waveforms together. Why mA; fs = 20 kHz; ΔVo = 50 mV.
does the output voltage ripple waveshape
Mode of implementation: The above circuit is
differ significantly from the triangular
studied by
inductor current ripple waveshape?
a. Hardware bread-boarding
(e) Tabulate the steady-state input and output
voltages for various duty ratios D. From the Tasks for study:
tabulation, find the relationship between (a) Rig up the above circuit with appropriate
duty ratio D, input voltage and output values for the components.
voltage. (b) The inductor L has to be fabricated by
(f ) Tabulate the steady-state input and output winding copper wire on an EE-10 ferrite
currents for various duty ratios D. From the core. A paper may be inserted in the cen-
tabulation, find the relationship between tral limb of the EE core as air gap to obtain
duty ratio D, input current and output the proper inductor value of the inductor.
current? To experimentally measure the inductor
(g) What is the effect of load variation on the value, a small test jig has to be built using a
inductor current ripple amplitude? MOSFET switch with the test inductor
(h) What happens to the inductor current as connected between the positive rail and
the load is decreased? What is the mini- the drain of the MOSFET. A freewheeling
mum load to maintain continuous current diode–resistor should be connected across
in the inductor? the inductor. The current through the
(i) Tabulate and plot Vo/Vi versus the duty inductor can be measured by including a
ratio for the converter. Plot duty ratio on very small resistor in the source of the
Chapter 05.indd 277 3/19/2009 1:30:35 PM

2N4033 1.7 mH
560 Ω
Vin FR304 100 μF Ro
220 Ω
From TL494
Figure 5.69 Buck converter circuit.
switch. Switching pulses are given to the load to maintain continuous current in the
gate of the MOSFET and the voltage inductor?
across the inductor and the current through (h) Tabulate and plot Vo/Vin versus the duty
the inductor are measured in an oscillo- ratio for the converter. Plot duty ratio on
scope. Using Faraday’s equation, calculate the x-axis and Vo/Vin on the y-axis. Based
the value of the inductor. on the plot, specify the range of valid duty
(c) Observe the voltage waveforms across the ratio for the converter.
diode, inductor and output. (i) Find out the value of K for the circuit. Plot
(d) Observe the currents through L, switch Kbound versus d for the converter.
and the load.
4. The circuit shown in Figure 5.70 is a boost con-
(e) Tabulate the steady-state input and
verter circuit or a step-up DC–DC converter. The
output voltages for various duty ratios D.
transistor or BJT is operated as a switch, that is,
From the tabulation, find the input–
it is either in ON-state or OFF-state. Instead of
output voltage relationship and compare
a BJT, one may use a MOSFET or IGBT.
with the idealized input–output voltage
relationship. Mode of implementation: The above circuit is
(f ) What is the effect of load variation on the studied by
inductor current ripple amplitude? a. Simulation in SciLab
(g) What happens to the inductor current as b. Simulation in SPICE
the load is decreased? What is the minimum c. Hardware bread-boarding
L
iin iL b id c io
a
ic
Vi Q RL Vo
PWM C
generator
Figure 5.70 Boost converter circuit.
Chapter 05.indd 278 3/19/2009 1:30:35 PM

Tasks for study: From the plots, comment on the value of

(a) Rig up the above circuit with appropriate Kbound for the converter.
values for the components.
5. The circuit shown in Figure 5.71 is a buck–
(b) Observe the voltage waveforms across the
boost converter circuit. The transistor or BJT is
diode, inductor and output.
operated as a switch, that is, it is either in ON-
(c) What is the average voltage waveform
state or OFF-state. Instead of a BJT, one may
across the inductor during steady state?
use a MOSFET or IGBT.
(d) Observe the inductor current and the
output voltage waveforms together. Mode of implementation: The above circuit is
(e) Tabulate the steady-state input and output studied by
voltages for various duty ratios D. From a. Simulation in SciLab
the tabulation, find the relationship b. Simulation in SPICE
between duty ratio D, input voltage and c. Hardware bread-boarding
output voltage.
(f ) Tabulate the steady-state input and output Tasks for study:
currents for various duty ratios D. From (a) Rig up the above circuit with appropriate
the tabulation, find the relationship values for the components.
between duty ratio D, input current and (b) Observe the voltage waveforms across the
the output current. diode, inductor and output.
(g) What is the effect of load variation on the (c) What is the average voltage waveform
inductor current ripple amplitude? across the inductor during steady state?
(h) What happens to the inductor current as (d) Observe the inductor current and the
the load is decreased? What is the mini- output voltage waveforms together.
mum load to maintain continuous current (e) Tabulate the steady-state input and output
in the inductor? voltages for various duty ratios D. From
(i) Tabulate and plot Vo/Vi versus the duty the tabulation, find the relationship
ratio for the converter. Plot duty ratio on between duty ratio D, input voltage and
the x-axis and Vo/Vi on the y-axis. Based on output voltage.
the plot, specify the range of valid duty (f ) Tabulate the steady-state input and output
ratio for the converter. currents for various duty ratios D. From
(j) What is the value of the conduction the tabulation, find the relationship
parameter? Plot K bound versus d, 1 – d, between duty ratio D, input current and
(1 – d )2 and d(1 – d )2 for the converter. output current.
iin a id c io
Q
iL
ic
Vi L RL Vo
C
PWM
generator
Figure 5.71 Buck–boost converter circuit.
Chapter 05.indd 279 3/19/2009 1:30:35 PM

(g) What is the effect of load variation on the Tasks for study:
inductor current ripple amplitude? (a) Rig up the above circuit with appropriate
(h) What happens to the inductor current as values for the components.
the load is decreased? What is the mini- (b) Observe the voltage waveforms across the
mum load to maintain continuous current BJT, primary, secondary, the secondary
in the inductor? freewheeling diode and output.
(i) Tabulate and plot Vo/Vi versus the duty (c) Observe the current waveforms through
ratio for the converter. Plot duty ratio on the various components of the circuit.
the x-axis and Vo/Vi on the y-axis. Based on (d) Keeping the duty cycle at 0.4, monitor the
the plot, specify the range of valid duty voltage waveform across the BJT. What is the
ratio for the converter. effect of duty cycle on the waveshape? What
( j) What is the value of the conduction param- should be the voltage rating of the BJT?
eter? Plot Kbound versus d, 1 – d, (1 – d )2 (e) For a given load, Io, what should be the
and d(1 – d )2 for the converter. From the current rating of the BJT?
plots, comment on the value of Kbound for (f ) Why should duty cycle be less than 0.5? What
the converter. happens if duty cycle is greater than 0.5?
(g) What is the primary current waveshape for
6. The circuit shown in Figure 5.72 is a forward
duty cycle of 0.75? What will be the pri-
converter circuit. The transistor or BJT is oper-
mary current waveshape for duty cycle of
ated as a switch, that is, it is either in ON-state
0.75 if a practical transformer is used?
or OFF-state. Instead of a BJT, one may use a
(h) For a given load, Io, what should be the
MOSFET or IGBT. current ratings of the various diodes?
Mode of implementation: The above circuit is (i) What is the input–output voltage and cur-
studied by rent transfer ratios under steady state?
a. Simulation in SciLab 7. The circuit shown in Figure 5.73 is a flyback
b. Simulation in SPICE converter circuit. The transistor or BJT is
g
1:1:n
id1 iL io
d e
ip
id2 L
Vi ic
C RL Vo
a
c
b
PWM
Q id3
generator
Figure 5.72 Forward converter circuit.
Chapter 05.indd 280 3/19/2009 1:30:35 PM

id b io
ip 1:n
Vi ic
C RL Vo
a
PWM
Q
generator
Figure 5.73 Flyback converter circuit.
operated as a switch, that is, it is either in ON- (d) Keeping the duty cycle at 0.4, monitor the
state or OFF-state. Instead of a BJT, one may voltage waveform across the BJT. What is
use a MOSFET or IGBT. the effect of duty cycle on the waveshape?
(e) For a given load, Io, what should be the
Mode of implementation: The above circuit is
current rating of the BJT?
studied by
(f ) Is there a duty cycle limitation of 0.5? What
a. Simulation in SciLab
happens if duty cycle is greater than 0.5?
b. Simulation in SPICE
(g) Does increase in load cause increase in the
primary current?
Tasks for study: (h) As current flow in primary and secondary
(a) Rig up the above circuit with appropriate is mutually exclusive in time, how does
values for the components. loading affect the primary current and
(b) Observe the voltage waveforms across the hence primary power?
BJT, primary, secondary and output. (i) What is the transformer turns ratio?
(c) Observe the current waveforms through (j) What is the input–output voltage and cur-
the various components of the circuit. rent transfer ratios under steady state?

1. A chopper can be constructed using an 4. To handle all four quadrant operation,
switch. basic choppers are required.
2. The chopper does not contain an output 5. In any converter topology, the inductor should
. be placed in such a manner where its current will
not become at instant of operation.
3. A single basic chopper consisting of the
single pole double throw switch can handle 6. In any converter topology, the capacitor should
only quadrants of operation. be placed in such a manner that its
Chapter 05.indd 281 3/19/2009 1:30:36 PM

does not become discontinuous at instant of 20. If the current through the inductor does not
operation. become zero at any instant in a switching
period, then the converter operation is referred
7. The buck converter gives an output voltage
to as the .
that is always than the input voltage.
21. If the current through the inductor becomes
8. In a buck converter, the inductor current is
zero for some portion of the switching period
composed of the DC current part
then the converter operation is referred to as
and the AC current part.
the .
9. The capacitor value is calculated by applying
22. If the input and output voltages at any given
the rule.
instant are constant, then the inductor ripple
10. The area under the positive portion of the current amplitude is with load.
capacitor current waveform implies
23. For a particular converter, if the conduction
of the capacitor.
parameter value is greater than Kbound for the
11. The area under the negative portion of the converter at a given operating duty cycle, then
capacitor current waveform implies the converter is in .
of the capacitor.
24. For a particular converter, if the conduction
12. The boost converter gives an output voltage parameter value is lesser than Kbound for the
that is than the input voltage. converter at a given operating duty cycle, then
the converter is in .
13. The inductor current is composed of
which is the DC current part and 25. The buck-based isolated topologies consist of
which is the AC part. the output converter, the input side
and the transformer.
14. The diode current is composed of the DC
current part and the AC 26. In isolated converters, the circuit is
current part. needed to convert the DC input source to an
AC pulse voltage before feeding to the trans-
15. The buck–boost converter gives an output volt-
former.
age that is either or than
the input voltage. 27. In the forward converter, the flux in the core
increases during the DTs duration of
16. The inductor current is composed of the
the switching period.
values of the input and output current as the
DC part and ΔiL as the part. 28. In the forward converter with demagnetizing
winding the leakage inductance is minimized
17. The diode current is composed of the DC
by winding the primary and the demagnetizing
current part and the AC
windings in a fashion.
current part.
29. In the forward converter with primary and
18. The switch element in any converter is selected
demagnetizing windings having equal number
based on the i–v static characteristic
of turns, the maximum duty cycle is .
requirement.
30. The forward converter utilizes only the
19. Continuous conduction mode and discontinu-
half of the core magnetization as the magnetiz-
ous conduction mode are defined with respect
ing current and the core flux are .
to the .
Chapter 05.indd 282 3/19/2009 1:30:36 PM

31. The push–pull, half-bridge and full-bridge con- 37. For a given power, the average current flowing
verters magnetize the core in both through the primary of the full-bridge con-
directions to better utilize the core. verter is of that flowing through the
half-bridge converter.
32. The switching frequency of the primary side
switches is the frequency of the 38. The current-carrying capacity for the full-
inductor ripple current. bridge converter switches is that of
the switches of the half-bridge converter of the
33. In the half-bridge converter, the voltage across same output power rating.
the primary of the transformer is the
DC link voltage. 39. Flyback converter is an isolated converter that
is based on the converter topology.
34. The voltage developed across the OFF switches of
the half-bridge converter is maximum 40. The flyback converter has one of the
voltage. component count.
35. In the full-bridge converter, the voltage 41. In the high-boost converter, at a duty cycle of
across the primary of the transformer is the 0.5, the output voltage is .
. 42. In soft switching converters, the switching loss
36. The voltage developed across the OFF is .
switches of the full-bridge converter is 43. The ZCS and ZVS converters are called also
maximum voltage. called resonant converters.
1. What is duty ratio or duty cycle? ation wherein the power flow can be
bi-directional? Explain.
2. Can a chopper handle bi-directional power
flow? Illustrate with an application example. 10. In a buck converter, what is the rate at which the
inductor current ripple rises and falls?
3. Explain the operation of the dual chopper.
11. Consider the buck converter circuit sche-
4. What is volt-second balance?
matic of Figure 5.14 wherein the first-level
5. What is charge balance? non-idealities are included. What is the input–
output relation? By what factor is the input–
6. Is volt-second balance and charge balance appli-
output relation different from the input–output
cable under transient conditions? Explain.
relationship of the ideal buck converter?
7. Explain the operation of the buck converter What is the efficiency of the buck converter
and illustrate the operation with the inductor circuit with non-idealities?
current and the switch waveforms.
12. Explain the operation of the boost converter
8. What is inductor current freewheeling? Why and illustrate the operation with the inductor
does it occur? current and the switch waveforms.
9. Figure 5.13 gives the buck converter realiza- 13. Figure 5.17 gives the boost converter real-
tions. Can they be used for two-quadrant oper- izations. Can they be used for two-quadrant
Chapter 05.indd 283 3/19/2009 1:30:36 PM

operation wherein the power flow can be DTs (1 − D)Ts

bi-directional? Explain.
S1 S2
14. In a boost converter, what is the rate at which
the inductor current ripple rises and falls?
N1 L1 Vo
Vi C
15. Consider the boost converter circuit schematic
of Figure 5.18 wherein the first-level non-
N2 L2
idealities are included. What is the input–output
relation? By what factor is the input–output
relation different from the input–output rela-
tionship of the ideal boost converter? What is Figure 5.74 DC–DC converter topology for
the efficiency of the boost converter circuit with Question 19.
non-idealities?
20. For the DC–DC converter topology shown in
16. Explain the operation of the buck–boost con-
Figure 5.75, switch S1 is connected to the
verter and illustrate the operation with the
input during DTs duration and switch S2 is
inductor current and the switch waveforms.
connected to the output during (1 – D)Ts
17. Figure 5.21 gives the buck–boost converter duration of a switching period. N1 and N2 are
realizations. Can they be used for two-quadrant the number of turns on the tapped inductors
operation wherein the power flow can be bi- L1 and L2 wound on the same core. Using the
directional? Explain. volt-second and charge balance rule, derive the
steady-state input–output relationship. Plot
18. Consider the buck–boost converter circuit the idealized i–v characteristic requirement for
schematic of Figure 5.22 wherein the first-level the switches S1 and S2 and indicate the most
non-idealities are included. What is the input– appropriate power semiconductor switches
output relation? By what factor is the input– based on the static characteristic.
output relation different from the input–output
relationship of the ideal buck–boost converter? N1 N2 (1 − D)Ts
What is the efficiency of the buck–boost con-
verter circuit with non-idealities? L1 L2
S2
19. For the DC–DC converter topology shown
S1
in Figure 5.74, switch S1 is connected to the Vi C Vo
input during DTs duration and switch S2 is DTs
connected to the output during (1 – D)Ts
duration of a switching period. N1 and N2
are the number of turns on the tapped
inductors L1 and L2 wound on the same core. Figure 5.75 DC–DC converter topology
Using the volt-second and charge balance for Question 20.
rule, derive the steady-state input–output
relationship. Plot the idealized i–v character- 21. For the DC–DC converter topology shown in
istic requirement for the switches S1 and S2 Figure 5.76, a single pole triple throw switch
and indicate the most appropriate power (SPTT) is connected as shown. In a switching
semiconductor switches based on the static period, the pole P is connect to throw T1 for a
characteristic. duration D1Ts and is connected to throw T2
Chapter 05.indd 284 3/19/2009 1:30:36 PM

P D2Ts
L T2
T1
Vi T3 C Vo
D1Ts
Figure 5.76 DC–DC converter topology of Question 21.
for a duration D2Ts. For the remaining dura- 32. Explain the operation of the forward converter
tion of the switching period the pole P is con- and illustrate with inductor current, trans-
nected to throw T3. Using the volt-second and former primary, switch current and switch
charge balance rule, derive the steady-state voltage waveforms.
input–output relationship. Plot the idealized
33. For a forward converter, include the first-level
i–v characteristic requirement for the switches
non-idealities like the ON-state drops of the
of the SPTT switch and indicate the most
switches, transformer leakage, winding resis-
appropriate power semiconductor switches
tances and equivalent series resistance of the
based on the static characteristic.
output capacitor. Derive the input–output
22. What is conduction parameter? What is its sig- relationship using the volt-second balance and
nificance? charge balance rule.
23. Derive Kbound for the buck, boost and buck– 34. What are the advantages of the forward con-
boost converters and plot Kbound versus the verter with demagnetizing winding over the
duty cycle. diode–resistor reset topology?
24. Derive the input–output relationship for 35. Explain the operation of the forward converter
a buck converter operating in DCM. with demagnetizing winding.
25. Derive the input–output relationship for 36. What are the advantages of the dual switch for-
a boost converter operating in DCM. ward converter and explain its functioning?
26. Derive the input–output relationship for a 37. Explain the output inductor freewheeling pro-
buck–boost converter operating in DCM. cess in the case of the push–pull, half-bridge
and full-bridge converters. Draw the secondary
27. Why is isolation needed between the output
side diode current waveforms.
circuit and the input circuit?
38. What is flux walking in push–pull circuit?
28. What are the categories of isolated converters?
What is its effect in the functioning of the
29. What are the types of buck-based isolated con- circuit?
verters?
39. Explain the operation of the half-bridge
30. In forward converter, what is meant by core topology.
flux resetting?
40. Explain the operation of the full-bridge
31. What are the various methods of achieving core topology.
flux resetting?
Chapter 05.indd 285 3/19/2009 1:30:36 PM

41. How is flux walking problem solved in half- 48. Discuss the voltage rating of the IGBT and
bridge and full-bridge converters? diode switches of a high-boost converter.
42. Discuss the operation of the Cuk converter. 49. Can the inductors of the high-boost converter
be coupled together onto a common core? Dis-
43. What are the benefits and drawbacks of the
cuss.
Cuk converter with respect to the buck-based
converters? 50. What is the motivation for using soft switching
in converters?
44. How does one achieve zero ripple at the input
or output inductor currents of the Cuk con- 51. What are zero-current switching and zero-
verter? voltage switching?
45. Can both the input and the output inductor 52. Explain the ZCS and ZVS switch topologies.
current ripples be made zero simultaneously?
53. Explain the operation of ZCS and ZVS for
Explain.
boost and buck–boost converters.
46. What conditions should be satisfied for cou-
54. What are the four operative modes of the ZCS
pling the inductors of a converter onto to a
buck converter?
common core?
55. What are the four operative modes of the ZVS
47. Discuss the operation of the high-boost
buck converter?
converter.
| PROBLEMS
1. The input to a chopper is from a 100 V DC 5. A boost converter is supplied with an input
source. The chopper is switched at a frequency voltage that varies between 5 V and 10 V. The
of 100 kHz with a pulse width of 4 μs. What is output is required to be regulated at 15 V. Find
the average output voltage of the chopper? the duty cycle range.
2. A buck converter is supplied with an input 6. A boost converter that is switching at 50 kHz is
voltage that varies between 20 V and 30 V. The supplied with an input voltage that varies between
output is required to be regulated at 5 V. Find 5 V and 10 V. The output is required to be regu-
the duty cycle range. lated at 15 V. A load resistor of 15 Ω is connected
across the output. If the maximum allowable
3. A buck converter that is switching at 50 kHz is
inductor current ripple is 10% of the average
supplied with an input voltage that varies
inductor current, estimate the value of the induc-
between 20 V and 30 V. The output is required
tance to be used in the boost converter.
to be regulated at 5 V. A load resistor of 5 Ω is
connected across the output. If the maximum 7. For Problem 6, if the output voltage ripple
allowable inductor current ripple is 10% of the specifi cation is 1% of the output voltage,
load current, estimate the value of the induc- estimate the output capacitor value.
tance to be used in the buck converter.
8. A buck–boost converter is supplied with an input
4. For Problem 3, if the output voltage ripple voltage that varies between 5 V and 10 V. The
specification is 1% of the output voltage, esti- output is required to be regulated at 15 V. Find
mate the output capacitor value. the duty cycle range.
Chapter 05.indd 286 3/19/2009 1:30:36 PM

9. A buck–boost converter that is switching at 50 tance? Is the developed voltage within the
kHz is supplied with an input voltage that IGBT rating?
varies between 5 V and 10 V. The output is
required to be regulated at 15 V. A load resistor 16. A forward converter is switched at 50 kHz with a
of 15 Ω is connected across the output. If the duty cycle of 0.3. It is supplying a 50 W load at
maximum allowable inductor current ripple is an output voltage of 5 V. The input to the forward
10% of the average inductor current, estimate converter is derived from the 230 V mains by
the value of the inductance to be used in the using a capacitor input filter rectifier. A demagne-
buck–boost converter. tizing winding with the same number of turns as
the primary winding is used to reset the core flux.
10. For Problem 9, if the output voltage ripple What is the voltage developed across the primary
specification is 1% of the output voltage, esti- switch during the OFF-state?
mate the output capacitor value.
17. A push–pull converter is switched at 50 kHz
11. For a buck converter supplying a 10 A load, with a duty cycle of 0.4. It is supplying a 48 W
the inductor current ripple is designed to be load. The turn ratio of the center-tapped trans-
10% of the maximum load current. Calculate former is 5. The input to the push–pull con-
the load current at which the converter is at the verter is a 12 V battery. The output inductor
boundary of CCM and DCM. current ripple is 10% of the load current.
a. What is the output voltage?
12. A forward converter is switched at 50 kHz b. What is the peak current through
with a duty cycle of 0.3. It is supplying a the primary switch?
50 W load at an output voltage of 5 V. The c. What is the voltage developed across the
input to the forward converter is derived primary switch when it is OFF?
from the 230 V mains by using a capacitor
input filter rectifier. Estimate the turns ratio 18. A half-bridge converter is switched at 50 kHz
of the transformer. with a duty cycle of 0.4. It is supplying a 48
W load. The turn ratio of the center-tapped
13. For Problem 12, the peak magnetizing current transformer is 5. The DC link of the half-
is 20 mA. If the output inductor current ripple bridge converter is from a 12 V battery. The
is 10% of the average inductor current, calcu- output inductor value is 10 mH. The output
late the peak current requirement for the pri- inductor current ripple is 10% of the load
mary switch (IGBT). current.
14. For Problem 12, if the transformer core flux a. What is the output voltage?
reset is done by using a diode–resistor con- b. What is the peak current through the
nected across the primary, calculate the power primary switch?
dissipated in the flux reset resistor Rf if the c. What is the voltage developed across the
transformer is ideal. primary switch when it is OFF?
d. What is the average current of the sec-
15. For Problem 13, if the voltage rating of the ondary diodes?
IGBT is 1200 V, calculate the range of the flux e. What is the rms current of the secondary
reset resistor Rf . If the value of Rf is selected at diodes?
the midpoint of the obtained range, what is the f. What is the peak inverse voltage of the
voltage developed across the IGBT switch secondary diodes?
during the turn-OFF when (a) transformer is g. What is the range of value for the flux
ideal and (b) transformer has leakage induc- walking capacitor so that the DC voltage
Chapter 05.indd 287 3/19/2009 1:30:37 PM

developed across the capacitor does not g. What is the range of value for the flux
exceed 10% of the DC-link voltage? walking capacitor so that the DC voltage
developed across the capacitor does not
19. A full-bridge converter is switched at 50 kHz
exceed 10% of the DC-link voltage?
with a duty cycle of 0.4. It is supplying a 48 W
load. The turn ratio of the center-tapped trans- 20. A flyback converter is switched at 50 kHz with
former is 5. The DC link of the full-bridge a duty cycle of 0.4. It is supplying a 48 W load.
converter is from a 12 V battery. The output The turn ratio is 5. The input voltage of the
inductor value is 10 mH. The output inductor flyback converter is derived from a 12 V bat-
current ripple is 10% of the load current. tery. The flyback inductor value as viewed from
a. What is the output voltage? the primary side is 10 mH.
b. What is the peak current through the a. What is the output voltage?
primary switch? b. What is the peak current through
c. What is the voltage developed across the the primary switch?
primary switch when it is OFF? c. What is the peak current through the
d. What is the average current of the sec- secondary diode?
ondary diodes? d. What is the voltage developed across
e. What is the rms current of the secondary the primary switch when it is OFF?
diodes? e. What is the peak inverse voltage of the
f. What is the peak inverse voltage of the secondary diode?
secondary diodes?
| ANSWERS
Fill in the Blanks
1. SPDT 16. average; AC ripple 28. bifilar
2. filter 17. load; capacitor 29. 0.5
3. two 18. ideal 30. positive; unidirectional.
4. two 19. inductor current 31. positive and negative
5. discontinuous 20. continuous conduction 32. half
6. voltage mode 33. half
7. smaller 21. discontinuous conduction 34. DC link
8. load; capacitor mode 35. DC-link voltage
9. amp-second or charge 22. invariant 36. DC link
balance 23. continuous conduction 37. half
10. charging mode 38. less than
11. discharging 24. discontinuous conduction 39. buck–boost
12. larger mode 40. least
13. Iin; input; ΔiL; ripple 25. buck; chopper; isolation 41. infinite
14. load; capacitor 26. input chopper 42. zero
15. smaller; larger 27. linearly 43. quasi
Chapter 05.indd 288 3/19/2009 1:30:37 PM

DC–AC Switched-Mode Converters
Learning Objectives
CHAPTER
6
understand the different types of inverters.
distinguish between square wave and PWM operation and understand the effect of
PWM on the output harmonics.
obtain the Fourier series for a PWM wave pattern.
understand the operation of various inverter schemes.
D C–AC switched-mode converters are circuit configurations that convert DC into AC. They are popu-
larly called inverters. The inverters are used in many power electronic applications such as AC motor
controls, pumping applications, uninterruptible power supplies, electronic ballasts, electric vehicles, fre-
quency converters, induction heating and so on. An inverter is a circuit or system that delivers AC power
when energized from a source of DC power. Alternately, one may say that inverters are DC-to-AC convert-
ers providing the reciprocal function of rectification. Rectifiers change AC into DC whereas the inverters
behave in exactly the opposite manner by converting DC into AC.
DC–AC converters appear in many applications with different nomenclatures. However, they can be
classified into three broad categories, namely,
1. choppers;
2. oscillators;
3. inverters.
As discussed in the previous chapter, the choppers convert the input DC voltage to pulsed output. The out-
puts of the choppers have a non-zero average. The terms oscillators and inverters have been used inter-
changeably in the literature. The distinction between the two categories is somewhat arbitrary. The
distinction is more in connotation as there is no clear line of demarcation between oscillators and inverters.
In general, the term inverter is used where the emphasis is to provide AC power for some other circuit or
equipment. Here the focus is on parameters like efficiency, control, protections, etc. A circuit where the
emphasis is on producing an AC output as a reference signal, wherein the focus is on frequency stability and
drift, has traditionally been called an oscillator even though the operational principles for the oscillator and
the inverters are similar.
Chapter 06.indd 289 3/12/2009 6:27:39 PM

An inverter is an electronic circuit which converts a DC voltage source to an AC voltage source of

specified magnitude and frequency. The AC output waveform may be a square wave, quasi-square wave or
sine wave. The primary component in the inverter is the power semiconductor switch. The number of
switches can vary from one switch for a self-oscillating-type inverter to a very large number of switches for a
multi-level inverter. Based on the strategy used for driving the power semiconductor switches ON and OFF,
the inverters are broadly classified as
1. self-oscillating type;
2. driven type.
In the self-oscillating-type inverters the power switching elements are turned ON and OFF by the inherent
non-linear characteristics of the magnetic and the semiconductor devices. However, this type of inverter
cannot be easily controlled. In the driven type of inverter the switching elements are turned ON and OFF
by external control circuits. The outputs of the switched-driven inverters are amenable for control. In the
sections to follow, the switching inverters belonging to both these categories are discussed.
6.1 Inverter Topologies
T he inverter is composed of only power semiconductor switches and their drive circuits. Any of the con-
trolled power semiconductor switches discussed in Chapter 1 can be used. Until the advent of the fully
controlled switches like the power bipolar junction transistors (BJTs), metal oxide semiconductor field effect
transistors (MOSFETs) and insulated gate bipolar transistors (IGBTs), the inverters primarily consisted of
the thyristor-based switches. However, now the MOSFETs and IGBTs are more popular in the inverter
topologies due to control on both the turn-OFF and turn-ON of the device coupled with better switch
transition characteristics. The drive circuits for the inverter switches are based on the discussion of Chapter 2.
Topologically, the inverters can be classified into the single-phase inverters and the three-phase inverters
depending on whether they are feeding a single-phase load or a three-phase load. Both these classes of
inverter topologies will be discussed in the following sub-sections.
Single-Phase Topologies
Basic Two-Level Structure
The simplest inverter topology is composed of a single pole double throw (SPDT) switch. The self-oscillating
types of inverters use this single SPDT switch configuration. The driven types, however, range from the
simplest single SPDT configuration to more complex multiple throw switch configurations. Figure 6.1(a)
shows the schematic of the single SPDT configuration inverter. The voltage across the load at any given
instant is dependent on the switch position.
V1, V2 and V3 are derived from DC sources such that V2 < V3 < V1. If the pole of the SPDT switch is
connected to V1, then the output voltage Vo is V1 − V3; if the pole of the SPDT switch is connected to V2,
then the output voltage Vo is V2 − V3. There are only two levels that the output can assume and therefore
this is a two-level inverter. The topology of Figure 6.1(a) is a generic configuration. A specific configuration
using a single DC input source Vdc is shown in Figure 6.1(b). Here, V1 = Vdc, and V2 and V3 are connected
to the circuit ground. The output voltage is measured with respect to the circuit ground. The output voltage
Vo can assume either Vdc or 0 V depending on the switch pole position. Such a two-level inverter wherein
one of the output voltage levels is 0 V is called a chopper. As the output voltage swings between 0 and Vdc, it
contains a finite non-zero average value.
Chapter 06.indd 290 3/12/2009 6:27:40 PM

DC–AC Switched-Mode Converters 291
V1 V1
P P
S S
Load Vo Load Vo
V2 = 0
V3 = 0
V2 V3
Figure 6.1 Simple single SPDT switch topology: (a) Generic configuration; (b) inverter
configuration with V2 and V3 connected to the circuit ground.
Figure 6.2(a) is the implementation schematic of the two-level chopper topology depicted in Figure 6.1(b).
S1 and S2 are the two power semiconductor switches used to implement the SPDT switch. When S1 is ON,
Vo is at V1 and when S1 is OFF and S2 is ON, Vo is 0 V. For inductive loads, the internal body diode of the
IGBT provides the inductive energy freewheeling path. When S1 is ON and S2 is OFF, the load current of an
inductive load flows from the pole of the SPDT switch to the ground. When S1 is turned OFF and S2 is
turned ON, the inductive load current freewheels through the body diode of S2 till it reduces to zero.
Figure 6.2(b) shows the implementation schematic of the generic two-level topology. One end of the
load is connected to the pole “p” (midpoint) of S1 and S2 switch arm and the other end of the load is con-
nected to the midpoint of the capacitor arm as shown in Figure 6.2(b). The midpoint of the capacitor arm
is at potential V3. The output voltage can assume either of the two levels (Vdc − V3) or (−V3). Usually the
capacitor values for C1 and C2 are the same and therefore V3 = Vdc/2. In such a case the two levels of the
inverter output voltage are Vdc/2 and −Vdc/2.
V1 = Vdc
V1 = Vdc
S1 S1 C1
Vo
P P V3
Load
Load Vo C2
S2 S2
V2 = 0
(a) (b)
Figure 6.2 Implementation schematic of (a) the chopper; (b) a two-level inverter.
Chapter 06.indd 291 3/12/2009 6:27:40 PM

Multi-Level Structures
To have better control on the harmonic content of the output voltage waveform, it is desirable to have the
output voltage that can assume many more levels than just two. In such cases, the simple two-level inverter
topology will not suffice. One will have to use more complex configurations to achieve multi-level outputs.
The number of switches for multi-level converters will increase depending on the number of levels required.
Figure 6.3(a) gives the topology of a generic four-level inverter. The output voltage can assume four different
levels depending on the switching pattern of the two SPDT switches. Table 6.1 gives the output levels in
relation to the switching pattern.
Figure 6.3(b) gives an implementation schematic of Figure 6.3(a) wherein there is only one input DC voltage
source. Here V1 = V3 = Vdc and V2 = V4 = 0 V. Based on these constraints, Table 6.1 is modified as Table 6.2
From Table 6.2, it can be observed that with a single input DC voltage source, the four-level structure of
Figure 6.3(a) reduces to a three-level inverter wherein the output voltage can assume only Vdc, 0 and −Vdc.
V1 = V3 = Vdc
V1 V3
S1 S3
V0
S1 S2
Load Load
S2 S4
V2 V4
V2 = V4 = 0
(a) (b)
Figure 6.3 Generic four-level topology; (b) implementation schematic of a three-level structure
by applying single input DC voltage source to the generic four-level topology.
Table 6.1 Output voltage levels in relation to the switching pattern

S1 connected to S2 connected to Output voltage
V1 V4 V1 − V4
V1 V3 V1 − V3
V2 V4 V2 − V4
V2 V3 V2 − V3
Chapter 06.indd 292 3/12/2009 6:27:40 PM

Table 6.2 Output voltage levels with single input DC voltage source
Vdc 0 Vdc
Vdc Vdc 0
0 0 0
0 Vdc −Vdc
The schematic of Figure 6.4(a) shows a single pole triple throw (SPTT) switch S wherein the pole P is
connected to one end of the load and the throws are connected to DC voltage sources V1, V2 and V3. The
other end of the load is connected to another DC voltage source V4. Figure 6.4(b) shows the schematic
wherein the SPTT switch is implemented as two SPDT switches. This forms a three-level arm.
The output voltage can assume three different levels depending on the switching pattern of the two
SPDT switches. Table 6.3 gives the output levels in relation to the switching pattern.
The three-level arm can be applied on either side of the load as shown in Figure 6.5(a). Figure 6.5(b)
shows the implementation with only SPDT switches. This forms the three-level arm bridge topology.
V1
(V1 or V2)
V1
S2
P (V1 or V2 or V3) P (V1 or V2 or V3)
S V2
V2 S1
Load V0 Load V0
V3 V4 V3 V4
(a) (b)
Figure 6.4 Three-level arm: (a) SPTT arm; (b) SPTT switch replaced by two SPDT switches.
Table 6.3 Output voltage levels in relation to the switching pattern for schematic of Figure 6.4
Pole of S2 V1 V1 − V4
Pole of S2 V2 V2 − V4
V3 V1 or V2 V3 − V4
Chapter 06.indd 293 3/12/2009 6:27:40 PM

V1 V4
V1 V4
S1 S3
V0 V0
Sa Sb
V2 V5 V2 S2 S4 V5
Load
Load
V3 V6 V3 V6
(a) (b)
Figure 6.5 Single-phase three-level arm bridge: (a) Schematic; (b) SPTT switches replaced by
SPDT switches.
The output voltage can assume nine different levels depending on the switching pattern of the SPDT
switches. Table 6.4 gives the output levels in relation to the switching pattern.
To maintain symmetry if V1 = V4, V2 = V5 and V3 = V6, from Table 6.4 we have V1 − V4, V2 − V5 and V3 − V6
are zero voltage levels and therefore the inverter output can assume only seven levels. Figure 6.6 shows the circuit
implementation of the multi-level inverter topology described in Figure 6.5(b).
The three-level arm can be extended to an n-level arm and applied to both sides of a load to form an
n-level arm bridge. This two-arm n-level bridge inverter is depicted in Figure 6.7. In general, for a bridge arm,
Table 6.4 Output voltage levels in relation to the switching pattern for schematic of Figure 6.5
S1 connected to S2 connected to S3 connected to S4 connected to Output voltage
V1 Pole of S1 V4 Pole of S3 V1 − V4
V1 or V2 V3 V4 Pole of S3 V3 − V4
V1 or V2 V3 V5 Pole of S3 V3 − V5
V1 Pole of S1 V4 or V5 V6 V1 − V6
V2 Pole of S1 V4 or V5 V6 V2 − V6
V1 or V2 V3 V4 or V5 V6 V3 − V6
Chapter 06.indd 294 3/12/2009 6:27:41 PM

V1 V4
S11 S31
S12 S21 S41 S32

V0
V2 Load V5
S22 S42
V3 V6
Figure 6.6 Circuit implementation of the multi-level inverter topology of Figure 6.5(b).
if there are n-input DC sources then there are n possible levels that the arm pole can assume. The n sources
are connected to the throws of a single pole n-throw switch. To build an equivalent single pole n-throw
switch, n − 1 SPDT switches are needed.
For the two-arm n-level bridge inverter topology as shown in Figure 6.7, the maximum number of
possible levels for each arm is n. The maximum number of levels that the output can assume is n ⋅ n = n 2.
It would be n2 only if all the voltages on both sides of the load are unique. However, if the arms are sym-
metric in voltages, then the combinations wherein the same n voltages are applied to both bridge arms in a
symmetrical manner will lead to zero load voltage. Under this condition, there will be n possibilities wherein
the voltages across the load is zero. Thus, out of the possible n2 possibilities, n possibilities are removed and
combine to form a zero voltage level. Therefore, the number of possible output voltage levels in this case is
n2 − n + 1 .
V11 V21
V12 V22
V0
S1
Load
V13 V23
V1n V2n
Figure 6.7 Two-arm n-level arm bridge inverter topology.
Chapter 06.indd 295 3/12/2009 6:27:41 PM

If “m” is the required number of output voltage levels, then
m = n2 − n + 1
Therefore the number of distinct voltage sources required is
1 ± 4m − 3
n= (6.1)
2
The number of distinct voltage sources should be an integral number; therefore, the solution of Eq. (6.1)
should be a positive integer. If both the solutions of the quadratic are not appropriate then the required
number of levels “m” specified is inappropriate.
Three-Phase Topologies
The previous sub-section discussed the inverter topologies for a single-phase load that is supplied, in gen-
eral, by a two-arm bridge wherein the single-phase load is connected in between the two arms of the
inverter. In the case of a three-phase inverter, the single-phase inverter is extended to three-phase loads by
including one more bridge arm. Thus, three-phase inverter topologies consist of three bridge arms wherein
each arm can be of arbitrary level. However, practical inverters consists of three bridge arms wherein all the
bridge arms are identical and the throws of the single pole multi-throw switches are supplied by the same
voltage sources across all three bridge arms. Figure 6.8(a) shows a two-level three-arm bridge converter
supplying a three-phase load. Figure 6.8(b) shows the circuit implementation. The three-arm bridge can
either supply a three-phase star-connected load as shown in Figure 6.9(a) or a three-phase delta-connected
load as shown in Figure 6.9(b). Instead of a two-level arm, one may synthesize a multi-level three-phase
inverter by using a multi-level structure for the arms of the inverter along lines discussed for the single-phase
inverter topologies.
Vdc
S1 S3 S5
c
b N
Sa Sb Sc
a
N S2 S4 S6
(a) (b)
Figure 6.8 (a) Inverter with three-arm bridge; (b) implementation schematic of the three-arm
bridge topology.
Chapter 06.indd 296 3/12/2009 6:27:41 PM

Vdc Vdc
S1 S3 S5 S1 S3 S5
a a
b N b
c c
S2 S4 S6 Star S2 S4 S6 Delta
load load
(a) (b)
Figure 6.9 Three-arm bridge connected to (a) three-phase star load; (b) three-phase delta load.
6.2 Self-Driven Inverters
T he self-driven inverters are a class of DC–AC converters wherein the power semiconductor devices are
operated without externally applied control signals to the bases or gates of the devices. Based on the
operation mode of the power semiconductor devices, they can be broadly classified as follows:
1. Power Semiconductor Devices Operated in the Linear Region: These types are used as oscillators for
reference generation, linear audio applications, etc. wherein the signal power is very low.
2. Power Semiconductor Devices Operated Alternatively in Saturation and Cut-Off Regions: These
types are called switching self-driven inverters. They are used for supplying power to power supplies that
need voltage translation.
In this section, only the switching type of self-driven inverters will be discussed, wherein the power semicon-
ductor switches are driven to saturation and cut-off alternately.
Generally, in inverters the transformers are operated in the linear magnetic region. The saturation of the
magnetic core is usually avoided because the operation in this region greatly increases the hysteresis loss
component that deteriorates the efficiency. Though the transformer driven into magnetic saturation becomes
a source of power dissipation, the combination of a transformer driven to saturation and power transistors
can produce an inverter with reasonably high efficiency as well as high reliability and low component count.
Core saturation is the mechanism responsible for switching and timing in these types of inverters.
The important features of saturable-core inverters are:
1. Switching transitions occur due to the characteristics of the magnetic core material and the static char-
acteristics of the power transistors. Core saturation alone cannot account for the operation, nor can
only the transfer characteristics of the power transistor.
Chapter 06.indd 297 3/12/2009 6:27:42 PM

2. Despite positive feedback, the oscillation is not of the LC type.

3. The frequency of the oscillation of such an inverter is dependent on the magnetic characteristics of the
core material and is independent of transistor parameters and temperature in an ideal circuit. The ideal
circuit also exhibits zero regulation. However, in a practical circuit, the frequency of oscillation is depen-
dent on the transistor parameters and temperature too.
4. The ideal saturable-core inverter generates a frequency that is directly proportional to the input DC
voltage. The wave shape is a symmetrical rectangular or square wave with a 50% duty cycle.
Saturable-Core Inverter
The circuit schematic shown in Figure 6.10 is a self-driven inverter wherein the oscillations are induced due
to the non-linear characteristics of the magnetic core.
Circuit Operation
When Vin is applied at switch ON of the inverter, current flows through R either through the path Nb2–R2–Q2
(base–emitter) or through the path Nb1–R1–Q1 (base–emitter). The path which is decided at start-up depends
on the characteristics of Q1 and Q2. It should be noted that Q1 and Q2 will never have identical characteristics.
R1
Nb1
Q1
DF1
Np1
− + Vo
Ns
Np2
D Vin
DF2
Q2
Nb2
R2
Figure 6.10 Saturable-core inverter topology.
Chapter 06.indd 298 3/12/2009 6:27:42 PM

Let Q2 have characteristics such that its Vce is slightly lower than that of Q1 for a given base current. Then at
switch ON, current from Vin flows through R and Nb1–R1–Q1 (base–emitter) and Nb2–R2–Q2 (base–emitter).
If Nb1 = Nb2 and R1 = R2, at start both transistors get equal base currents. However due to the non-identical
characteristics of Q1 and Q2 as stated earlier, Vce2 will be lesser than Vce1. The voltage across Np1 is Vin − Vce1
which tries to make the non-dot end positive. The voltage across Np2 is Vin − Vce2 and it tries to make the dot
end positive. As Vce2 is lesser than Vce1, the dot end of the windings will tend to become more positive. As a
consequence, from the depicted “dot” polarity, Nb2 “dot” is more positive compared to Nb1 non-dot end. This
tends to turn Q2 more towards conduction and Q1 more towards the OFF-state. This re-generatively builds up
until Q2 is in saturation and Q1 is in cut-off state. At this point the base current takes the path Nb2–R2–Q2
(base–emitter)–D.
When Q2 is in saturation, Vp2 (the voltage across Np2 is a constant) of magnitude Vin (neglecting the
saturation drops across the collector–emitter of the transistors), the current through the winding Np2, that
is, ic2, keeps increasing linearly as shown in Figure 6.11 because
dic2
Vin = Lp2
dt
VNb2
ib2
ic2
2Vin
Vce2
tsat
Figure 6.11 Waveforms for transistor Q2.
Chapter 06.indd 299 3/12/2009 6:27:42 PM

where Lp2 is the inductance of winding Np2.

As the collector current increases, the transformer begins to saturate at time tsat. As a result the current
ic2 starts to shoot up as indicated in Figure 6.11 due to reduction in permeability. The induced voltage across
Nb2 reduces as the transformer core goes into saturation, thereby decreasing the base drive for Q2. This
causes Q2 to come out of saturation region. A reduction in induced voltage at the dot end of the winding
implies an equivalent increase at the non-dot end of the base windings. The Nb1 polarity is such that it is
180o out-of-phase with respect to Nb2. Therefore Q1 tries to cut-in. This process re-generatively builds up
until Q1 goes to saturation and Q2 goes OFF.
The above process is repeated with transistor Q1 also and this continues indefinitely, thereby giving rise
to oscillations. The waveforms for Q1 are similar to that of Q2 as shown in Figure 6.11. However, Q1 wave-
forms are 180o out-of-phase with respect to the waveforms of Q2.
Oscillation Frequency
Much insight into the behavior of the saturable-core inverter can be gained from the basic equation defining
its frequency of oscillation. From the theory on magnetics (discussed in Chapter 7), it can be shown that
E = 4 K f NAc Bm f (6.2)
where E is the root mean square (rms) value of the induced voltage; Kf the form factor; f the frequency in
Hz; N the number of turns in the winding; Ac the core cross-section area in m2; Bm the maximum operating
flux density in Tesla. Rearranging Eq. (6.2), the oscillation frequency can be determined from
E
f = (6.3)
4 K f NAc Bm
For this particular application of saturable-core inverter, the waveform is a square waveform and therefore
Kf = 1 and E = Vin. Further, Bm (the maximum flux density) has to be chosen equal to the material satura-
tion flux density Bsat because it is required that the core saturates every switching cycle. Therefore Bm = Bsat
for these inverters.
Design Equations
In designing the saturable-core inverter, first determine the transistor ratings and select a suitable transistor:
Icm > Iin (6.4)
where
Iin = Pin/Vin; Pin = Po/h
h is the efficiency of the transformer − 80% efficiency is a conservative value that can be used for design; Icm
is the continuous maximum rating of the transistor. Referring to the waveforms shown in Figure 6.11, the
voltage rating for the transistor is given as
Vceo > 2Vin (6.5)
Chapter 06.indd 300 3/12/2009 6:27:42 PM

A transistor should be chosen such that the inequalities in Eqs. (6.4) and (6.5) are satisfied. After having
chosen the transistor, decide on the turns ratio for the base–primary winding such that,
Vb N b2 N b1
nb = = = (6.6)
Vin N p2 N p1
where Vb is the voltage across the base winding (taken to be around 3 V). The transformer is designed using
the area product technique that will be discussed in Chapter 7. The area product for this transformer is
derived in Appendix I. It is given as follows:
Po ⎛ 2 2nb ⎞
Ap = ⎜ + + 1⎟ (6.7)
4 K w JBm f s ⎜ η ηhFE min ⎟
⎝ ⎠
where Kw is the window factor = 0.4; J the current density which is typically 3 × 106 A/m2
Bm = Bsat of the material; fs the oscillation frequency; Po the output power; h the efficiency of the inverter
which is usually around 0.8. The number of turns is given as follows [refer to Eq. (6.2)]:
Vinmax
N p1 = (6.8)
4Bm Ac f s
N p2 = N p1
⎛ Vo ⎞
N s = ⎜⎜ ⎟⎟ N p1 (6.9)
⎝ Vinmin − Vcesat ⎠
where Vo is the secondary voltage and
N b1 = N b2 = nb N p1 (6.10)
The resistors R1 and R2 can be calculated as follows:
Vb − Vbesat − VD
R1 = R2 = (6.11)
Ib
where
I in
Ib =
hFEmin
and dissipation in R1 and R2 are
PR1 = PR2 = Ibrms2R1 (6.12)
Chapter 06.indd 301 3/12/2009 6:27:42 PM

where
Ib
I brms =
2
The resistor R is required only at the time of starting the oscillations, when it is required to just cut-in one
of the transistors till the regenerative action takes over. The value of R is given by
Vinmin − Vbe-cutin
R= (6.13)
I b-cutin
where
⎛ I ⎞
I b-cutin = 0.2 ⎜⎜ in ⎟⎟
⎝ hFEmin ⎠
and the power dissipation is given by
(Vinmax + VD )2 (6.14)
PR =
R
Saturable Base Drive Inverter

In the previous section, the output transformer is designed such that it goes into saturation region and, hence
compared to a transformer which operates in the linear region, the hysteresis loss is much greater. Further the
power switches have to handle high peak currents at the time of core saturation. To avoid the short comings,
the output transformer is designed to operate in the linear region, but another transformer is used for base
drives of the power switches, as indicated in Figure 6.12, which is designed to go into saturation region. As a
R1
Q1
DF1
R3
R
Np1
Nb1
− +
Npb Ns Vo
Vin Np2
Nb2 Vb D
DF2
Q2
R2
Figure 6.12 Self-driven inverter with saturable base drive transformer.
Chapter 06.indd 302 3/12/2009 6:27:43 PM

consequence, better efficiency is obtained using this two-transformer technique. The operation is exactly simi-
lar to the single transformer saturable-core transformer inverter, except that for the base drive, a separate
transformer is used which is saturable and that for the power output the transformer is designed to operate in
the linear region.
The output transformer is designed as a push–pull transformer and the base drive transformer is designed
along similar lines except that Bm is taken equal to Bsat of the core material. Here,
Vb − Vbe − VD
R1 = R2 = (6.15)
Ib
where Vb is the voltage developed across Nb1 and Nb2 of the saturable-core base drive transformer. The col-
lector current is given as
P
I c = I in = o (6.16)
ηVin
and the base current Ib is given as
Ic
Ib = (6.17)
hFEmin
The following points should be noted while designing the base drive transformer.
1. The VA capacity is given as PT = VbIb.
2. It is usual to set VNpb, that is, voltage across primary of the base drive transformer, to Vin. As Q1 and Q2
collector–emitter voltages swing between 0 and 2Vin, the resistor R3 also drops a voltage of Vin across itself.
3. Npb is calculated using the Faraday’s Law given by
Vin
N pb = (6.18)
4Bm Ac f s
4. The secondary turns of the base drive transformer are given by
⎛V ⎞
N b1 = N b2 = ⎜⎜ b ⎟⎟ N pb (6.19)
⎝ Vin ⎠
The value of Vb can be chosen to be around 3 V.

5. The primary current of the base drive transformer is given as
PT
I Npb = (6.20)
Vin
and
Vin Vin2
R3 = = (6.21)
I Npb PT
At higher input voltages, Npb may be large. The turns may be reduced in the base drive transformer by
adding a low-voltage winding in the output transformer as shown in Figure 6.13.
Chapter 06.indd 303 3/12/2009 6:27:43 PM

R1
Q1
DF1
R3
Nb1 Np1 Ns
Npb − +
Vo
Vin
Nb2 Vb D Np2
DF2
Q2
R2
Figure 6.13 Low-voltage base drive transformer with additional output transformer winding.
6.3 Driven Inverter
I n the case of driven inverters, the power switches are driven through external control circuits where the
ON-time and the OFF-time are controllable. Because of the controllable nature of the power switch
states, any desired output wave shape, amplitude and frequency can be easily achieved. There are various
control strategies through which one could achieve either a quasi-square wave output or a waveshape with
specific harmonic minimization or even a sinusoidal waveshape.
For any control strategy, the power circuit configuration used is generally one of the topologies discussed
in Section 6.1. The following three configurations are commonly used for the driven inverter circuits:
1. the push–pull configuration;
2. the half-bridge configuration;
3. the full-bridge configuration.
Push–Pull Configuration
The push–pull configuration deviates from the other configurations in that the load is connected across the
throws of the SPDT switches rather than at the pole. The schematic is shown in Figure 6.14(a). It consists
of a SPDT switch with split load as shown. The split load is connected across the throws of the switch S. The
input voltage source Vi is connected between the split load and the pole of the switch S. Figure 6.14(b) gives
the schematic of the push–pull topology wherein the load is replaced by a transformer with the actual load
Chapter 06.indd 304 3/12/2009 6:27:43 PM

P S1
S Load Vo
Vi Vi
S2
(a) (b)
Q1
DF1
− + B
Vo
Vin
C
DF2
Q2
(c)
Figure 6.14 (a) Generic push–pull configuration with split load; (b) push–pull configuration with
transformer load; (c) circuit implementation of the push–pull configuration with
transformer load.
connected in the secondary of the push–pull transformer. The SPDT switch is replaced by two single pole
single throw (SPST) switches S1 and S2. Figure 6.14(c) gives the circuit implementation of the push–pull
DC–AC inverter wherein the switches S1 and S2 of Figure 6.14(b) are replaced by power semiconductor
switches Q1 and Q2, respectively. The switches Q1 and Q2 can be BJTs or MOSFETs or IGBTs.
Consider the circuit shown in Figure 6.14(c) wherein the load is inductive. Let Q2 be in the ON condi-
tion. Then VBC = Vin and Vce1 = VAC = 2Vin. If Q2 is switched OFF, VBC becomes negative because di/dt is
negative as the current in the winding is cut-off and causes the voltage polarity reversal due to Ldi/dt effect.
By magnetic induction, VBA is now positive which acts against Vin, causing the magnetic energy stored in the
inductor to freewheel through the path AB–Vin–DF1–AB.
Chapter 06.indd 305 3/12/2009 6:27:43 PM

Q1
drive
Q2
drive
2Vin + VD
2Vin
VCE of Vin
Q2
VCE of
Q1
Freewheeling action of DF1
Figure 6.15 Typical waveforms for the push–pull configuration.
Now when Q1 is turned ON, VBA = Vin and Vce2 = 2Vin. On switch-OFF of Q1, as discussed before, the
magnetic energy freewheels through the path CB–Vin–DF2–CB. Figure 6.15 illustrates the operation of the
push–pull converter by depicting the typical waveforms.
The ratings for the power switching devices Q1 and Q2 are as follows:
Vceo rating > 2Vin
Icm rating > Po/(hVin)
where Po is the output power and h is the efficiency of the transformer.
Half-Bridge Configuration
The half-bridge inverter configuration is based on the generic two-level inverter topology of Figure 6.2. The
implementation circuit of the half-bridge inverter is shown in Figure 6.16. Generally C1 = C2 and R1 = R2 in
half-bridge inverter circuits. This means that the potential at point B is Vin/2.
Chapter 06.indd 306 3/12/2009 6:27:44 PM

Vin
Q1
D1 C1 R1
A B
Q2
D2 C2 R2
Vo
Figure 6.16 Circuit schematic of the half-bridge inverter configuration.
Circuit Operation
Let Q1 be ON to start with. The current flows from A to B in the primary of the load transformer. VAB is
positive. On switching OFF Q1, VBA becomes positive because the equivalent primary inductance present
between A and B will cause point B to be more positive with respect to point A due to Ldi/dt effect. The
inductive current freewheels through the path AB–C2–D2–AB, where the current is forced against the volt-
age across C2 (i.e., Vin/2) and the path AB–C1–Vin–D2–AB, where the current is forced against Vin − VC1
(i.e., Vin/2).
Similarly when Q2 is ON, the current flows from B to A, VBA is positive. On switching OFF Q2, VAB
becomes positive because voltage across the equivalent primary inductance between the points A and B will
reverse polarity according to Faraday’s Law (v = Ldi/dt) due to cutting off the current through the switch Q2.
The current freewheels through the paths BA–D1–C1–BA and BA–D1–Vin–C2–BA.
The power switching transistors should be selected such that
Vceo rating > Vin
and
Icm rating > Po/(hVin/2)
R1 and R2 are provided to divide the DC-link voltage Vin equally between the two capacitors C1 and C2.
Generally, R1 and R2 are designed by allowing about 2% of the full load to flow through them. Therefore,
Vin / 2
R1 = R2 =
0.02 I cm
Chapter 06.indd 307 3/12/2009 6:27:44 PM

Full-Bridge Configuration
The full-bridge inverter configuration is based on the generic four-level two-arm bridge topology. If there is
only one input voltage source, then as discussed in Section 6.1, the full-bridge inverter configuration reduces
to a three-level two-arm structure. The circuit implementation of the three-level two-arm full-bridge inverter
is shown in Figure 6.17.
Circuit Operation
To start with, let Q1 and Q4 be turned ON. VAB is positive and current flows from A to B in the primary of
the load transformer. On switching OFF Q1 and Q4, VBA becomes positive due to the sudden break in the
current and consequent large negative di/dt that causes reversal of the voltage polarity across the primary of
the transformer. This forces the current to freewheel through the path AB–D3–Vin–D2–AB against the DC-
link voltage, Vin. Similarly, when Q3 and Q2 are turned ON, VBA is positive and current flows from B to A.
On switching OFF Q3 and Q2, VAB becomes positive due to the inductance present between points A and
B. This forces the current to freewheel through the path BA–D1–Vin–D4–BA against the DC-link voltage
Vin. Here the transistors are selected such that
Vceo rated > Vin
Icm rated > Po/(hVin)
It should be noted that in the case of a half-bridge configuration the voltage across points A and B (the cen-
tral load arm) swings between +Vin/2 and –Vin/2, whereas in the case of a full-bridge configuration, the
voltage across the points A and B swings between +Vin and –Vin. Therefore for a given power Po, the current
rating required for the power devices in the half-bridge configuration is double that for the full-bridge con-
figuration. Therefore, for higher powers, the full-bridge configuration is generally preferred.
Vin
Q1 Q3
D1 D3
A B
Q2 D2 D4 Q4
Vo
Figure 6.17 Circuit schematic of the full-bridge inverter configuration.
Chapter 06.indd 308 3/12/2009 6:27:44 PM

6.4 Quasi-Square Wave Inverter
A waveform of the form shown in Figure 6.18 is called a “quasi-square wave”. This is a waveshape that is
easily achieved in a simple manner and hence it is popular in low-power, low-quality back-up power
supplies. The quasi-square waveform has a large harmonic content and hence it is not suitable for trans-
former or motor loads as it would heat up the core material. However, it has its niche application as a back-
up power supply for loads that have the rectifier–capacitor filter as the frontend.
An inverter which generates an output waveshape of the form shown in the Figure 6.18 is called a
“quasi-square wave” inverter. Any of the basic configurations (push–pull, half-bridge or full-bridge) can be
used with appropriate base drive signals for the power switches to generate the quasi-square wave. The
parameter of the quasi-square wave is the dead band angle q as indicated in Figure 6.18.
Circuit Operation
With reference to a full-bridge configuration like that shown in Figure 6.17, the drive signals for the power
switches Q1, Q2, Q3 and Q4 to generate a quasi-square waveform are as shown in Figure 6.19. The switches
Q1 and Q2 are switched 180° out-of-phase with respect to each other and Q3 and Q4 are similarly switched
180° out-of-phase with respect to each other. However the switching pattern for Q3 and Q4 are shifted in-
phase by an angle a with respect to the switching pattern of Q1 and Q2. The angle a is twice the dead band
angle q. When Q1 and Q4 are ON, then VAB is positive as shown in Figure 6.19. When Q4 goes OFF, Q3
would turn ON. During this time, Q1 and Q3 will be ON. This will ensure that the voltage VAB is zero.
If the load is inductive, the current will freewheel through Q1 and D3. When Q3 and Q2 are ON, VAB is neg-
ative, as indicated. When Q3 turns OFF, Q4 will become ON. During this time, Q2 and Q4 are ON ensuring
zero voltage across VAB. The current now freewheels through Q2 and D4 if the load is inductive.
Output Regulation
To perform the output voltage regulation, the output voltage is sensed, rectified, filtered and processed. This
filtered DC voltage is used for controlling the output voltage. The output quantity that can be controlled is
one of the following:
1. output voltage flat top peak value;
2. output voltage half-cycle average value;
3. output voltage rms value.
f
V
q 2q
wt
q
−V
wT
Figure 6.18 Quasi-square waveform.
Chapter 06.indd 309 3/12/2009 6:27:45 PM

Q1
drive
Q2
a
drive
Q3
drive
Q4
drive
VAB
Figure 6.19 Drive signals for quasi-square waveform generation.
If the duty ratio of the rectified output waveform is D, then

Vo-avg = VD
Vo-rms = V D
Vo-avg = Vo-rms D
Depending on the quantity to be controlled, the reference is the desired flat top peak value or the aver-
age value or the rms value. The rectified and filtered output waveform is accordingly processed depending on
the control requirement.
The Fourier series representation for the quasi-square waveform shown in Figure 6.18 is given as
4 ∞ ⎛V⎞
f (ω t) = ∑ ⎜ ⎟ cos(nθ )sin(nω t)
π n =1 ⎝ n ⎠
(6.22)
where n is the harmonic number. It can be observed from Eq. (6.22) that the output voltage can be con-
trolled in two ways:
Chapter 06.indd 310 3/12/2009 6:27:45 PM

Main / grid
AC−DC
rectifier
Vref DC-to-DC DC-link

+ Controller Inverter
− converter to inverter
Rectifier
Attenuator
and filter
Figure 6.20 Quasi-square wave regulation based on DC-link voltage control.
1. As the amplitude is a function of the DC link, the output can be controlled by controlling the DC-link
voltage.
2. As the amplitude is also a function of the angle q, the output can be controlled by controlling q, which
is also called variable duty cycle control.
The block schematics for both these types of control are shown in Figures 6.20 and 6.21, respectively.
Figure 6.20 shows the block schematic of a DC-link control for a quasi-square wave inverter. Here the aver-
age of the output is controlled by controlling the DC-link voltage. The output voltage is rectified and
filtered to obtain the average value of the output voltage waveshape. This voltage is compared with a desired
reference value for the output average, Vref. The error is passed through a controller like a proportional or a
proportional–integral controller which is used to modulate the duty cycle of the switches in the DC–DC
converter. The DC–DC converter gets its energy input from the output of a grid/mains rectified capacitor-
filter circuit. The output voltage of the DC–DC converter is varied through duty cycle control which in turn
feeds the quasi-square wave inverter, thereby controlling the average value of the output quasi-square wave.
In Figure 6.21 the block diagram for duty cycle control is shown, wherein the angle a is varied. The
control of this angle is obtained by comparing a DC signal, which in this case is the controller output with
Vref Drive for

a angle
+ Controller inverter Inverter
− generation
switches
Rectifier
Attenuator
and filter
Figure 6.21 Quasi-square wave regulation based on duty cycle control.
Chapter 06.indd 311 3/12/2009 6:27:45 PM

a = 2q Sawtooth
Controller carrier
output
Q1
drive
Q2
drive
Q3
drive
Q4
drive
Figure 6.22 Drive signal generation by duty cycle control for a quasi-square wave full-bridge inverter.
a ramp as shown in the Figure 6.22. The a angle generation block converts the controller output to duty
cycle variations and generates the drive signals for the switches of the full-bridge inverter.
In the case of the DC-link control, the power is handled by both the DC–DC converter and the inverter.
Therefore, this method needs two power blocks of the same rating. Further, this also implies that there are
losses in the DC–DC converter and also the inverter, thereby giving reduced efficiency as compared to the
variable duty ratio control wherein there is only one power block. Thus, the variable duty cycle control is
more common in practice.
6.5 Three-Phase Inverter Topologies
T he three-phase inverter topologies require three-arm bridge structures. They can be built using three two-
level single-arm bridge topologies as shown in Figure 6.2(b). Consider three two-level single-arm bridge
configuration inverters joined together as shown in Figure 6.23. Merging the input supplies of the three
single-phase bridges together, one has the system as shown in Figure 6.24 wherein the load becomes a three-
phase star topology. The junction of the three-phase star load is the neutral point N. The neutral (N) of the
load is connected to the center of the split supply (O) which is called the pole voltage. Therefore, it may be said
that the load neutral is connected to the pole voltage. In Figure 6.24, in is the neutral current. However, in a balanced
Chapter 06.indd 312 3/12/2009 6:27:45 PM

+
− −
A B
O O
+
+
− − −
+
− C
Figure 6.23 Three two-level single-arm bridge configurations used to form a three-phase inverter.
three-phase system, the neutral current will be zero and therefore reduces to the configuration as shown in
Figure 6.25. In the configuration shown in Figure 6.25, the load neutral point is floating wherein the neutral
current is forced to zero for a star load. The load may be a star or an equivalent delta load.
Referring to Figure 6.24, the potential at “N” with respect to “O” can be determined by knowing the
potentials of A, B, and C with respect to O:
VAN = VAO − VNO (6.23)
+
−
N
O B
in
C
+
−
Figure 6.24 Three-phase inverter topology with neutral connected to pole (center of split supply).
Chapter 06.indd 313 3/12/2009 6:27:46 PM

Q1 Q3 Q5
C
+ B
VCC N
− A
Q2 Q4 Q6
Figure 6.25 Three-phase inverter topology with neutral floating.
VBN = VBO − VNO (6.24)
VCN = VCO − VNO (6.25)

Adding Eqs. (6.23)–(6.25), one obtains
VAN + VBN + VCN = VAO + VBO + VCO − 3VNO (6.26)
But VAN + VBN + VCN = 0 at every instant. Hence from Eq. (6.26),
V AO + VBO + VCO
VNO = (6.27)
3
Knowing the phase to center-point voltages, VAO, VBO and VCO, the neutral-to-center point voltage VNO
can be obtained from Eq. (6.27). The phase voltages VAN, VBN and VCN can then be estimated using
Eqs. (6.23)–(6.25). The above analysis is valid even for the configuration of Figure 6.25. Here the center
point O is assumed as an imaginary or virtual point with a potential of half the input DC voltage.
Six-Step Inverter
The six-step inverter is a three-phase inverter and as the name suggests, there are six steps in the phase-voltage
waveforms per cycle. The phase-voltage waveform for phase AN is shown in Figure 6.26. The drive signals for
the various switches of the three-phase bridge inverter topology are also depicted in Figure 6.26.
Based on the switching sequence for each arm of the inverter, the equivalent circuits for the load at
various switching instants are depicted in Figure 6.27. The entire cycle is divided into six operative modes.
Each mode is operative for 60o of the whole cycle. During mode 1, Q1, Q4 and Q5 are ON and the other
three switches are OFF. The equivalent circuit for the load phases is as shown in Figure 6.27 for mode 1. The
equivalent circuits given in Figure 6.27 at various intervals of times indicate the manner in which the phase-
voltage waveform changes at various instants of time. From Figure 6.27, it is seen that in mode 1, both
phases A and C are connected to VCC and the phase B is connected to ground point gnd. Therefore, two-third
of VCC drops across phase B and one-third of VCC drops across phases A and C with respect to the neutral
Chapter 06.indd 314 3/12/2009 6:27:46 PM

Q1 ON Q2 ON
Q4 ON Q3 ON Q4ON
Q5 ON Q6 ON Q5 ON
2VCC/3
VAN VCC/3
Mode 1 Mode 2 Mode 3 Mode 4 Mode 5 Mode 6
Figure 6.26 Switching sequence for the 3f inverter switches to generate the six-step
phase waveform.
point N as phases A and C are in parallel. By similar argument, the voltage values for all the other modes can
be obtained from Figure 6.27.
It can be observed that each arm is a two-level arm that produces only square waveforms. However, the
switching sequence for the switches in each arm is such that the square waveforms of the arms are displaced
with respect to each other by 120o. Thus it can be argued that if each arm generates a square wave and is dis-
placed with the waveform of the other two arms by 120 o as depicted in Figure 6.27, then a six-step waveform is
obtained across the phases of the three-phase load.
Current-Controlled Inverter
Another type of three-phase inverter in common use is the current-controlled pulse-width modulated
(PWM) inverter that consists of a PWM voltage source inverter fitted with current-regulating loops to pro-
vide a controlled current output (pulse-width modulation is discussed in greater detail in the next section).
It is generally used in AC motor applications. Normally, a sinusoidal reference current waveform, iref , is gene-
rated and fed to a comparator, together with the actual measured current of the motor. The simple approach
Chapter 06.indd 315 3/12/2009 6:27:46 PM

VCC VCC VCC
A C A A B
N N N
gnd B gnd B C gnd C
Mode 1 Mode 2 Mode 3
VCC VCC VCC
B B C C
N N N
gnd A C gnd A gnd A B
Mode 4 Mode 5 Mode 6
Figure 6.27 Equivalent circuits for the various modes of the six-step inverter.
uses the comparator error to switch the devices in the inverter, so as to limit the instantaneous current error.
Figure 6.28 shows the control for one inverter leg. If the motor phase current is more positive than the reference
current value, the upper device is turned OFF and the lower device is turned ON, causing the motor current
to decrease and vice-versa. The comparator is a hysteresis comparator that determines the permitted devia-
tion of the actual phase current from the reference value before an inverter switching is initiated. Thus, the
actual current tracks the reference current without significant amplitude error or phase delay. In three-phase
systems, there is usually an independent current controller for each inverter phase.
Chapter 06.indd 316 3/12/2009 6:27:47 PM

Drive
iref
+ Hysteresis
− comparator
Drive
Figure 6.28 A current control scheme showing one leg of a three-phase inverter.
Figure 6.29 illustrates the type of output current waveform obtained with the simple hysteresis or ON/
OFF current controller. A small dead band gives a near sinusoidal motor current with a small current ripple,
but results in a high switching frequency in the inverter. One should note that the switching frequency is
not constant for a given dead band but is modulated by the variations in motor inductance and back emf.
When the back emf of the motor is low, the switching frequency may rise excessively. In a three-phase
system without a neutral connection, the instantaneous current error can reach double the hysteresis band.
In addition, the variable switching frequency produces objectionable acoustic noise and despite its simplicity
this ON/OFF or “bang–bang” technique is seldom used in practice. However, this principle is used in certain
Dead band
iref
Figure 6.29 Typical phase-current waveshape of the current hysteresis controller.
Chapter 06.indd 317 3/12/2009 6:27:47 PM

AC motor control applications called direct torque control. Alternately the vector control technique is also
used for current control applications. This is discussed in Chapter 10.
A fixed switching frequency is preferred because the acoustic noise is less and the inverter switching
losses are more predictable. Figure 6.30 shows the current control technique in which the current error is
passed through a sample-and-hold (S/H) circuit and compared with a fixed frequency triangular carrier
wave. The current error is essentially the reference or the modulating signal. The resulting PWM signal,
whose duty cycle is such that the current error tends towards zero, controls the inverter switching.
6.6 Pulse-Width Modulation
I n the previous sections, the various topologies of the inverters starting from self-driven inverters to three-
phase inverters were discussed. Some of the common topologies like the push–pull, half-bridge and full-
bridge inverter structures were also discussed. Obtaining simple inverter output waveshapes like the quasi-square
wave in the case of single-phase inverters and the six-step waveshape in the case of the three-phase inverters
were discussed. However, for more complex output waveshapes, the inverter switches are controlled by means
of PWM drive waveforms. The nature and the quality of the output waveshape is controlled by the switching
sequence that is applied to the inverter switches. The main objective of the pulse-width modulation is to control
the fundamental component of the PWM waveform with respect to the applied DC voltage. This control
on the fundamental component is expressed in terms of a measure called the modulation control index m.
Higher the value of “m” lower will be the relative harmonic content. As m decreases, there will be an increase
in the relative harmonic content. The modulation control index is defined as
V1
m= (6.28a)
Vdc
where V1 is the peak value of the fundamental component of the PWM waveform; Vdc the value of the
applied input DC to the inverter.
For the harmonic content, a measure called the harmonic index is defined for each of the harmonic
component relative to the fundamental. The harmonic index hn for the nth harmonic is defined as
Vn
hn = (6.28b)
V1
Drive
iref Hyteresis
+ Control S/H +
− − comparator
Drive
Carrier
Figure 6.30 Carrier-based current control.
Chapter 06.indd 318 3/12/2009 6:27:47 PM

where Vn is the peak value of the nth harmonic component of the PWM waveform; V1 the peak value of the
fundamental of the PWM waveform.
For a practical system, the voltage generated by the inverter should fulfill the following two constraints:
1. The negative part of the waveform should be a mirror reflection of the positive part of the waveform.
This means that the generated waveform should be symmetrical about the time axis. This will ensure
that the even harmonics are absent in the resulting PWM waveform.
2. In the case of three-phase systems, the waveforms generated should be symmetrical. This means that the
waveform of each phase should have the same amplitude as the other phases but should be displaced by
120o with respect to the other two phases.
The PWM can be categorized into the following two general classes, namely,
1. PWM without harmonic elimination;
2. PWM with harmonic elimination.
In the former class, the PWM is performed without any formal effort to cancel specific harmonics whereas
in the latter class, the PWM is performed such that specific harmonics can be eliminated. This section dis-
cusses the PWM without harmonic elimination and the following sections will discuss on the issue of PWM
with specific harmonic elimination.
The PWM without harmonic elimination can be further classified into the following types:
1. quasi-square modulation;
2. end pulse modulation;
3. center pulse modulation;
4. sinusoidal pulse-width modulation;
5. PWM by phase modulation;
6. space-vector modulation.
The Fourier series representation of the waveforms will be used to analyze and discuss the various
modulation methods. The Fourier series representation of any general waveform is given by
a0 ∞ ∞
f (ω t ) = + ∑ an cos(nω t ) + ∑ bn sin(nω t ) (6.29)
2 n =1 n =1
where
2π
1
a0 =
2π ∫ f (ω t )dω t
0
2π
2
an =
2π ∫ f (ω t )cos(nω t )dω t
0
2π
2
bn =
2π ∫ f (ω t )sin(nω t )dω t
0
Here a0 gives the measure of the average value of the waveform. In the discussions to follow, as the wave-
forms that are considered are AC waveforms with zero average, a0 value is considered to be zero. The
waveforms that are considered in the sections to follow have quarter wave symmetry. This means that
Chapter 06.indd 319 3/12/2009 6:27:47 PM

the waveform is symmetrical about a vertical line at q = p/2. In such cases, an or bn can be evaluated con-
sidering only quarter of the periodic waveform. Thus for waveforms with quarter wave symmetry,
π /2
2
an =
π /2 ∫ f (ω t )cos(nω t )dω t
0
π /2
2
bn =
π /2 ∫ f (ω t )sin(nω t )dω t
0
The waveforms that are discussed in the sections to follow have the negative parts that are a mirror reflection
of the positive parts. Therefore the even harmonics are absent. This implies that the harmonic number n will
take only odd values like 1, 3, 5, 7, 9, etc.
The PWM patterns that will be discussed in the following sub-sections and the next section are generally
obtained by programming the timers in a digital processor. Either a microcontroller or a digital signal proces-
sor is generally used for the purpose. However, the PWM sequence can also be obtained by comparison of a
sawtooth or a triangular wave carrier with an appropriate analog signal to generate the various PWM patterns.
The PWM signal is given as input to the drive circuitry of the power switches of the inverter. A block
schematic of a typical PWM inverter is shown in Figure 6.31. In Figure 6.31, the output waveshape that is
pulse-width modulated is filtered to obtain a sinusoidal output voltage. This voltage is fed back for control
purposes. It is rectified and filtered to obtain the average value of the output waveform. This DC voltage is
compared with a desired reference voltage. The error voltage is passed through a proportional–integral controller.
Bridge
AC rectifier Vo
Inverter
1-phase and filter
230 V rms
Vref
+ Controller PWM generator
−
Reference wave
generation
Rectifier
and filter
Figure 6.31 Typical block schematic of a PWM inverter.
Chapter 06.indd 320 3/12/2009 6:27:48 PM

Drive
Dead time
PWM signal
logic
Drive
Figure 6.32 PWM signal driving one arm of an inverter.
The controller output is used for controlling the amplitude of a reference wave. This controlled sine wave
reference is compared with a carrier frequency waveform to generate the PWM waveform that is used for
switching the inverter switches.
Figure 6.32 illustrates the schematic for driving one arm of the inverter bridge. The PWM signal that is
generated by a specific algorithm is passed through a dead time logic. This logic ensures that at every voltage
transition of the PWM pattern, sufficient dead time is provided for the ON-state switch to turn OFF and the
OFF-state switch to subsequently turn ON, such that the positive rail of the inverter bridge is not short-circuited
to the negative rail. The signal is then used to drive the inverter switches of a specific arm. For single-phase
inverters, there are two bridge arms wherein one bridge arm gets a PWM signal that is generated by a reference
modulating wave that is phase shifted by 180o with respect to the reference modulating wave for the other arm.
In the case of three-phase inverters, there are three bridge arms, one each for phases A, B and C. Each
arm is given a PWM signal wherein the PWM signal is generated by a reference modulating wave that is
phase shifted 120o with respect to the reference modulating waves of the other two arms.
Quasi-Square Modulation
Quasi-square modulation is the same as the duty cycle control as discussed in the section “Quasi-Square
Wave Inverter”. The quasi-square waveform is again depicted here in Figure 6.33 wherein Vdc is the applied
DC voltage to the inverter and d is the angular pulse width for the quasi-square wave. The pulse width d is
varied or modulated to obtain the variation in modulation index and the harmonic content.
The Fourier series representation of the quasi-square waveforms is given by Eq. (6.31). Referring to
Figure 6.33, it can be observed that the waveform has odd symmetry about the vertical axis at 0. The wave-
form further has quarter wave symmetry about q = p/2; therefore
θ π /2
2 2
π / 2 ∫0 ∫ Vdc sin(nω t )dω t
bn = 0 sin(nω t )dω t +
π /2 θ
Chapter 06.indd 321 3/12/2009 6:27:48 PM

f
Vdc d
0 p
p wt
q 2
Figure 6.33 Quasi-square wave.
4Vdc π /2
bn = [ − cos(nω t ) θ ]
nπ
4Vdc
bn =[cos nθ − cos(nπ / 2)]
nπ
Now cos(nπ / 2) = 0 for n = 1, 3, 5, 7, …. . The waveform does not contain even harmonics as the positive
and negative parts are mirror images. Thus one obtains
4Vdc
bn = cos nθ for n = 1, 3, 5, … (6.30)
nπ
∞
4Vdc
f quasi =
nπ
∑ [ cos(nθ )sin(nω t )] for n = 1, 3, 5, … (6.31a)
n =1
The angle,
π δ
− θ= (6.31b)
2 2
It can be observed from Eq. (6.31a) that for q = 0, fquasi becomes a square waveform fsquare that is given as
∞
4Vdc
f square =
nπ
∑ [ sin(nω t )] for n = 1, 3, 5, …
n =1
The peak value of the fundamental component of fquasi is given as
4Vdc 4Vdc ⎛ π δ ⎞ 4V ⎛δ ⎞
V1 = cos(θ ) = cos ⎜ − ⎟ = dc sin ⎜ ⎟ (6.32)
π π ⎝2 2⎠ π ⎝2⎠
From Eq. (6.32), the modulation control index m is obtained as
V1 4 ⎛ δ ⎞
m= = sin ⎜ ⎟ (6.33)
Vdc π ⎝ 2 ⎠
Chapter 06.indd 322 3/12/2009 6:27:48 PM

The harmonic indices are given as

Vn cos(nθ )
hn = = for n = 3, 5, 7, … (6.34)
V1 n sin(δ / 2)
The modulation control index and the harmonic indices are dependent on the angular pulse width d and
therefore q. The plot of the modulation control index and the harmonic indices are shown in Figure 6.34.
The third harmonic and the harmonics that are multiples of three are not plotted as they will be eliminated
in a three-phase system. It can be observed that at low modulation control index (m) the relative harmonic
content is large. For values of m less than 0.7, the relative harmonic content starts becoming significantly
high. As d tends towards p, the modulation control index tends towards that of a square wave.
For three-phase circuits, the phase-to-neutral voltages are quasi-square waveforms. Each phase waveform
is displaced by 120° with respect to the other two phase waveforms. The line-to-line voltage waveform is the
difference between two phase-voltage waveforms and is as depicted in Figure 6.35. Figure 6.35 shows
the A phase and B phase voltage. The line-to-line voltage between A and B phases is also shown. The
line-to-line voltage waveforms contain the same harmonics as the phase-voltage waveforms; however,
the third harmonic and the harmonics that are multiples of three are eliminated. Further, the line-to-line
voltage waveforms are 3 times that of the phase-to-neutral waveforms. Thus
∞
4 3Vdc
f quasi- 3φ =
nπ
∑ cos(nθ )sin(nω t ) for n = 1, 5, 7, 11, …
n =1
1.4
1.2
m m=
1.0
h5 =
h7 =
0.8 h11 =
h7
0.6
0.4 h11
h5
0.2
0.0
0 20 40 60 80 100 120 140 160 180
Pulse width d in degree
Figure 6.34 Modulation control index and harmonic indices for quasi-square modulation.
Chapter 06.indd 323 3/12/2009 6:27:49 PM

VAN
wt
π 2π
2 3
VBN
wt
VAB
wt
Figure 6.35 Phase- and line-voltage waveforms for quasi-square modulation.
End Pulse Modulation

End pulse modulation is another pattern of modulation wherein each half-cycle starts and ends with an
interval where the voltage is of opposite sign. The end pulse waveshape is shown in Figure 6.36. The end
pulse width q is varied or modulated to obtain the variation in modulation control index and the relative
harmonic content.
The Fourier series representation of the end pulse waveform is given by Eq. (6.36). Referring to
Figure 6.36, it can be observed that the waveform has odd symmetry about the vertical axis at 0. The waveform
further has quarter wave symmetry about q = p/2; therefore only bn coefficients exist. Thus,
f
Vdc
End pulse
0
π p wt
2
End pulse
q q
Figure 6.36 End pulse waveform.
Chapter 06.indd 324 3/12/2009 6:27:49 PM

θ π /2
2 2
bn = ∫
π /2 0
( −Vdc )sin(nω t )dω t +
π /2 ∫ Vdc sin(nω t )dω t
θ
−4Vdc ⎡ θ 4V π /2
bn = − cos(nω t ) ⎤ + dc ⎡− cos(nω t ) ⎤
nπ ⎣ 0 ⎦ nπ ⎣ θ ⎦
4Vdc
bn = ( 2 cos nθ − 1) for n = 1, 3, 5, 7, … (6.35)
nπ
∞
4Vdc
f endpulse =
nπ
∑ (2 cos nθ − 1)sin(nω t ) for n = 1, 3, 5, … (6.36)
n =1
It can be observed from Eq. (6.36) that for q = 0, fendpulse becomes a square waveform fsquare that is given as
∞
4Vdc
f square =
nπ
∑ sin(nω t ) for n = 1, 3, 5, …
n =1
The peak value of the fundamental component of fendpulse is given as

4Vdc
( 2 cos θ − 1)
V1 = (6.37)
π
From Eq. (6.37), the modulation control index (m) is obtained as
V1 4
m= = ( 2 cos θ − 1) (6.38)
Vdc π
Vn 2 cos nθ − 1
hn = = (6.39)
V1 n(2 cos θ − 1)
where n = 3, 5, 7, …. The plot of the modulation control index and the harmonic indices are shown in
Figure 6.37. The relative harmonics that are not multiples of three are plotted. It can be observed that as
the modulation control index (m) decreases, the relative harmonic content increases. For values of m less
than 1.1, the relative harmonic content starts becoming significantly higher. As q tends towards p/6 the
modulation control index drops to zero as can be seen from Eq. (6.38). It can be observed by comparing
Figures 6.34 and 6.37 that the range of the modulation control index is larger for the quasi-square modu-
lation for a given relative harmonic content.
For three-phase circuits, the phase-to-neutral voltages are the end pulse waveforms as depicted in
Figure 6.38. Each phase waveform is displaced by 120° with respect to the other two-phase waveforms.
The line-to-line voltage waveform is the difference between two-phase voltage waveforms and is as
depicted in Figure 6.38. Figure 6.38 shows the A phase and B phase voltage. The line-to-line voltage
between A and B phases is also shown. The line-to-line voltage waveforms contain the same harmonics as
the phase-voltage waveforms but the third harmonic and the harmonics that are multiples of three are
eliminated. The Fourier series expansion for the line-to-line voltage waveforms is given as
∞
4 3Vdc
f endpulse- 3φ =
nπ
∑ (2 cos nθ − 1)sin(nω t ) for n = 1, 5, 7, 11, …
n =1
Chapter 06.indd 325 3/12/2009 6:27:49 PM

1.4
1.2
1.0 m
m=
h5 = h11
0.8 h7 = h5
h11 =
0.6
h7
0.4
0.2
0.0
0 10 20 30 40 50 60
End pulse width q in degree
Figure 6.37 Modulation control index and harmonic indices for end pulse modulation.
VAN V
0
p 2p wt
2 3
VBN
0
V wt
2V
VAB
0
wt
Figure 6.38 Phase- and line-voltage waveforms for end pulse modulation.
Chapter 06.indd 326 3/12/2009 6:27:49 PM

Center Pulse Modulation

Center pulse modulation is another variant of modulating the square wave wherein each half-cycle consists
of an interval in the center of the half-cycle where the voltage is of opposite sign. The center pulse waveshape
is shown is Figure 6.39. The center pulse width d is varied or modulated to obtain the variation in modula-
tion control index and the relative harmonic content.
The Fourier series representation of the center pulse waveform is given by Eq. (6.41). Referring to
Figure 6.39, it can be observed that the waveform has odd symmetry about the vertical axis at 0. Therefore the
Fourier coefficients an = 0. The waveform further has quarter wave symmetry about q = p/2, therefore
θ π /2
2 2
π / 2 ∫0 ∫ (−Vdc )sin(nω t )dω t
bn = Vdc sin(nω t )dω t +
π /2 θ
−4Vdc ⎡ θ 4V π /2
bn = − cos(nω t ) ⎤ + dc ⎡− cos(nω t ) ⎤
nπ ⎣ 0 ⎦ nπ ⎣ θ ⎦
4Vdc
bn = (1 − 2 cos nθ ) for n = 1, 3, 5, 7, … (6.40)
nπ
∞
4Vdc
f centerpulse =
nπ
∑ (1 − 2 cos nθ )sin(nω t ) for n = 1, 3, 5, … (6.41a)
n =1
δ = π − 2θ (6.41b)
It can be observed from Eq. (6.41a) that for q = p/2, d = 0 and fcenterpulse becomes a square waveform fsquare
that is given as
∞
4Vdc
f square =
nπ
∑ sin(nω t ) for n = 1, 3, 5, …
n =1
The peak value of the fundamental component of fcenterpulse is given as

4Vdc 4Vdc ⎡ ⎛ π δ ⎞⎤ 4Vdc ⎛ δ⎞
V1 = (1 − 2 cos θ ) = ⎢1 − 2 cos ⎜ − ⎟⎥ = ⎜ 1 − 2 sin ⎟ (6.42)
π π ⎣ ⎝ 2 2 ⎠⎦ π ⎝ 2⎠
f
Center pulse
Vdc
d
0 wt
q p p
2
Center pulse
Figure 6.39 Center pulse waveform.
Chapter 06.indd 327 3/12/2009 6:27:50 PM

From Eq. (6.42), the modulation control index m is obtained as

V1 4 ⎛ δ⎞
m= = ⎜ 1 − 2 sin ⎟ (6.43)
Vdc π ⎝ 2⎠
Vn 1 − 2 cos nθ
hn = = (6.44)
V1 n[1 − 2 sin(δ / 2)]
where n = 3, 5, 7, …. The plot of the modulation control index and the harmonic indices are shown in
Figure 6.40. It can be observed that as the modulation control index m decreases, the relative harmonic con-
tent increases. For values of m less than 1.1, the relative harmonic content starts becoming significantly
higher. The plot shows that there is a near linear relation between the modulation control index m and the
pulse width d. This implies that the control voltage and the fundamental of the output voltage are also
linear.
The end pulse and center pulse modulation can have multiple pulses to cancel some specific harmonics.
This would mean that there can be more than one end/center pulses. The harmonic elimination method
uses multiple pulses to eliminate certain harmonics. This is discussed in the next section while discussing the
pulse-width modulation with harmonic elimination.
1.4
m=
h5 =
1.2 h7 =
h11 =
1.0 h7
m
0.8
h5
0.6
h11
0.4
0.2
0.0
0 10 20 30 40 50 60
Center pulse width d in degree
Figure 6.40 Modulation control index and harmonic indices for center pulse modulation.
Chapter 06.indd 328 3/12/2009 6:27:50 PM

VAN
0
π/2 2π/3 wt
VBN
0
wt
VAB
0
wt
Figure 6.41 Phase- and line-voltage waveforms for center pulse modulation.
For three-phase circuits, the phase-to-neutral voltages are the end pulse waveforms as depicted in
Figure 6.41. Each phase waveform is displaced by 120o with respect to the other two-phase waveforms.
The line-to-line voltage waveform is the difference between two-phase voltage waveforms and is as
depicted in Figure 6.41. Figure 6.41 shows the phase A and phase B voltages. The line-to-line voltage
between A and B phases is also shown. The line-to-line voltage waveforms contain the same harmonics as
the phase-voltage waveforms but the third harmonic and the harmonics that are multiples of three are
eliminated. The line-to-line voltage is expressed as
4 3Vdc ∞
f centerpulse- 3φ =
nπ
∑ (1 − 2 cos nθ )sin(nωt ) for n = 1, 5, 7, 11, …
n =1
Sinusoidal Pulse-Width Modulation

The sinusoidal output waveshape is one of the most common requirements for the loads. This is usually
accomplished by means of driven inverter topologies. To obtain a sinusoidal output waveshape, the square,
quasi, end or center PWM waveform that is generated from the inverter is passed through a filter such that
the harmonics are attenuated and only the fundamental is allowed to pass through to the load. In the case of
the square, quasi-square, end pulse or center pulse waveforms, the significant harmonic after the fundamen-
tal is the third harmonic in the case of single-phase loads and the fifth harmonic in the case of three-phase
loads as discussed in the previous sub-sections. This implies that the filter should have a sharp cut-off profile
to significantly attenuate the amplitudes of the third and higher harmonics.
In the sinusoidal pulse-width modulation technique, the pulse width of a high-frequency switching
signal, called the carrier, is varied in accordance to the amplitude of a low-frequency signal, called the
modulating signal, as shown in Figure 6.42.
Chapter 06.indd 329 3/12/2009 6:27:50 PM

Modulating signal (fundamental)
High-frequency carrier
Figure 6.42 Unipolar pulse-width modulation: Sinusoidal fundamental modulating the

high-frequency carrier.
It is seen from the frequency spectrum that the major harmonic after the modulating signal frequency is
at the carrier frequency. If the modulating signal is a sinusoid of 50 Hz, then one gets a sinusoidal PWM
waveform with a fundamental at 50 Hz. If the carrier frequency is 20 kHz, then the major harmonic after
50 Hz will be at 20 kHz. Therefore, to filter out the carrier from the PWM, the filter requirements are less
stringent compared to filtering out the harmonics in a square or a quasi-square waveform. As the filter is
supposed to filter out only the high-frequency carrier, the filter size will be small.
The sinusoidal PWM waveform as depicted in Figures 6.42 and 6.43 has quarter wave symmetry about
the p/2 vertical and odd symmetry about the 0 vertical. Therefore, the Fourier coefficient an = 0 and bn
contains only odd harmonics.
fc
fm
1
0
p /2 wt
−1
V+
0
q1 q2 q3
wt
V−
Figure 6.43 A portion of the bipolar sinusoidal pulse-width modulated waveform.
Chapter 06.indd 330 3/12/2009 6:27:51 PM

The angular pulse width is proportional to the amplitude of the modulating signal value at the specific
q values. The sinusoidal modulating signal is compared with a high-frequency triangle or sawtooth wave-
form. The points of intersection between the sinusoidal modulating signal and the triangle carrier are the
angles at which the voltage transitions take place. Every voltage transition is called a “notch”. The corres-
ponding angles at which these transitions occur are called the “notch angles”. The voltage transits between
two levels from V+ to V– and vice-versa. V+ is usually the input DC voltage level Vdc and V– is usually –Vdc
or 0. At every notch, if the voltages transit between positive and negative voltage, then such a PWM is called
bi-polar PWM as depicted in Figure 6.43. At every notch if the voltages transit between zero and either positive
or negative voltage, then such a PWM is called a unipolar PWM as illustrated in Figure 6.42.
The carrier frequency or the frequency of the switching pulses is
1
ωc = 2π f c ; Tc =
fc
and the modulating frequency or the frequency of the fundamental sinusoid is
1
ωm = 2π f m ; Tm =
fm
The ratio of the carrier frequency to the modulating frequency is called the frequency ratio r that is given as
ωc f
r= = c (6.45)
ωm f m
Referring to Figure 6.43, the Fourier representation of the sinusoidal PWM for “p” notches in a quarter
cycle is given as
∞
f sinusoidal = ∑ bn sin(nω t ) (6.46)
n =1
where
θ θ π /2
2 1 2 2 2
bn = ∫
π /2 0
V + sin( nω t )dωt + ∫
π /2θ
V− sin(nω t )dωt + ... +
π /2 ∫ V+ sin(nω t )dω t
1 θp
p p
4V+ 4V−
bn =
nπ
∑ cos nθi − cos nθi +1 + nπ
∑ cos nθj − cos nθj +1 (6.47)
i =0 j =1
for i = even (0, 2, 4, 6, 8, …); j = odd (1, 3, 5, 7, …); n = odd (1, 3, 5, 7, …). Referring to Figure 6.43, the
notch angles are determined by two waveforms: (a) the triangular carrier and (b) the modulating signal. The
points of intersection of these two waveforms give the notch angles. Therefore, based on the modulation
signal, the amplitude of the modulating signal at the notch angle is
f = M sin θi (6.48)
where M is the modulation index which is defined as the ratio of the peak amplitude of the fundamental
signal to the peak amplitude of the carrier signal. Based on the carrier signal and referring to Figure 6.43, the
amplitude of the carrier signal at the notch angle is
⎛ 2r ⎞
f = ( −1)i ⎜ 2i − θi ⎟ (6.49)
⎝ π ⎠
Chapter 06.indd 331 3/12/2009 6:27:51 PM

1.2
r=7
M=1
0.8
0.6
0.4
0.2
−0.2
0 1 3 5 7 9 11 13
Harmonic number n
Figure 6.44 Harmonic line spectrum for bipolar PWM.
where 2r/p is the slope of the triangular carrier. At the intersection points, Eqs. (6.48) and (6.49) will be
equal. These two equations should be used to compute qi iteratively for the purpose of determining the
notch angles for analysis. The notch angles thus determined can be substituted in Eq. (6.47) to obtain the
various harmonic components. The harmonic line spectrum are plotted for a frequency ratio r = 7 and modu-
lation index M = 1. Figure 6.44 shows the harmonic line spectrum for the bipolar PWM and Figure 6.45
shows the harmonic line spectrum for the unipolar PWM. It can be observed that the unipolar PWM gives
a better modulation control index resulting in higher fundamental amplitude. The carrier amplitude is con-
siderably reduced. Figure 6.46 shows the modulation control index m and the harmonic indices for the 5th,
7th and 11th harmonics for a bipolar PWM. Likewise, Figure 6.47 shows the modulation control index and
the harmonic indices for the unipolar PWM. It can be observed that the bipolar PWM gives a better har-
monic reduction than the unipolar PWM.
General Notes on Sinusoidal PWM

1. It should be observed that for low values of r, the relative harmonic content is significantly high. Care
should be taken not to add even harmonics due to unsymmetry in the positive and negative half of the
PWM pattern. The value of frequency ratio r should be an integer. This means that the carrier and
the modulating waveforms are synchronous thereby avoiding discontinuities and harmonics due to dis-
continuities. The value of r should be odd. This will ensure that the number of pulses for the positive
half and the negative half are same and symmetric thereby avoiding even harmonics.
Chapter 06.indd 332 3/12/2009 6:27:51 PM

1.2
r=7
0.8 M=1
0.6
0.4
0.2
−0.2
0 1 3 5 7 9 11 13
Harmonic number n
Figure 6.45 Harmonic spectrum for unipolar PWM.
2.5
2 M=1
1.5
m
1
0.5
h 5, h 7, h 11
0
5 10 15 20 25 30 35 40 45 50 55
Frequency ratio, r = fc/fm
Figure 6.46 Modulation control index and harmonic indices for bipolar PWM.
Chapter 06.indd 333 3/12/2009 6:27:52 PM

1.8
1.6
M=1
1.4
1.8
m
0.8
0.6 h5
h7 h11
0.4
0.2
0
5 10 15 20 25 30 35 40 45 50 55
Frequency ratio, r = fc/fm
Figure 6.47 Modulation control index and harmonic indices for unipolar PWM.
2. To ensure that the generated PWM is symmetric and does not contain even harmonics in single-phase
inverters, two 180o out-of-phase sine waves are used to modulate the same triangular carrier. The pulse
sequence generated by one sine wave is used to switch one arm of the full-bridge inverter and the pulse
sequence generated by the out-of-phase sine wave is used to switch the other arm of the full-bridge. This
will completely cancel out the even harmonics even if the frequency ratio is not odd.
3. The carrier can also be a sawtooth waveform. Either a positive slope or a negative slope sawtooth
waveform can be used for modulating the sine wave. However, the notch angles must be evaluated
accordingly for the sawtooth waveform to perform analysis.
4. For frequency ratios greater than r = 20, the relative harmonic content of the bipolar PWM reduces
significantly and the integer constraint on r can be relaxed.
5. Harmonic content of the harmonic number n = r is quite significant. This can be filtered out by a low-pass
filter as indicated in Figure 6.48. This harmonic can be significantly reduced by using unipolar PWM.
6. In the case of three-phase systems, the PWM pattern for each arm of the inverter is displaced by 120o
with respect to the other two arms. In the line-to-line voltage, the third harmonic and the harmonics
that are multiples of three will be eliminated. Thus if the frequency ratio is chosen as a multiple of three,
then n = r will be automatically eliminated in the line voltage waveforms for three-phase loads.
7. For modulation index, M > 1, the notches will gradually disappear and the fundamental voltage V1 will
no longer be proportional to M. The operation when M > 1 is called over-modulation.
There are many variations from the implementation point of view. For the bipolar PWM, a triangular carrier
that swings from –1 to +1 is used. However for implementing the unipolar PWM, steering and sequencing of the
Chapter 06.indd 334 3/12/2009 6:27:52 PM

Amplitude
Filter characteristics
fm requirement
fc − 2fm
fc − fm
fc
fc + fm
fc + 2fm
Frequency
Figure 6.48 Frequency spectrum of pulse-width modulated waveform.
inverter switching is needed. For a full-bridge inverter as depicted in Figure 6.49, the gate drive signal
strategy for generating a sinusoidal PWM waveform is shown in Figure 6.50. The high-frequency PWM
for every alternate half-cycle of the fundamental is given to Q1 and Q3 of the full-bridge inverter. Alternate
half-cycle pulses at fundamental frequency (10 ms pulse for a 50 Hz sine wave) are applied to Q2 and Q4.
The voltage across AB is a PWM voltage as shown in the waveforms of Figure 6.50. Here one should note
that the flux f in the transformer is given using
dφ
V AB = N (6.50)
dt
and therefore
1
N ∫ AB
φ= V dt (6.51)
As VAB is a PWM switching waveform, it is evident from Eq. (6.51), that the flux will be the integral of the
PWM waveform. The integral of the PWM waveform will have a frequency of the modulating signal and
hence the flux in the transformer core will also have the modulating signal frequency, that is, 50 Hz for a 50
Hz sine wave. Therefore, the transformer has to be designed for the modulating frequency. As a conse-
quence, the size of the transformer will become large.
In order to reduce the size of the transformer, configurations will have to be used such that the flux in
the transformer core will have the frequency of the carrier and not that of the modulating signal. One
of the ways to achieve high-frequency isolation is to shift the control of the output fundamental amplitude
to the control of the inverter DC-link voltage. The inverter is operated at a fixed modulation index to convert
the DC-link voltage to sinusoidal PWM output. The amplitude of the output is controlled by varying
Chapter 06.indd 335 3/12/2009 6:27:53 PM

Vin
Q1 Q3
A B
Q2 Q4
Figure 6.49 Full-bridge inverter with output filter.
Q2
drive
Q4
drive
Q1
drive
Q3
drive
VAB
Figure 6.50 Unipolar sinusoidal pulse-width modulated waveform generation in a full-bridge inverter.
Chapter 06.indd 336 3/12/2009 6:27:53 PM

DC–DC
converter
AC with high- Inverter Vo
1 phase frequency
230 V rms isolation
PWM generator
Vref with fixed
+ Controller
− modulation
index
Rectifier
and filter
Figure 6.51 PWM inverter schematic with high-frequency transformer for isolation.
the DC-link voltage value. The isolation transformer is within the DC–DC converter wherein the trans-
former can be designed for high carrier frequencies thereby reducing the size of the core. This scheme is
shown in the block diagram of Figure 6.51. But one should note that here two power stages are involved
(DC–DC converter and inverter) and therefore the efficiency is lower compared to the previous scheme
wherein only one power stage, that is, the inverter only is used.
PWM through Phase Modulation

This is another scheme wherein the isolation transformer is a high-frequency transformer. This scheme
(shown in Figure 6.52) has the following important features:
1. The transformer always sees high-frequency square wave of 50% duty cycle at the carrier frequency
across its primary and hence the transformer is compact in size.
2. The filter circuit sees a PWM waveform with a carrier frequency that is twice the switching frequency.
Hence the filter requirements become less stringent.
S1 S3 A
− +
Vo
Vin
S2 S4
Figure 6.52 Phase modulation inverter topology.
Chapter 06.indd 337 3/12/2009 6:27:53 PM

Principle of Operation
Let S1, S2, S3 and S4 be semiconductor switches. Switches S3 and S4 are, in particular, bi-directional
switches. Switches S1 and S2 are switched complimentary to each other. Switches S3 and S4 are switched
complimentary to each other. However, there exists a phase shift q between the primary-side switching and
the secondary-side switching of the transformer as shown in Figure 6.53.
From the waveforms shown in Figure 6.53, one can observe the following:
1. The voltage at A (VA) has a switching waveshape that is twice the switching frequency, hence the output
filter (L–C) design is at twice the switching frequency.
2. If the phase shift between the primary and the secondary patterns is modulated in a sinusoidal manner,
then VA is a sinusoidal PWM waveform.
3. The transformer always sees 50% duty cycle at switching frequency. Therefore, the transformer can be
designed for the switching frequency, which will result in a smaller sized transformer.
4. One should note that S3 and S4 should be bi-directional switches, that is, they should allow current to
flow in either direction. A bi-directional switch can be built as shown in Figure 6.54.
A bi-directional switch can be constructed using a BJT or a MOSFET or an IGBT as depicted in
Figure 6.54. A current flowing from A to B follows the path A–D1–Q–D4–B and a current flowing from B
to A follows the path B–D3–Q–D2–A. Note that in both the current directions, the current through the
transistor switch Q is always in the same direction. Though the push–pull configuration is useful to demon-
strate the principle of phase modulation, half-bridge or full-bridge configurations are usually preferred on
S1
drive
signal
S2
drive
signal
S3 q
drive
signal
S4
drive
signal
VA
Figure 6.53 Waveforms for the phase modulation inverter topology.
Chapter 06.indd 338 3/12/2009 6:27:53 PM

D1 D3
A Q B
D2 D4
Figure 6.54 Bi-direction switch using a BJT.
the primary side. On the secondary side, instead of a center-tapped winding, a full-bridge configuration
can be used because the voltage rating for the secondary switches in the full-bridge configuration will be
half that for the center-tapped winding configuration, given the same output voltage requirements.
Space-Vector Modulation
Space-vector modulation is another PWM generation principle. This technique is employed primarily for
three-phase inverter systems. However this technique can be adopted for two-phase system also and extended
to the poly-phase systems. Some of the features of this method that have made it a very popular three-phase
PWM generation method are:
1. It is inherently suitable for digital implementation.
2. It provides a higher DC-link voltage utilization compared to the sinusoidal PWM.
3. It provides lower relative harmonic content compared to the sinusoidal PWM.
4. The switching frequency of the inverter switches is half the carrier frequency. Therefore, the switching
losses are lesser as compared to a sinusoidal PWM with the same carrier frequency.
Primary Vectors
The three-phase system is described by a co-planar vector space with three axes. The three-phase inverters
have three bridge arms. Each arm represents an axis in the co-planar vector space. Therefore, there are three
axes, one for each arm of the bridge inverter that is uniformly distributed in the co-planar vector space at an
angular distance of 120o with respect to each other. Any vector in the co-planar vector space can be repre-
sented by a triple (a1, b1, c1). The co-planar vector space can be described by three primary vectors (0, 0, 1),
(0, 1, 0) and (1, 0, 0). The primary vectors of the co-planar vector space of the three-phase system are
depicted in Figure 6.55. Using the a-axis as the reference axis, the primary vectors in the polar form are rep-
resented as
(0, 0,1) ⇒ 1 ⋅ e j 0 = 1
(0,1, 0) ⇒ 1 ⋅ e j ( 2π /3) = e j ( 2π /3)
(1, 0, 0) ⇒ 1 ⋅ e j ( 4π /3) = e j ( 4π /3)
Chapter 06.indd 339 3/12/2009 6:27:54 PM

b-
ax
(0, 1, 0)
is
120°
240°
(0, 0, 1) a -axis
(1, 0, 0)
is
ax
c-
Figure 6.55 Primary vectors defining the co-planar vector space.
Any vector in the co-planar vector space is called a space vector and is represented in terms of the primary
vectors as
→
V = k1 ⋅ 1 + k2 ⋅ e j ( 2π /3) + k3 ⋅ e j ( 4π /3) (6.52)
where k1, k2 and k3 are scalar quantities along a, b and c axes, respectively.
Space Vectors
Figure 6.56 shows the three primary states of the inverter bridge arms that are connected to a three-phase
star load. Each arm is a two-level switch arm. When only one of the bridge arms is connected to the positive
rail, then such a state is a primary state. The state (0, 0, 1) occurs when the A phase switch is connected to
Vdc and the B and C phase switches are connected to the ground. Likewise, the state (0, 1, 0) occurs when
only the B phase switch is connected to the positive rail and the state (1, 0, 0) occurs when only the C phase
switch is connected to Vdc. The three primary states of the inverter are represented as vectors in the co-planar
vector space as
→
Va = Vdc ⋅ 1 + 0 ⋅ e j ( 2π /3) + 0 ⋅ e j ( 4π /3) = Vdc (6.53)
Similarly
→
Vb = Vdce j ( 2π /3) (6.54)
→
Vc = Vdc e j ( 4π /3) (6.55)
Each of the inverter bridge arms can assume two levels, either Vdc or 0. This implies that there are 23 states
(i.e., 8 states) in all. The eight different states of the inverter including the primary states are depicted in
Figure 6.57. The last two states, namely, state (0, 0, 0) and state (1, 1, 1) result in zero vectors in the co-planar
Chapter 06.indd 340 3/12/2009 6:27:54 PM

C B A C B A C B A
Vdc Vdc Vdc
N N N
State: 0 0 1 State: 0 1 0 State: 1 0 0
Figure 6.56 Primary states of a two-level three-arm bridge inverter.
vector space. This means that when all the switches are connected to Vdc or when all the switches are con-
nected the 0, the resulting space vector has amplitude of zero.
Figure 6.58 shows the space vectors that are obtained for the eight different states of the inverter bridge.
There are two zero space vectors and they are represented as
V00 = 0 ⋅ 1 + 0 ⋅ e j ( 2π /3) + 0 ⋅ e j ( 4π /3) = O ⇒ (0, 0, 0)
V01 = 1 ⋅ 1 + 1 ⋅ e j ( 2π /3) + 1 ⋅ e j ( 4π /3) = Ο ⇒ (1,1,1)
V1 = Vdc ⇒ (0, 0,1) (the primary vector along a-axis)
V3 = Vdc ⋅ e j ( 2π /3) ⇒ (0,1, 0) (the primary vector along b-axis)
V5 = Vdc ⋅ e j ( 4π /3) ⇒ (1, 0, 0) (the primary vector along c-axis)

The other three vectors can be constructed from the three primary vectors. They are
V2 = Vdc ⋅ 1 + Vdc ⋅Vdc ⋅ e j ( 2π /3) ⇒ (0,1,1)
V4 = Vdc ⋅ e j ( 2π /3) + Vdc ⋅Vdc ⋅ e j ( 4π /3) ⇒ (1,1, 0)
V6 = Vdc ⋅ 1 + Vdc ⋅ e j ( 4π /3) ⇒ (1, 0,1)

The six space vectors and the two zero space vectors form a hexagonal pattern as shown in Figure 6.58. The
hexagon is divided into six sectors of 60o each. This division is made to simplify the analysis. Analysis made
for any one sector can be applied to the other sectors with appropriate angular bias.
Space-Vector Modulation Principle

Consider a space vector of amplitude kVdc that is aligned along the a-axis as shown in Figure 6.59. Let this
constant amplitude space vector rotate with an angular speed of w starting from the a-axis. If an observer is
stationed on the a-axis and observes the projection of the vector tip onto an axis orthogonal to the a-axis and
Chapter 06.indd 341 3/12/2009 6:27:54 PM

C B A C B A C B A
Vdc Vdc Vdc
N N N
C B A C B A C B A
Vdc Vdc Vdc
N N N
C B A C B A
Vdc Vdc
N N
State: 0 0 0 State: 1 1 1
Figure 6.57 States of a two-level three-arm bridge inverter.
plots these projection as time evolves, a sinusoidal waveform with respect to time (VAN) is obtained as shown
and is given as
V AN = kVdc sin ω t (6.56)
Similarly, if an observer is stationed on the b-axis and observes the projection of the vector tip onto an axis
orthogonal to the b-axis and plots these projects as time evolves, a sinusoidal waveform with respect to time
Chapter 06.indd 342 3/12/2009 6:27:55 PM

b-
ax
Vdc(0,1,0) = V3
is
Vdc(0,1,1) = V2
II
III I
Vdc(1,1,0) = V4 V01(1,1,1) V00(0,0,0)
a-axis
Vdc(0,0,1) = V1
IV VI
V
Vdc(1,0,1) = V6
is Vdc(1,0,0) = V5
ax
c-
Figure 6.58 Space vectors in the co-planar vector space.
(VBN) is obtained. However, in this case the b-axis is positioned at an angular distance of 120o from the
a-axis in the co-planar vector space. If the vector completes a circle in a period T, then it would take a time
of T/3 to reach the b-axis. Thus the VBN waveform will lag the waveform VAN with respect to time or equiva-
lently the delay in angle is
T 2π T 2π
ω = ⋅ = radian
3 T 3 3
Thus, VBN is given as
⎛ 2π ⎞
VBN = kVdc sin ⎜ ωt − ⎟ (6.57)
⎝ 3 ⎠
wt
0°
12
b-
ax
is
N
VB
kVdc VAN
wt a-axis
0° K<1
24
VCN
is
wt
ax
c-
Figure 6.59 Amplitude–time projections on the three axes for a constant amplitude rotating vector in
the vector space.
Chapter 06.indd 343 3/12/2009 6:27:55 PM

Similarly, the time evolution of the projection of the space-vector tip onto an axis orthogonal to the
c-axis is given as
⎛ 4π ⎞
VCN = kVdc sin ⎜ ω t − ⎟ (6.58)
⎝ 3 ⎠
Thus one can conclude that if a constant amplitude space vector rotates at a constant speed, then the resulting
waveform with respect to time on the three axes are sinusoidal and displaced in time with respect to each other by
120°. Conversely, one can also conclude that if three-phase sinusoidal waveforms of frequency w and displaced
with respect to each other by 120° in time are applied at the three axes, then the resulting space vector is of con-
stant amplitude and rotates at a constant speed of w. This is the basic principle that is used in space-vector
modulation to obtain sinusoidal PWM.
The two-level three-phase inverter can produce only eight space vectors including the two zero vectors as
depicted in Figure 6.58. Using combinations of these eight space vectors, any given vector within the hexagon
can be obtained by time averaging in a carrier period. This means that in a carrier period Tc, the inverter spends
T1 duration of time in state (0, 0, 1) and T2 duration of time in state (0, 1, 1) and T0 duration of time in the zero
vector state (0, 0, 0) or (1, 1, 1). Time averaging the space vectors for a carrier period of Tc gives an equivalent
averaged vector kVdc as shown in Figure 6.60. This equivalent averaged vector is made to rotate at a speed wm.
Then the time evolution of the waveforms on the three axes is sinusoidal as described by Eqs. (6.56)–(6.58).
b-
ja-axis
ax
is
(T2)
Vdc(0,1,1)
kV dc
(Tc)
0 θ (T1)
(T0) Vdc(0,0,1) a-axis

is
ax
c-
Figure 6.60 Arbitrary space-vector realization based on time-averaged state vectors.
Consider the sector I of the hexagon as shown in Figure 6.60. If one has to realize a time-averaged space
vector of amplitude kVdc at an angle q from the a-axis, then the inverter must spend time in the bounding
states of the sector and the zero space-vector state. Let the inverter spend T1 duration of time in state (0, 0, 1),
T2 duration of time in state (0, 1, 1) and T0 duration of time in the zero space-vector state. Then equiva-
lently the inverter is said to output a time-averaged space vector of amplitude kVdc at an angle q from the
a-axis for the duration Tc, the carrier period. To derive the time durations, the time-averaged space vector
can be projected onto the a-axis and an axis orthogonal to the a-axis called the ja-axis as shown in Figure 6.60.
Referring to Figure 6.60, projecting onto the a-axis, one obtains
Chapter 06.indd 344 3/12/2009 6:27:55 PM

⎛π ⎞
kVdc (cos θ )Tc = VdcT1 + Vdc cos ⎜ ⎟ T2 (6.59)
⎝3⎠
Projecting onto the ja-axis, one obtains
⎛π ⎞
kVdc (sin θ )Tc = Vdc sin ⎜ ⎟ T2 (6.60)
⎝3⎠
From Eqs. (6.59) and (6.60), the two time durations T1 and T2 can be estimated. Thus, from Eq. (6.60),
2
T2 = kTc (sin θ ) (6.61)
3
where Tc is the carrier period and 0 ≤ k ≤ 1. From Eqs. (6.58) and (6.60), one obtains
⎡ 1 ⎤
T1 = kTc ⎢cos θ − sin θ ⎥
⎣ 3 ⎦
2 ⎡ 3 1 ⎤
T1 = kTc ⎢ cos θ − sin θ ⎥
3 ⎢⎣ 2 2 ⎥⎦
2 ⎛π ⎞
T1 = kTc sin ⎜ − θ ⎟ (6.62)
3 ⎝3 ⎠
The carrier period Tc is given as
Tc = T1 + T2 + T0
Therefore, the time spent by the inverter in the zero vector state is
T0 = Tc − T1 − T2 (6.63)
For sector I, the reference state is (0, 0, 1). For sector-II, state (0, 1, 1) becomes the reference state and the
Eqs. (6.61)–(6.63) are used to estimate the T1, T2 and T0 times. Likewise for every sector, Eqs. (6.61)–(6.63) are
used to estimate the T1, T2 and T0 times. However, the bounding states are different and therefore, the system
state will accordingly change depending on the sector of interest. If a space vector has an angle a measured from
the a-axis, then angle q for the calculation of the state times depends on the sector number and is given as
π
θ = α − ( s − 1) ⋅
3
where
⎛ α ⎞
s = truncate ⎜ ⎟ +1
⎝π /3⎠
is the current sector within which the space vector exists. The angle q will change at the rate of wm, the modu-
lating sinusoid frequency. Equations (6.61)–(6.63) are used to estimate the times T1, T2 and T0 for every
carrier period interval, which in turn will determine the inverter switch states during the carrier period.
The value of k which scales the primary space-vector amplitude can vary from 0 to 1. However, if the
space vector has to maintain a circular trajectory, then the amplitude of the space vector is limited
Chapter 06.indd 345 3/12/2009 6:27:56 PM

b -a
xi s
Vdc cos30° = 3/2 Vdc
60°
30°
kVdc a-axis
k= 3/2
xis
c -a
Figure 6.61 Maximum circular space-vector trajectory.
to ( 3 / 2)Vdc such that the hexagon sides are tangents to the circular trajectory as shown in Figure 6.61.
If the value of k is increased beyond 3 / 2 then the trajectory inscribed by the space vector will not be cir-
cular but clamped near the centers of the hexagon sides. Therefore, for k > 3 / 2 , additional harmonics will
be introduced and the inverter is said to be in over-modulation. As the value of k is increased to 1, the trajectory
gradually changes over to the hexagonal trajectory as shown in Figure 6.62. In the limit, the inverter can be in
each of the six states for an angular period of 60o. This will result in the classical six-step waveform as discussed
in the previous section.
Sequencing
Figure 6.63 shows the timing diagram wherein during the T1 period the PWM signal given to the three
phases is such that the inverter is in state (0, 0, 1). During T2 the inverter signals are (0, 1, 1) as indicated in
the figure. During the zero vector period, either (0, 0, 0) or (1, 1, 1) can be used. However, the best har-
monic and switching loss performance is obtained if T0 period is split into two and both zero vectors are
b -a
xis
k=1
a-axis
k= 3/2 = 0.866
= 3/2 < k < 1

xis
c -a
Figure 6.62 Maximum circular space-vector trajectory for over-modulation.
Chapter 06.indd 346 3/12/2009 6:27:56 PM

T1 T2 T0 T1 T2
A
1 1 1
0 1 0
B
0 0 0
C
(1,1,1) OR (0,0,0)
Figure 6.63 Choice of zero vectors.
used in a carrier period. Figure 6.64 shows two carrier period cycles wherein the T0 period is split into two
T0/2 periods that are appended to either side of T1 and T2 as shown. The zero vector periods are chosen such
that during any period only one inverter arm will switch. This leads to the A, B and C arm switching patterns
as indicated in Figure 6.64. Observe that due to the choice of the zero vectors, as indicated the switching
period of the inverter, Ts is twice that of the carrier period. Therefore, the inverter arms switch at half the
carrier period, thereby reducing the switching losses for a given carrier.
T0 /2 T1 T2 T0 /2 T0 /2 T2 T1 T0 /2
0 1 1 1 1 1 1 0
A
0 0 1 1 1 1 0 0
B
0 0 0 1 1 0 0 0
C
Tc Tc
Ts
Figure 6.64 Timing sequence.
Chapter 06.indd 347 3/12/2009 6:27:56 PM

The zero vectors are appropriately chosen for every sector of the hexagon such that the switching
frequency is half of the carrier frequency. The switching sequence for two carrier periods when the time-
averaged space vector is in various sectors are
Sector-I
(0, 0, 0) (0, 0,1) (0,1,1) (1,1,1) (1,1,1) (0,1,1) (0, 0,1) (0, 0, 0)
T0 T0 T0 T0
T1 T2 T2 T1
2 2 2 2
Sector-II
(1,1,1) (0,1,1) (0,1, 0) (0, 0, 0) (0, 0, 0) (0,1, 0) (0,1,1) (1,1,1)
T0 T0 T0 T0
T1 T2 T2 T1
2 2 2 2
Sector-III
(0, 0, 0) (0,1, 0) (1,1, 0) (1,1,1) (1,1,1) (1,1, 0) (0,1, 0) (0, 0, 0)
T0 T0 T0 T0
T1 T2 T2 T1
2 2 2 2
Sector-IV
(1,1,1) (1,1, 0) (1, 0, 0) (0, 0, 0) (0, 0, 0) (1, 0, 0) (1,1, 0) (1,1,1)
T0 T0 T0 T0
T1 T2 T2 T1
2 2 2 2
Sector-V
(0, 0, 0) (1, 0, 0) (1, 0,1) (1,1,1) (1,1,1) (1, 0,1) (1, 0, 0) (0, 0, 0)
T0 T0 T0 T0
T1 T2 T2 T1
2 2 2 2
Sector-VI
(1,1,1) (1, 0,1) (0, 0,1) (0, 0, 0) (0, 0, 0) (0, 0,1) (1, 0,1) (1,1,1)
T0 T0 T0 T0
T1 T2 T2 T1
2 2 2 2
It should be noted that the modulating signal for the space-vector modulation is a composite waveform
consisting of the sinusoidal fundamental and a third harmonic component that is introduced due to
splitting T0 into two T0/2 durations placed at either ends of T1 and T2 durations. The third harmonic
component tends to flatten the top of the modulating waveform, thereby increasing the modulation control
index. The equivalent modulating signal, compared with a triangular carrier and results in the space vector
modulated output, is shown in Figure 6.65. The equivalent modulating signal consists of the dominant
fundamental component fb and a third harmonic component fa. The equivalent modulating signal is the
algebraic sum of the functions fa and fb.
As can be observed from Figure 6.65, the modulating signal is the summation of the sinusoidal funda-
mental and a third harmonic triangular component. The third harmonic triangular component flattens the
top of the sinusoidal component. At a modulation index M = 1, the fundamental amplitude is 15% more as
Chapter 06.indd 348 3/12/2009 6:27:57 PM

Increase in fundament
1 Amplitude w.r.t sinusoidal PWM
f m = fa + f b
fb
0 wt
fa
−1
Figure 6.65 Equivalent modulating signal of space-vector PWM.
compared to that of the sinusoidal PWM. In sinusoidal PWM, the control index m = 1 compared to 4/p for
the unmodulated waveform. This is about 78.5% of the unmodulated waveform. In the case of the space-
vector PWM, the fundamental amplitude is 93% of the unmodulated waveform.
Multi-Level Inverters
The space-vector modulation concept can be extended to multi-level inverters too. Figure 6.66 shows the
space vectors for the four-level three-arm bridge converter.
As discussed for the two-level inverter, the co-planar vector space can be described by three primary vec-
tors (0, 0, 1), (0, 1, 0) and (1, 0, 0). Any vector in the co-planar vector space (i.e., space vector) is repre-
sented in terms of the primary vectors. Each of the axes consists of four levels: 0, k1Vdc, k2Vdc and Vdc. Thus
there are four primary space vectors for each of the three axes. These primary vectors can be vectorially
added to obtain the other states of the multi-level inverter. For sector-I, the state vectors along the a-axis and
the negative c-axis can be combined to form three more space-vector states as shown. Thus, each sector gets
divided into nine smaller sub-sectors.
b -axis Vdc (0,1,0)

Vdc (0,1,1)
k2 Vdc (0,0,1)
Vdc (0,0,1)
Vdc (1,1,0)
a-axis
0
k1 Vdc (0,0,1)
Vdc (1,0,0) Vdc (1,0,1)

c -axis
Figure 6.66 State vectors in the co-planar vector space for four-level three-arm bridge.
Chapter 06.indd 349 3/12/2009 6:27:57 PM

As there are more states available, there is more flexibility in obtaining the equivalent time-averaged
rotating space vector. Different choice of states and sequencing can lead to different optimizations. As there
are more states available for the inverter, the carrier frequency to the fundamental modulating frequency
ratio (r) can be significantly reduced.
6.7 Pulse–Width Modulation with Harmonic Elimination
T here are many applications wherein harmonics are detrimental to the functioning of the system. In
applications involving transformers and AC motors, the harmonics will lead to excess core losses and
consequently heat up the core material which leads to deterioration of the magnetic core. Further, in some
applications some harmonics lead to undesirable effects. In the case of single-phase AC motors like fans, the
harmonics can lead to torque pulsations and audible noise generation. Therefore to reduce the adverse
effects sometimes few specific undesirable harmonics like third, fifth, seventh, etc. need to be eliminated.
Consider the quasi-square waveform as shown in Figure 6.67. The fundamental component is given by
setting n = 1. This is given as
4V
f fundamental = cos θ sin ω t (6.64)
π
The control is performed by duty cycle control as this approach gives a greater control over the harmonic
elimination rather than the DC-link control. Using this approach, from Eq. (6.64), it can be observed that
there is only one variable that can be controlled and that is q. With this, one can either
control the amplitude of the fundamental only
OR
control the amplitude of one particular harmonic
OR
eliminate one particular harmonic
wt
Figure 6.67 Quasi-square waveform and its fundamental.
Chapter 06.indd 350 3/12/2009 6:27:58 PM

For example, if one desires to eliminate the third harmonic, then one has to set
4V 1
⋅ ⋅ cos(3θ )sin(3ω t ) = 0
π 3
which implies that 3q = 90o or q = 30o. Therefore, if one sets q = 30o for the quasi-square waveform, then
the third harmonic gets eliminated. Thus, one can generalize that by setting q = 90o/n, the nth harmonic can
be eliminated. If one also needs to control the amplitude of the fundamental in addition to eliminating one
harmonic, then two control variables are needed, one to eliminate the harmonic and one to set the funda-
mental to the desired level.
Harmonic Elimination with Multiple Notches

With a quasi-square waveform, there is only one control variable and hence either the fundamental can be
controlled or a harmonic can be eliminated. However, if one needs to control many parameters of the
output waveform then a more complex waveform structure is needed. Consider the waveform shown in
Figure 6.68 that contains two control variables, q1 and q2. Note that the waveform has quarter-wave sym-
metry. This implies that the Fourier series would then contain only “sine” terms. q1 and q2 are called the
“NOTCH” angles. At every notch, the voltage transits from either positive to negative or negative to posi-
tive resulting in voltage reversal. If the “notches” q1 and q2 are specified, then the whole waveshape becomes
fully determined.
For the waveform of Figure 6.68 with two notches q1 and q2, the average value in a period is zero and
therefore
a0 = 0
The waveform of Figure 6.68 has quarter wave symmetry and therefore,
an = 0
0 q1 q2 p/2 wt
Figure 6.68 Two-notch pulse-width modulated waveform.
Chapter 06.indd 351 3/12/2009 6:27:58 PM

Further as the waveform of Figure 6.68 is symmetrical about the x-axis, it contains only odd harmonics.
Thus for a quarter wave symmetrical waveform containing only odd harmonics,
π /2
2
bn =
π /2 ∫ f (ωt )sin(nωt )dωt
0
4V
bn = [1 + 2(− cos nθ1 + cos nθ2 )] for n = 1, 3, 5, … (6.65)
nπ
For example, if one desires to set the fundamental amplitude to “x” and also to eliminate the third harmonic
for the two-notch PWM waveform, then it is required that
4V
b1 = x = [1 + 2(− cos θ1 + cos θ2 )]
π
4V
b3 = 0 = [1 + 2( − cos 3θ1 + cos 3θ2 )]
3π
4V
f1 = [1 + 2(− cos θ1 + cos θ2 )] − x = 0 (6.66)
π
4V
f2 = [1 + 2(− cos 3θ1 + cos 3θ2 )] = 0 (6.67)
3π
Solving Eqs. (6.66) and (6.67) for the two unknowns q1 and q2, one obtains q1 and q2 such that the two-
notch waveform has a fundamental amplitude of “x” and the third harmonic is absent. This principle can be
extended to waveshapes having more than two notches. Let there be “p” notches per quarter cycle, then,
“p” harmonics can be eliminated
OR
“( p − 1)” harmonics can be eliminated with a specified value for the fundamental amplitude
For “p” notches per quarter cycle with quarter wave symmetry,
4V ⎡ p
⎤
bn =
nπ
⎢1 + 2 ∑ (−1)k cos nθk ⎥ (6.68)
⎣ k =1 ⎦
where qk is the kth notch angle. From Eq. (6.68), one obtains “p” equations which will allow q1, q2, …, qp
to be estimated for the specified constraints on the selected harmonics.
Determination of Notch Angles

This section summarizes the harmonic elimination process by providing a step-by-step approach to deter-
mine the notch angles for specific harmonic constraints.
Step 1: Decide on the number of notches required per quarter cycle. This can be specified based on the
number of harmonics one needs to eliminate and whether fundamental amplitude requires to
be set to a desired value or not.
Step 2: Using Eq. (6.68), obtain the p equations in p unknown notch angles q1, q2, …, qp.
Chapter 06.indd 352 3/12/2009 6:27:58 PM

Step 3: The p equations obtained from Step (2) form a set of non-linear equations. The solution to the
set of p equations is based on the extension of the Newton–Cotes algorithm to non-linear
equations. The solution is an iterative algorithm that is given as
⎡ θ1 ⎤ ⎡ θ1 ⎤ ⎡ f 1(θ1 ,...,θ p ) ⎤
⎢ . ⎥ ⎢ . ⎥ ⎢ . ⎥
⎢ ⎥ = ⎢ ⎥ − [ J ]−1 ⎢ ⎥ (6.69)
⎢ . ⎥ ⎢ . ⎥ ⎢ . ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢⎣θ p ⎥⎦k +1 ⎢⎣θ p ⎥⎦k ⎢⎣ f p (θ1 ,...,θ p )⎥⎦k
where J is the Jacobian matrix for the solution of non-linear equations and is given by
⎡ ∂f 1 ∂f 1
(
⎢ ∂θ θ1 ,...,θ p ) . .
∂θ p
( )
⎤
θ1 ,...,θ p ⎥
⎢ 1 ⎥
⎢ . . ⎥
J =⎢ ⎥ (6.70)
⎢ . . ⎥
⎢ ∂f ∂f p ⎥
(
⎢ p θ1 ,...,θ p ) . . ( )
θ1 ,...,θ p ⎥
⎣⎢ ∂θ1 ∂θ p ⎥⎦
k
Step 4: Start the iterative algorithm by making an initial guess for the p unknown notch angles, that is
⎡ θ1 ⎤ ⎡ a1 ⎤
⎢ . ⎥ ⎢. ⎥
⎢ ⎥ =⎢ ⎥
⎢ . ⎥ ⎢. ⎥
⎢ ⎥ ⎢ ⎥
⎢⎣θ p ⎥⎦k = 0 ⎢⎣ a p ⎥⎦
where a1 to ap are the initial guess values for q1 to qp, respectively.

Step 5: Iteratively evaluate
⎡ θ1 ⎤
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎢⎣θ p ⎥⎦k +1
till the function matrix

⎡ f 1(θ1 ,...,θ p ) ⎤ ⎡ ε1 ⎤
⎢ . ⎥ ⎢ . ⎥
⎢ ⎥ <⎢ ⎥
⎢ . ⎥ ⎢ . ⎥
⎢ ⎥ ⎢ ⎥
⎢⎣ f p (θ1 ,...,θ p )⎥⎦k ⎢⎣ε p ⎦⎥
where ei are very small values that are used to break the iteration loop once the error in the
function matrix value at instant k are less than the eI values.
Chapter 06.indd 353 3/12/2009 6:27:58 PM

EXAMPLE 6.1 Let it be required that the fifth and the seventh harmonics be eliminated from a
PWM waveform.
Step 1: It is required that two harmonics are to be eliminated and there is no
constraint on the amplitude of the fundamental. This implies that two not-
ches per quarter cycle are sufficient. Hence, p = 2.
Step 2: Using Eq. (6.68), one obtains
f1 = cos(5q1) − cos(5q2) − 0.5 = 0
f2 = cos(7q1) − cos(7q2) − 0.5 = 0
Step 3: The Jacobian J is obtained as
⎡ ∂f 1 ∂f 1 ⎤
⎢ ⎥
∂θ ∂θ2 ⎥ ⎡ −5 sin(5θ1 ) 5 sin(5θ2 ) ⎤
J =⎢ 1 =⎢ ⎥
⎢ ∂f ∂f 2 ⎥ ⎣−7 sin(7θ1 ) 7 sin(7θ2 )⎦
⎢ 2 ⎥
⎢⎣ ∂θ1 ∂θ2 ⎥⎦
Formulate the iterative equation to solve for q1 and q2. It is given as
⎡θ1 ⎤ ⎡θ1 ⎤ −1 ⎡ f 1 ⎤
⎢θ ⎥ = ⎢θ ⎥ − [ J ] ⎢ f ⎥ (6.71)
⎣ 2 ⎦k +1 ⎣ 2 ⎦k ⎣ 2 ⎦k
Step 4: To start the iterative solution as given by Eq. (6.71), initial values of q1 and
q2 are required. Therefore, make an initial guess for q1 and q2 at k = 0.
⎡f ⎤
Evaluate the Jacobian matrix J and ⎢ 1 ⎥ with values of q1 and q2 at k = 0.
⎣ f2 ⎦
Using Eq. (6.71), evaluate
⎡θ1 ⎤
⎢θ ⎥
⎣ 2 ⎦k =1
Step 5: Iteratively evaluate
⎡θ1 ⎤
⎢θ ⎥
⎣ 2 ⎦k +1
till each element of
⎡ f1 ⎤
⎢f ⎥
⎣ 2 ⎦k
is less than a predetermined value, say less than 0.0001. Then the corre-
sponding values of q1 and q2 at iteration level k + 1 will be the required
notch angles q1 and q2.
Implementation: One of the simple ways to implement this harmonic elimination technique is to store the
values of [q1, …, qp] in a read only memory (ROM). The notch angles can be loaded into a compare register
sequentially and cyclically to generate the “p” notches per quarter wave waveshape. Due to the computation
involved in calculating the Jacobian and solving the non-linear equation, the harmonic elimination method
is normally used with pre-computed values for specific harmonics and fundamental frequency and stored in
Chapter 06.indd 354 3/12/2009 6:27:58 PM

the ROM. However today with high compute power and low-cost digital signal processors available, it is
possible to dynamically estimate the notch angles for various harmonics and specific fundamental frequency
on-the-fly. Further, for the three-phase loads the computational overhead is much reduced as the third har-
monic and the harmonics that are multiples of three are anyway eliminated. This means that the dominant
harmonics that need to be eliminated are the 5th, 7th, 11th and 13th harmonics. This implies that only four
notches are needed and a 4 × 4 Jacobian matrix inverse needs to be evaluated.
Staircase PWM
Another harmonic elimination method is the staircase PWM technique. This method combines the princi-
ples of the sinusoidal PWM and the harmonic elimination with notch method to give a simple PWM tech-
nique with higher modulation control index than the bipolar PWM. The carrier is a triangular waveform
that transits between –1 and +1 like in the case of the sinusoidal PWM carrier. The modulation signal is
instead a staircase waveform. The levels of the staircase waveform are decided based on the harmonics to be
eliminated in a method similar to the notch-based harmonic elimination method. Figure 6.69 depicts the
triangular carrier and a two-level staircase waveform superimposed on the carrier wherein the modulation
index M is less than 1. The angles corresponding to the points of intersection are the notch angles at which
the voltage transition takes place.
The carrier frequency or the frequency of the switching pulses is
1
ωc = 2π f c ; Tc =
fc
and the modulating frequency or the frequency of the fundamental is
1
ωm = 2π f m ; Tm =
fm
As discussed for the sinusoidal PWM, the ratio of the carrier frequency to the modulating frequency is called
the frequency ratio r that is given as
ωc f
r= = c (6.72)
ωm f m
M<1
1 m0
m2
m1 p/2
0
wt
−1 m1
m2 p/r
V+
d1 d2
0
q1 q2 q3 q4 wt
V−
Figure 6.69 Staircase pulse-width modulation.
Chapter 06.indd 355 3/12/2009 6:27:59 PM

Referring to Figure 6.69, the Fourier representation of the staircase PWM for “p” notches in a quarter cycle
is given as
∞
f staircase = ∑ bn sin(nωt ) (6.73)
n =1
where
θ θ π /2
2 1 2 2 2 p +1
bn = ∫
π /2 0
( −V )sin(nω t )dω t +
π / 2 θ∫
V sin(nω t )dω t + +
π /2 ∫ (−1) V sin(nω t )dω t
1 θp
4V p
4(−V ) p
bn =
nπ
∑ cos nθi − cos nθi+1 + ∑ cos nθj − cos nθj+1
nπ j=1
(6.74)
i =0
for i = even (0, 2, 4, 6, 8, …); j = odd (1, 3, 5, 7, …); n = odd ( 1, 3, 5, 7, …).

4V
bn = [cos n0 − 2 cos nθ1 + 2 cos nθ2 − 2 cos nθ3 + + ( −1) p 2 cos nθp ]
nπ
4V
bn = [1 − 2 cos nθ1 + 2 cos nθ2 − 2 cos nθ3 + + ( −1) p 2 cos nθp ]
nπ
−8V
bn = [cos nθ1 − cos nθ2 + cos nθ3 − cos nθ4 + + ( −1) p +1 cos nθp − 0.5]
nπ
−8V ⎡ ⎛ θ1 + θ2 ⎞ ⎛ θ2 − θ1 ⎞ ⎛ θ3 + θ 4 ⎞ ⎛ θ 4 − θ3 ⎞ ⎤
bn = ⎢2 sin ⎜ n 2 ⎟ sin ⎜ n ⎟ + 2 sin ⎜ n ⎟ sin ⎜ n ⎟ + + (− )0.5⎥
nπ ⎣ ⎝ ⎠ ⎝ 2 ⎠ ⎝ 2 ⎠ ⎝ 2 ⎠ ⎦
−16V ⎡ ⎛ θ1 + θ2 ⎞ ⎛ θ2 − θ1 ⎞ ⎛ θ3 + θ 4 ⎞ ⎛ θ 4 − θ3 ⎞ ⎤
bn = ⎢sin ⎜ n 2 ⎟ sin ⎜ n ⎟ + siin ⎜ n ⎟ sin ⎜ n ⎟ + + (−)0.25⎥ (6.75)
nπ ⎣ ⎝ ⎠ ⎝ 2 ⎠ ⎝ 2 ⎠ ⎝ 2 ⎠ ⎦
For a specific level of the staircase waveform, the notch width will be the same at any comparison point with
the triangular carrier. Thus for a given level, the notch width can be defined as follows:
1. Level-1: Notch angle width is δ1 = θ2 − θ1 .
2. Level-2: Notch angle width is δ2 = θ4 − θ3 and so on.
The mid-points of the pulse widths are given as
θ1 + θ2 ; θ +θ
μ1 = μ2 = 2 3 ; …
2 2
The Fourier series coefficient bn for the staircase waveform as given in Eq. (6.75) can be re-written as
−16V ⎡ k1 ⎛ δ ⎞ 1 2
k +k
⎛ δ ⎞ ⎤
bn = ⋅ ⎢∑ sin nμi sin ⎜ n 1 ⎟ + ∑ sin nμi sin ⎜ n 2 ⎟ + + (−)0.25⎥ (6.76)
nπ ⎢⎣ i =1 ⎝ 2 ⎠ i =k1 +1 ⎝ 2 ⎠ ⎥⎦
where k1 is the number of pulses resulting from level-1 staircase intersection with the carrier; k2 the
number of pulses resulting from level-2 staircase intersection with the carrier and so on. For k1 = k2 = 1,
the Eq. (6.76) reduces to
Chapter 06.indd 356 3/12/2009 6:27:59 PM

−16V ⎡ ⎛ δ ⎞ ⎛ δ ⎞ ⎤
bn = ⎢sin nμ1 sin ⎜ n 1 ⎟ + sin nμ2 sin ⎜ n 2 ⎟ + + ( −)0.25⎥
nπ ⎢⎣ ⎝ 2⎠ ⎝ 2 ⎠ ⎥⎦
For an m-level staircase waveform, at M = 1, the mth level will be at par with the peak amplitude of the car-
rier and therefore, there will be no notches corresponding to the mth level. In a 3-level staircase waveform,
there will be notches corresponding to only two levels for M = 1. To eliminate the fifth and the seventh har-
monics, from Eq. (6.76), one has to set
k +k
⎛ k1 ⎞ δ ⎛ 1 2 ⎞ δ
f 1 = ⎜ ∑ sin 5μi ⎟ sin 5 1 + ⎜ ∑ sin 5μi ⎟ sin 5 2 − 0.25 = 0 (6.77)
⎝ i =1 ⎠ ⎜
2 ⎝ i =k1 +1 ⎟ 2
⎠
k +k
⎛ k1 ⎞ δ ⎛ 1 2 ⎞ δ
f 2 = ⎜ ∑ sin 7μi ⎟ sin 7 1 + ⎜ ∑ sin 7μi ⎟ sin 7 2 − 0.25 = 0 (6.78)
⎝ i =1 ⎠ ⎜
2 ⎝ i =k1 +1 ⎟ 2
⎠
For k1 = k2 = 1, Eqs. (6.77) and (6.78) reduce to
⎛ δ ⎞ ⎛ δ ⎞
f 1 = sin(5μ1 )sin ⎜ 5 1 ⎟ + sin(5μ2 )sin ⎜ 5 2 ⎟ − 0.25 = 0 (6.79a)
⎝ 2⎠ ⎝ 2 ⎠
⎛ δ ⎞ ⎛ δ ⎞
f 2 = sin( 7 μ1 )sin ⎜ 7 1 ⎟ + sin( 7 μ 2 )sin ⎜ 7 2 ⎟ − 0.25 = 0 (6.79a)
⎝ 2⎠ ⎝ 2 ⎠
The midpoints of the pulse widths m1 and m2 are the center points of the triangular carrier and therefore they
are known quantities:
π /2 π π
μi = + i = ( 2i + 1) for i = 0, 1, 2, 3, … (6.80)
r r 2r
Here d1 and d2 are the unknown quantities that are estimated by solving Eqs. (6.77) and (6.79) using the
method outlined in section on “Determination of Notch Angles” for the harmonic elimination with multi-
ple notches.
The levels of the staircase waveforms are proportional to the pulse widths di. If the level is at the zero
axis, then the intersection points with the triangular carrier are on the zero axis and the pulse width is
equal to p/r. If the level is at +1, then the intersection with the carrier is at a point on the peak of the
triangle and the pulse width is zero. As the carrier is a linearly rising waveform, the relationship between
the pulse width and staircase level for the modulation index is illustrated in Figure 6.70. The height of the
ith level is
⎛ δ ⎞
mi = M ⎜⎜ 1 − i ⎟⎟ (6.81)
⎝ π /r ⎠
where mi is the height of the ith level of the staircase PWM waveform.
Summary of Staircase PWM Design

1. The fundamental component of the staircase PWM waveform has a control index m that is higher than
the sinusoidal PWM for a given modulation index M.
Chapter 06.indd 357 3/12/2009 6:27:59 PM

1
Level Carrier
m1
d1
M1
0 m0 m1 m2 wtm
p/r
−1
Figure 6.70 Triangular carrier along with the pulse width and level relationship.
2. For an n-level staircase PWM, n − 1 harmonics can be eliminated at a modulation index of M = 1. For
modulation indices less than 1, the harmonics that are designed to be eliminated at M = 1 will remain
eliminated or negligible.
3. To design the staircase PWM pattern for eliminating specific harmonics, first choose the number of
levels for the staircase PWM.
4. Then choose the frequency ratio r.
5. Then choose the number of pulses needed for every level, that is, k1, k2, ….
6. Evaluate the pulse-width center points m1, m2, etc. as given by Eq. (6.80).
7. Formulate the n − 1 equations for eliminating the n − 1 harmonics along similar lines like Eqs. (6.78)
and (6.79).
8. Solve the n − 1 non-linear equations as discussed in the previous sub-section for harmonic elimination
with multiple notches and obtain the required pulse widths for the staircase PWM.
9. From the estimated values of the pulse widths, calculate the heights of the levels of the staircase
PWM.
The vast expanse of DC–AC converters or inverters one desires at the output. Therefore, time spent on
is rather difficult to discuss in a single chapter. the study of PWM is considered to be well spent.
However, the essentials have been discussed in this Most of the pulse-width modulators were based
chapter that will give a flavor of the various issues on operational amplifier (op-amp) based circuits.
and the insights gained will help in pursuing However, today most if not all algorithms are
advanced inverter concepts. One of the most signifi- embedded into the digital processors. This chapter
cant issues that needs to be assimilated is the con- does not discuss the algorithmic implementation of
cept of pulse-width modulation (PWM). It is the the PWM, but focuses more on the Fourier analysis
PWM that eventually gives one the waveform that aspects. This will provide a sound base for the design
Chapter 06.indd 358 3/12/2009 6:27:59 PM

of the PWM patterns. The algorithmic implementa- The laboratory exercises suggested in the next
tion aspect is dealt in Chapter 12 after a formal dis- section are mainly for the beginners to get a good
cussion on the discrete computation issues. understanding of the basics of PWM and dead
Harmonic elimination methods that are dis- time delay aspects for single-phase inverters. This is
cussed in this chapter were implemented previously in no way an exhaustive list of experiments. Based
by lookup table methods where the patterns were on the topics covered in the chapter, one may
pre-computed for various fundamental amplitudes extend to three-phase and multi-level inverter
and stored in ROM or EPROM. Depending on the experiments along similar lines as suggested for
control input, the specific pattern for the required single-phase inverter. It should be noted that the
fundamental was read out from the stored memory more popular PWM methods like the space-vector
location. However, today, with even low-priced PWM are not easy to implement with analog
DSPs having good computational speeds, on-the-fly devices like the op-amp. They are best implemented
calculation of the required PWM pattern (notches as embedded algorithms on a digital processor as
or staircase levels) for harmonic elimination is a fea- discussed in Chapter 12.
sible possibility.
1. Consider the block schematic of the pulse- Tasks for study:
width modulator as shown in Figure 6.71. It (a) Based on the block schematic of Figure 6.71,
consists of a comparator wherein the modulat- rig up the pulse-width modulator circuit
ing signal is one input and it is compared with using a Schmitt comparator, one signal
the carrier signal connected to the other input. generator for carrier signal and another
The output of the comparator is a PWM signal. signal generator for the modulating signal.
The PWM signal is passed through a low-pass The low-pass filter can be any standard
filter to obtain the filtered output. op-amp based filter like a Sallen-Key
filter.
Comparator
(b) Set the carrier signal to be a triangular
waveform swinging from –1 V to +1 V
Vm + LPF Vo with a carrier frequency of 20 kHz. Set the
− Vpwm
Modulating low-pass filter cut-off at 2 kHz. Set the mod-
signal ulating signal generator to be a sinusoidal
waveform of frequency 50 Hz with the
amplitude at 0.5 V peak.
Vc (c) Observe the PWM output Vpwm and the
(Carrier) filtered output Vo. Compare Vo and Vm
and observe the differences.
Figure 6.71 Pulse-width modulator.
(d) For hardware experiment, the data may be
Mode of implementation: The above circuit copied to a computer in ASCII format.
can be studied by The data may then be imported to SciLAB
a. Simulation in Spice or MATLAB. Check and compare the line
b. Simulation in SciLAB spectrum of the harmonic content in Vm,
c. Hardware bread-boarding Vpwm and Vo.
Chapter 06.indd 359 3/12/2009 6:28:00 PM

(e) Change the amplitude of the modulating (b) RT and CT are used to set the carrier fre-
signal and repeat Steps (3) and (4). quency. In this case the carrier is a saw-
(f ) Keeping the carrier frequency at 20 kHz tooth waveform swinging from 0 to 3 V.
and without changing the low-pass filter Therefore, the modulating signal that is
cut-off, increase the modulating signal fre- given to the DTC pin should have a bias at
quency gradually and observe the effect on 1.5 V and the sinusoidal waveform with a
Vpwm and Vo. Observe the effect on the peak value less than 1.5 V should be super-
harmonic contents of Vpwm and Vo also. imposed on the bias such that a level-
(g) Using Vpwm as input, rig up a gating or shifted sine wave is obtained.
steering circuit for the four switches of a (c) Observe the PWM output, Vpwm and
full-bridge inverter. Vpwm .
2. Consider the block schematic of the pulse-width 3. The top switch and the bottom switch of an
modulator as shown in Figure 6.72. It consists inverter arm have finite rise and fall times. To
of a TL494 IC with the modulating signal given ensure that the turning-OFF device has sufficient
as the DTC pin of the IC. Vpwm and Vpwm are time to turn OFF, the turn-ON signal for the
the two out-of-phase PWM signals that are other switch of the arm is delayed by a few micro-
required to be given to the inverter switches. seconds. This way a dead time is ensured at every
switch transition that prevents the DC link from
shorting to the ground. The dead time delay is
can be studied by
incorporated in a digital processor by using a low
bit counter. In hardware implementation, one of
the ways to implement dead time delay is as
Tasks for study: shown in Figure 6.73. It consists of Schmitt AND
(a) Rig up the pulse-width modulator circuit gates connected as shown. The RC is connected
as shown in Figure 6.72. to provide the delay for every rising edge.
12 V
Vref
−
Vpwm
+
1 kΩ 12 V
−
TL494
1 kΩ
+ Vpwm
DTC
RT
CT S/P
2.2 kΩ
10 KpF
Figure 6.72 Pulse-width modulator based on TL494 IC.
Chapter 06.indd 360 3/12/2009 6:28:00 PM

To Q1, Q4 drive
2.2 KΩ
Vpwm
Inf
Dead time delay circuit
To Q2, Q3 drive
2.2 KΩ
Vpwm
Inf
Figure 6.73 Dead time delay circuit.
Mode of implementation: The above circuit 4. The schematic for the phase-modulated PWM
can be studied by is shown in Figure 6.74. The TL494 PWM cir-
a. Simulation in Spice cuit is same as shown in Figure 6.73. The
b. Hardware bread-boarding output of the PWM circuit is given to a JK
flip–flop IC that is connected to behave as a
Tasks for study:
toggle flip–flop. The output of the flip–flop is
(a) Rig up the pulse-width modulator circuit
passed through a dead time delay circuit as
followed by the dead time logic as shown
shown in Figure 6.73. Only here there are four
in Figure 6.73.
sets of Schmitt gate segments, one for each
(b) The PWM outputs, Vpwm and Vpwm are
signal line. The outputs from the dead time
given as input to the dead time logic.
delay circuit is used to drive the primary and
(c) Observe Q1 and Q2 drive signals on the
secondary-side switches of the phase-modulated
oscilloscope both during the rising and
PWM inverter.
falling transitions and note the dead time.
(d) Change the RC time constant and observe Mode of implementation: The above circuit
the effect on the dead time. can be studied by
(e) Likewise observe the dead time between a. Simulation in Spice
the Q3 and Q4 drive signals. b. Hardware bread-boarding
J JK-FF
Q1 drive
H K Q1 Primary
Vpwm side
CLK Q2 drive switches
S Q1
TL494 R Delay
ckt. circuit Q3 drive
CLK Q2 Secondary
S side
Vpwm R switches
Q4 drive
J Q2
H K
Figure 6.74 Drive signal generation for phase-modulated PWM.
Chapter 06.indd 361 3/12/2009 6:28:00 PM

Tasks for study: respect to the positive rail and the n-channel
(a) Rig up the pulse-width modulator circuit MOSFETs can be driven with respect to the
followed by the toggle flip-flop and the negative rail. The load is a resistor connected
dead time logic as shown in Figure 6.74. across points A and B as shown.
(b) The PWM outputs, Vpwm and Vpwm , are
given as input to the JK flip–flop con- can be studied by
nected as a toggle flip–flop. a. Simulation in Spice
(c) Observe Q1 and Q2 drive signals on the b. Hardware bread-boarding
oscilloscope both during the rising and
falling transitions and note the dead Tasks for study:
time. (a) Rig up the inverter circuit using p-channel
(d) Likewise observe the dead time between and n-channel MOSFETs as shown in
the Q3 and Q4 drive signals. Figure 6.74.
(e) Plot the waveforms of the TL494 outputs, (b) Using the PWM circuits of the previous
flip-flop outputs and delay circuit outputs. laboratory experiments, modify the drive
(f ) Vary the modulating signal amplitude and logic for the p-channel MOSFETs and
frequency and observe the effect on the apply the signals to the inverter switches.
drive signals. (c) Observe the switching voltage waveform
across Q1 and Q2 and that across Q3 and Q4.
5. The circuit schematic shown in Figure 6.75 is (d) Observe the waveform during the switch-
a simple inverter for laboratory study wherein ing transitions. If the RC time constant of
the top switches are p-channel MOSFETs the delay circuit is reduced, what is the
and the bottom switches are n-channel MOS- effect during the switching transitions?
FETs. The DC link is a 15 V DC source. The (e) Observe the voltage across the output load.
p-channel MOSFETs can be driven with If the load resistor is replaced by a resistor
15 V
Q1 drive Q3 drive
Vo
A B
Q2 drive Q4 drive
Figure 6.75 Laboratory study inverter with simple non-isolated drive circuits.
Chapter 06.indd 362 3/12/2009 6:28:01 PM

in series with a capacitor, then what is the (f ) Apply bipolar and unipolar PWM to the
waveform across the capacitor? What is the above inverter and study the outputs with
effect of variation in the modulating signal R load, RC load and RL load. What are the
amplitude and frequency on the output current freewheeling paths in the inverter
waveform? for the RL load?

1. DC–AC switched-mode converters are called 12. The OFF-state voltage across the switches in a
. half-bridge inverter is the input DC
voltage.
2. The simplest inverter is composed of an
switch. 13. The voltage applied across the bridge arms in a
half-bridge inverter is the input DC
3. The three-phase inverter topologies consists of
voltage.
bridge arms.
14. The OFF-state voltage across the switches in a
4. Switching transitions occur from the
full-bridge inverter is the input DC
of the core and the characteristics of
voltage.
the power transistors.
15. The voltage applied across the bridge arms in a
5. In saturable-core inverter, the oscillation is not full-bridge inverter is the input DC
of the type. voltage.
6. The ideal saturable-core inverter generates a 16. For a given load power, the current rating
frequency that is proportional to the required for the power devices in the half-
input DC voltage. bridge configuration is that for the
7. The saturable-core inverter generates a wave- full-bridge configuration.
shape that is a with a duty 17. The quasi-square waveshape becomes a
cycle. square waveform when the dead band angle
8. Better can be achieved using the two is .
transformer saturable base drive inverter as 18. The efficiency of the duty cycle topology is
compared to the saturable output transformer than that of the DC-link control
topology. topology.
9. In driven inverters, the power switches are 19. The three-phase inverter topologies require
driven through control circuits where bridge structures.
the ON-times and the OFF-times are .
20. The three-phase inverter topologies can be built
10. The push–pull configuration is different from using three structures.
the other configurations in that the load is con-
nected of the SPDT switch. 21. If each arm of a three-arm bridge inverter
generates a wave and is displaced
11. The OFF-state voltage across the switches in with the waveform of the other two arms by
the case of a push–pull inverter is the , then a six-step waveform is obtained
input DC voltage. across the phases of the three-phase load.
Chapter 06.indd 363 3/12/2009 6:28:01 PM

22. The peak value of the six-step waveform that is 34. Modulation index is defined as the ratio of the
obtained across the phases of the three-phase peak amplitude of the signal to the
load is times the input DC-link peak of the signal.
voltage applied to the inverter.
35. In sinusoidal PWM, for low values of frequency
23. The current-controlled inverter uses the ratio, the relative harmonic content is .
to generate the switching signals for
36. To ensure that the number of pulses for the
the switches in the inverter bridge arms.
positive half and the negative half are same
24. One of the main drawbacks of the current- and symmetric the frequency ratio should be
controlled inverter is the in the an number.
switching frequency of the inverter switches.
37. If the frequency ratio for sinusoidal PWM is
25. In a current-controlled inverter as the hysteresis greater than 20, then the relative harmonic
band is made smaller, the switching frequency content of the bipolar PWM significantly
becomes . .
26. The main objective of the pulse-width modula- 38. In sinusoidal PWM, the harmonic content of
tion is to control the component of the harmonic number equal to the frequency
the pulse-width modulated waveform with ratio is quite .
respect to the applied DC voltage.
39. In a full-bridge inverter with sinusoidal PWM,
27. The modulation control index gives a measure the transformer across the bridge should be
of the component present in the designed for the frequency of the
output PWM waveform. signal.
28. The modulation index gives a measure of the 40. Space-vector modulation is a technique
relative peak amplitude of the signal employed for inverter systems.
with respect to the carrier signal.
41. Space-vector PWM provides a DC-
29. If the positive part of the waveform is a mirror link voltage utilization compared to the sinu-
image of the negative part, then the waveform soidal PWM.
is said to be about the time axis and
42. Space-vector PWM provides relative
such a waveform does not contain
harmonic content compared to the sinusoidal
harmonics.
PWM.
30. If the average value of the waveform is zero, the
43. The switching frequency of the inverter
Fourier series coefficient, is also zero.
switches is the carrier frequency in
31. If a waveform has odd symmetry about the space-vector PWM.
origin, the Fourier series coefficients for the
44. The switching loss is in space-vector
terms are all zero.
PWM-based system compared to a sinusoidal
32. In the case of three-phase inverters, the PWM PWM-based system with the same carrier
signal for an arm is generated by a reference frequency.
modulating wave that is phase shifted
45. If a constant amplitude space vector rotates at a
with respect to the reference modulating waves
constant speed, then the resulting waveforms
of the other two arms.
with respect to time on the three equally spaced
33. The ratio of the carrier frequency to the modu- spatial axes are and displaced in time
lating frequency is called the . with respect to each other by .
Chapter 06.indd 364 3/12/2009 6:28:01 PM

46. If three-phase sinusoidal waveforms of fre- 48. In harmonic elimination, at every notch, the
quency w and displaced with respect to each voltage transits from either positive to negative
other by 120o in time are applied at the three or negative to positive resulting in .
equally spaced spatial axes, then the resulting
49. Staircase PWM combines the principles of the
space vector is of amplitude and
PWM and the harmonic elimination
rotates at a constant speed of .
by method.
47. In a quasi-square waveform, there is only the
50. Staircase PWM gives modulation
dead band angle that can be used as a parame-
control index than the bipolar PWM.
ter either to control the amplitude or
to eliminate a .
1. What is the distinction between choppers, 11. What are the advantages of the saturable base
oscillators and inverters? drive inverter when compared to the saturable
output inverter?
2. How are the inverters classified based on the
manner in which the power semiconductor 12. Explain the operation of the saturable base
devices are switched? drive inverter.
3. Explain how a single pole double throw switch 13. What are the configurations generally used for
can be used as a two-level chopper and a two- driven inverter circuits?
level inverter.
14. Explain the operation of the push–pull
4. How many SPDT switches are needed for a inverter.
three- and four-level inverter? Explain.
15. Discuss the current freewheeling through the
5. What is the maximum number of the output body diodes of the switches in a push–pull
voltage levels for a two-arm bridge inverter? If inverter when the load is inductive.
the inverter is supplied by only one input DC
16. Explain the operation of the half-bridge inverter.
source, what is the possible number of levels
for the two-arm bridge inverter? 17. Discuss the current freewheeling through the
body diodes of the switches in a half-bridge
6. Discuss the multi-level inverter structures.
inverter when the load is inductive.
7. Discuss the three-phase inverter topologies.
18. Explain the operation of the full-bridge
8. Distinguish between driven and self-driven inverter.
inverters.
19. Discuss the current freewheeling through the
9. Explain the operation of the inverter wherein body diodes of the switches in a full-bridge
the output transformer is a saturable-core inverter when the load is inductive.
transformer.
20. Draw the quasi-square waveshape of amplitude
10. How does the saturable-core inverter start? 100 V and dead band angles of 60o and 30o.
Chapter 06.indd 365 3/12/2009 6:28:01 PM

21. Discuss the switching sequence for the switches 35. What is center pulse modulation?
of a full-bridge inverter to generate a quasi-
36. What is the difference between unipolar and
square waveshape output.
bipolar sinusoidal PWM waveform?
22. Give a block schematic to generate the signals
37. How can the size of the inverter’s output trans-
for the switches of a full-bridge inverter to
former be reduced?
generate a quasi-square waveform.
23. How is output regulation achieved for a quasi- 38. What is the principle of PWM through phase
square wave inverter? Explain the DC-link modulation? What are its advantages over sinu-
control and duty cycle control methods. soidal PWM?
24. Discuss the three-arm bridge inverter used for 39. Discuss the generation of the drive signals for
three-phase loads. switches of the phase-modulated inverter?
25. Explain the gate drive signals that need to be 40. What are the advantages of space-vector
provided for the switches of a three-phase PWM?
inverter to generate a six-step waveform. 41. What are the primary vectors in the case of a
26. Explain the various modes of the six-step three-phase PWM system?
inverter operation. 42. For a three-phase system, how many states can
27. Define modulation control index. be defined and what are they?
28. Define relative harmonic content. 43. What is space-vector hexagon? Into how many
sectors can it be divided? Discuss.
29. Distinguish between modulation control index
m and the modulation index M. 44. Discuss the sequencing of the inverter states
30. What are the various ways in which PWM is such that the inverter switching frequency is
classified? half the space-vector carrier frequency.
31. A waveform has a Fourier series given by 45. What is over-modulation?

4Vdc ⎛ sin 3ωt sin 5ωt sin 7ωt ⎞ 46. In space-vector PWM, if the time spent in the
⎜ sin ωt + + + +⎟
π ⎝ 3 5 7 ⎠ zero state is split into two equal halves and
appended on either side of the active bound-
and another waveform has a Fourier series
ing sector state times, then what is the effect
given by
on the harmonics of the modulating signal?
4Vdc ⎛ cos 3ωt cos 5ωt cos 7ωt ⎞ What is the shape of the modulating signal?
⎜ cos ωt − + − +⎟
π ⎝ 3 5 7 ⎠ Discuss.
Plot the two waveforms and explain the 47. Discuss space-vector modulation for multi-
difference. level bridge arm structures for the inverter.
32. In a quasi-square waveform, if the dead band 48. Discuss harmonic elimination method with
angle becomes zero, what happens to the wave- multiple notches?
form?
49. Discuss harmonic elimination method by stair-
33. What is end pulse modulation? case PWM?
34. What happens to the modulation control index 50. What is the difference between harmonic elim-
when the angular end pulse width is 60o for an ination method with multiple notches and
end pulse modulation? Why? staircase PWM?
Chapter 06.indd 366 3/12/2009 6:28:01 PM

| PROBLEMS
1. For a two-arm seven-level bridge inverter topol- 10. For a sinusoidal pulse-width modulation, the
ogy, what is the number of distinct source frequency of the triangular carrier is 20 kHz
potentials required? and the frequency of the modulating signal is
50 Hz. What is the slope of the triangular car-
2. A saturable output transformer inverter is sup-
rier?
plied from a 12 V battery. The output trans-
former core is a ferrite-based EE core with a 11. For a sinusoidal PWM having a modulation
core cross-sectional area of 31 mm2. The total index of 1 and the frequency ratio of 7, calcu-
number of turns on the primary is 200. If the late and compare the modulation control indi-
saturation flux density is 0.3 T, then calculate ces for unipolar and bipolar PWM.
the switching frequency of the inverter.
12. It is required to generate a space vector of
3. A saturable output transformer inverter having amplitude 0.5 at an angle of 15o (anti-clockwise)
an efficiency of 80% is driving a 25 W load. from the a-axis. How much time should the
The energy is drawn from a 12 V battery. Find inverter system spend in the bounding states to
the current and voltage rating of the primary- generate this equivalent space vector in a period
side switches. of 50 μs? What are the bounding states?
4. What is the modulation control index for a 13. It is required to generate a space vector of
square waveform? amplitude 0.5 at an angle of 150o (anti-clockwise)
5. A quasi-square waveform has a dead band angle from the a-axis. How much time should the
of 30o. What is its modulation control index? inverter system spend in the bounding states to
generate this equivalent space vector in a period
6. For a quasi-square waveform having a dead band of 50 μs? What are the bounding states?
angle of 60o, calculate the modulation control
index and tabulate the relative harmonic indices 14. It is required to generate a space vector at an
for the third, fifth, seventh and ninth harmonics. angle of 30o (anti-clockwise) from the a-axis
such that the vector tip just touches the hex-
7. For a quasi-square waveform having a dead agonal boundary. How much time should the
band angle of 30o, tabulate the relative har- inverter system spend in the bounding states
monic indices for the third, fifth, seventh and to generate this equivalent space vector in a
ninth harmonics. Compare with the relative period of 50 μs? What are the bounding
harmonic indices of the previous problem and states?
explain why the third and harmonics that are
multiples of three are insignificant. 15. In a quasi-square waveform, if the fifth har-
monic needs to be eliminated, then what is the
8. For an end pulse waveform having an angular dead band angle?
end pulse width of 30o, calculate the modula-
tion control index and tabulate the relative har- 16. Determine the number of notches and the
monic indices for the third, fifth, seventh and notch angles of a multiple notch harmonic
ninth harmonics. elimination PWM pattern such that the 5th,
7th and 11th harmonics are eliminated.
9. For a center pulse waveform having an angu-
lar center pulse width of 30o, calculate the 17. Determine the number of levels and the heights
modulation control index and tabulate the of the levels of a staircase PWM pattern such
relative harmonic indices for the third, fifth, that the 5th, 7th and 11th harmonics are elim-
seventh and ninth harmonics. inated.
Chapter 06.indd 367 3/12/2009 6:28:01 PM

| ANSWERS
Fill in the Blanks
1. inverters 18. better 35. high
2. SPDT 19. three-arm 36. odd
3. three 20. two-level single-arm 37. low
4. magnetic characteristics; static 21. square; 60o 38. high
5. LC resonance 22. 2/3 39. modulating
6. directly 23. hysteresis comparator 40. poly-phase
7. symmetrical rectangular; 50 24. variations 41. higher
8. efficiency 25. higher 42. lower
9. external; controllable 26. fundamental 43. half
10. across the throws 27. fundamental 44. lesser
11. twice 28. modulating 45. sinusoidal; 120o
12. same as that of 29. symmetric; even 46. constant; w
13. half 30. ao 47. fundamental; specified har-
14. same as that of 31. cosine monic
15. same as that of 32. 120o 48. voltage reversal
16. double 33. frequency ratio 49. sinusoidal; multiple notches
17. 0o 34. fundamental; carrier 50. higher
Chapter 06.indd 368 3/12/2009 6:28:01 PM

Design of Magnetics
Learning Objectives
CHAPTER
7
understand the principles of magnetic components.
design the magnetic components like transformers, inductors and current transformers
(CTs) for a given application.
M agnetic components are an important and integral part of any power electronic system. The magnetic
components are categorized into two broad classes: (a) energy-transfer devices and (b) energy-storage
devices.
The energy-transfer devices transfer the power from one energy port called the primary to another energy
port called the secondary without ideally storing or losing any energy in the process of transfer. These devices
are called the transformers. The transformers are further classified into two sub-classes: (a) potential trans-
formers (PTs) and (b) current transformers (CTs). In PTs, the energy or power that is drawn by the primary
is from a voltage source. This is translated as a scaled voltage to the secondary load. In the case of CTs, the
energy or power that is drawn by the primary is from a current source. This gets translated as a scaled current
at the secondary.
The energy-storage devices store the kinetic energy by virtue of a current flowing through it. These devices
are called inductors. This chapter will discuss both these classes of magnetic components and their design
principles.
7.1 Magnetic Concepts
T here are two fundamental laws of electromagnetism that are used to link the electric domain quantities
and the magnetic domain quantities. They are:
1. Ampere’s Law.
2. Faraday’s Law.
These two laws are the governing principles based on which all other relationships in the magnetic domain
are derived. Ampere’s law links the current through the coil of the electric domain and the magneto-motive
force (mmf ) of the magnetic domain. A current flowing through a loop of conductor consisting of N turns
produces a magnetic field of intensity H. This is depicted in Figure 7.1.
Chapter 07.indd 369 3/14/2009 12:53:58 PM

lm
H
N
i
Figure 7.1 Current through a loop producing magnetic field H.
The Ampere’s Law is expressed as

lm
mmf = ∫ Hdl = Ni (7.1)
0
where H is the magnetic field intensity (A/m); lm the magnetic path length (m); N the number of turns in
the loop or coil; i the current in the coil. If the magnetic field intensity H is uniform along the magnetic
path length, then Eq. (7.1) reduces to
Hl m = Ni
Ni mmf
H= = (7.2)
lm lm
Faraday’s law links the voltage induced across the coil in the electric domain and the rate of change of flux
in the magnetic core. Referring to Figure 7.1, any change in the flux f in the core induces an electromotive
force (emf ) e across the coil. The induced voltage is proportional to the rate of change of flux in the mag-
netic core. Faraday’s law is expressed as
dφ
emf = e = N (7.3)
dt
Figure 7.2 illustrates the energy interaction between the electric and the magnetic domains. The interac-
tion between the two domains is through an energy port. Energy can flow from the electric domain to the
magnetic domain and vice-versa. The energy or power variables in the electric domain are the voltage e and
the current i. The corresponding power variables in the magnetic domain are the mmf and the rate of change
of flux df/dt.
From the Ampere’s law given in Eq. (7.1),
mmf
N=
i
Substituting for N in the Faraday’s law given in Eq. (7.3), one obtains
dφ
e × i = mmf × = Power (7.4)
dt
Chapter 07.indd 370 3/13/2009 7:08:42 PM

Design of Magnetics 371
Electric Magnetic
domain N domain
e, i mmf, df
dt
Figure 7.2 Interaction between electric and magnetic domains.
Thus from Eq. (7.4), it can be observed that e is the potential variable, i is the flow variable in the electric
domain, mmf is the potential variable and df/dt is the flow variable in the magnetic domain. Further observe
that the product of the potential variable and the flow variable is always power. Thus, based on Eqs. (7.1),
(7.3) and (7.4), the following analogies, as given in Table 7.1, can be expressed between the variables of the
electric domain and that of the magnetic domain.
Dissipative Component
In Table 7.1, the resistance in the electrical domain is a dissipative component. It dissipates energy as heat.
If i is the root mean square (rms) current flowing through R, then
T
1
T ∫0
i2R = e × i × dt = Power dissipated (7.4a)
Table 7.1 Analogous variables in electric and magnetic domains

S. No. Electric domain Magnetic domain
1 e, v – voltage (volts) mmf – magneto-motive force (amp-turns)
2 i – current (amps) df/dt – rate of change of flux (weber/s)
3 Volts × Amperes = Watts Ampere-turns × Weber/s = Watts
4 Resistance – R =
∫
(1 / T ) v × i × dt
Rmag =
∫
(1 / T ) mmf (dφ / dt )dt
=
∫
(1 / T ) Hl m dφ
2 2
i (dφ / dt ) (dφ / dt )2
5 Charge – q = idt∫ φ=
dφ
∫ dt dt
1 1 dφ 1
C∫ Λ ∫ dt
6 Capacitance effect – v = idt Ni = mmf = dt = ∫ dφ
Λ
where Λ is permeance
1
ℜ= = Reluctance
Λ
Chapter 07.indd 371 3/13/2009 7:08:42 PM

In the magnetic domain, the power dissipated is given as

T
1 dφ 1
T ∫0
Pdis = mmf × × dt = ∫ H × l m × Ac × dB
dt T
where Bm is the maximum flux density discussed in the next sub-section. ∫ H × dB is the area of the B–H
loop. This means that in a period T the energy corresponding to the area of the B–H loop is dissipated. Then
an equivalent dissipative component Rmag can be defined in the magnetic domain wherein
2 Bm
⎛ dφ ⎞ Vc
⎜ ⎟ × Rmag =
⎝ dt rms ⎠ T ∫ H × dB (7.4b)
− Bm
where Vc = Ac × l m is the core volume.
Flux and Flux Density

Figure 7.3 illustrates the concept of the flux density. The flow in the magnetic domain is the rate of change
of flux. The magnetic core has a cross-sectional area Ac as shown. The flow df/dt is in a direction normal to
the cross-sectional area. The flow per unit cross-sectional area is called the flow density. It is given as
1 dφ d(φ / Ac ) dB
× = = (7.5a)
Ac dt dt dt
where B = φ / Ac is called the flux density. The magnetic field intensity H gives rise to a certain flux density
B (tesla) in the core. The relationship between B and H is given by
B = mH (7.5b)
where m is called the permeability of the core and is given as μ = μo . μ r . mo is the permeability of free space
which is equal to 4p × 10–7 H/m and mr is the relative permeability of the core material. Note that m is not
a constant quantity. It has been found to vary non-linearly with H. Graphically this relationship between B
and H is as shown in Figure 7.4. Figure 7.4(a) shows the B–H curve that is linearized. The slope of the B–H
curve is the permeability m. Beyond a specific magnetic intensity, the flux density B clamps to a value Bsat
and does not increase any further. At this instant the core is said to have saturated. When the core is in the
saturated condition, the permeability m is zero. When the core is saturated, for any change in the magnetic
field intensity H, there is no change in B and therefore f. Thus df/dt is zero and the power associated with
the saturated core in the magnetic domain is zero. The hysteresis characteristics of practical cores are depicted
df /dt
Ac
i
Figure 7.3 Flux density.
Chapter 07.indd 372 3/13/2009 7:08:42 PM

Table 7.2 Saturation flux density of common core materials

Core Type Bsat (T)
Ferrite 0.3
Si steel 1.2 (CRGO), 1 (CRNGO)
Powdered iron 1.6
Amorphous glass 1.6
Mu metal 1
Notes: CRGO – cold rolled grain oriented; CNRGO – cold rolled non-grain oriented.
in Figure 7.4(b). Observe that a practical core exhibits hysteresis. The area within the hysteresis loop is the
energy lost in every cycle and is called the hysteresis loss. The saturation flux density Bsat is a property of the
material used. Table 7.2 gives the saturation flux density of common core materials.
The charge in the electrical domain is the integral of the flow variable, that is current. The integral of the
flow variable in the magnetic domain is f. From the Ampere’s law [Eq. (7.2)] and the relationship as given
in Eq. (7.5), one obtains
Ni
μ ⋅ Ac ⋅ φ =
lm
Ni Ni
φ= = = mmf ⋅ Λ (7.6)
where l m / ( μo ⋅ μr ⋅ Ac ) ℜ
ℜ = l m / ( μo ⋅ μ r ⋅ Ac ) ⇒ Reluctance
1
Λ= ⇒ Permeance
ℜ
B B
Bsat Bsat
m m
0 H 0 H
−Bsat −Bsat
(a) (b)
Figure 7.4 (a) B–H curve (piece-wise linear); (b) practical B–H curve or B–H loop.
Chapter 07.indd 373 3/13/2009 7:08:43 PM

Volt-Second Balance
The flux f can be expressed in terms of the properties of the core as
φ = Ac B (7.7)
The core cross-section area Ac is a constant for a specified core. However, the flux density varies depending
on the level of the magnetic field intensity H. From the Faraday’s law, the induced emf is given by Eq. (7.3).
Using Eq. (7.7), the Faraday’s law can be re-written as
dB
e = NAc (7.8)
dt
The flux density B for the core can be obtained by integrating the above equation. Thus,
1
NAc ∫
B= e ⋅ dt (7.9)
From Eq. (7.9) it can be observed that if the average value of e is non-zero, then ∫ e ⋅ dt accumulates towards
infinity. This implies that the flux density B in the core continuously increases with time. However, from
Figure 7.4 it is clear that there is a limit on the value of B, as given by Bsat. Therefore, for ∫ e ⋅ dt to be finite,
the average value of e must be zero. If the frequency of the induced emf (e) is f, then the flux density that
builds up in half a period can be estimated from Eq. (7.9) as
1 ⎛2 ⎞ T 1 T
B= ⎜ ∫ e ⋅ dt ⎟ = E avg (7.10)
NAc ⎝T ⎠ 2 NAc 2
where T = 1/f ; Eavg is the average voltage of e in a period T/2. Thus,
E avgT = 2 NAc B (7.11)
Equation (7.11) defines the volt-second product for a coil wound on a core. For a transformer of a given
rated voltage, if a lower frequency is applied, the EavgT will increase and may saturate the core. When the
core saturates, the windings offer zero impedance, which in turn causes huge currents to flow in the wind-
ings and the source, damaging either the source or the windings. For example, consider a transformer with
the following rating: Primary: 230 V, 50 Hz.
If one applies 10 Hz at 230 V, then as the period at 10 Hz is higher than that at 50 Hz, from Eq. (7.11)
it is clear that the core will tend to saturate. Therefore, one should never apply a voltage of frequency lower
than that it is designed for. However, if the voltage is reduced proportionately, then a lower frequency can
be applied. This means that, as given by Eq. (7.11), as long as one does not exceed the volt-time product or
the volt-second product for the winding, the transformer will not saturate.
For one complete cycle the flux or the flux density should not build up. This means that the voltage
across the coil should have a zero average over a full cycle. Thus
1 ⎛1 ⎞ 1
B= ⎜ ∫ e ⋅ dt ⎟ T ; ∫ e ⋅ dt = 0
NAc ⎝ T ⎠ T
The above equations should be satisfied to ensure that the flux density B does not build up over a full cycle.
This means that a magnetic winding will not support an average voltage across it. The average voltage across the
magnetic winding under equilibrium conditions will always be zero. This is called the principle of volt-second
balance.
Chapter 07.indd 374 3/13/2009 7:08:43 PM

7.2 Inductor
Inductor Value
Equation (7.6) gives the relationship between flux, mmf and permeance. Thus,
φ = Λ ⋅ N ⋅i
Differentiating the equation and multiplying by N, one obtains
dφ di
N = Λ⋅N 2 ⋅ (7.12)
dt dt
From the Faraday’s law as given in Eq. (7.2), N (dφ / dt ) is the induced voltage e across the coil. Thus,
di di
e = Λ⋅N 2 ⋅ = L⋅ (7.13)
dt dt
Equation (7.13) is the derived form of Faraday’s law as viewed from the electric domain. As the induced
voltage e is the same whether viewed from the magnetic domain or the electric domain, the inductor value
L is given by comparing Eqs. (7.12) and (7.13). L can be expressed in terms of the physical properties of the
core by substituting the expression for permeance from Eq. (7.6) into Eq. (7.13):
μo ⋅ μr ⋅ Ac ⋅ N 2
L = Λ⋅N 2 = (7.14)
lm
Energy Storage
In an inductor, the energy is stored by virtue of the current flowing through the inductor in the electrical
domain. If L is the value of the inductor as viewed in the electrical domain, then the energy EL stored in the
inductor when a current i flows through it is given by
1 2
EL = Li (7.15)
2
The same energy EL as viewed within the magnetic domain is
2
1 2 1 ⎛ mmf ⎞ 1
EL = Li = L ⎜ ⎟ = ⋅ Λ ⋅ mmf
2
(7.16)
2 2 ⎝ N ⎠ 2
where Λ = L / N 2 from Eq. (7.14). Thus, it can be observed from Eq. (7.16) that the energy-storage mech-
anism within the magnetic domain is a potential storage similar to that of capacitance storage in electric
domain. Thus, the permeance is equivalent to a capacitance within the magnetic domain. From Eq. (7.6),
the mmf dynamics is given as
1 1 1 dφ
mmf = ⋅ φ = ⋅ ∫ dφ = ⋅ ∫ ⋅ dt (7.17)
Λ Λ Λ dt
Further, compare Eq. (7.17) with the following electric domain variables:
1 1 1 dq 1
v= ⋅ q = ⋅ ∫ dq = ⋅ ∫ ⋅ dt = ⋅ ∫ i ⋅ dt (7.18)
C C C dt C
Chapter 07.indd 375 3/13/2009 7:08:43 PM

where q is the charge. From Eqs. (7.17) and (7.18), it can be observed that the permeance Λ is like a capaci-
tance and the energy-storage mechanism within the magnetic domain is like a capacitive energy storage, that
is, potential energy storage even though it appears as a kinetic or inductive energy storage as viewed in the
electric domain. For a given core with a specified core cross-section area Ac and magnetic length lm, the per-
meance is dependent on the permeability of the core m. Beyond a particular value of magnetic field intensity,
Ni/lm, the core will saturate. The permeability of the saturated core is zero. Therefore, the permeance becomes
zero and the core in saturated state is incapable of storing energy.
An air gap is generally introduced for inductive applications wherein the core needs to store energy. If an
air gap of lg is introduced in the magnetic path, then the reluctance of the magnetic path is given as
lm lg 1 ⎛ lm ⎞
ℜ= + = ⎜⎜ + l g ⎟⎟ (7.19)
μo μr Ac μo Ac μo Ac μ
⎝ r ⎠
The relative permeability mr is very large as compared to lm. Therefore, l m / μ r l g . The reluctance given in
Eq. (7.19) can be written as
lg
ℜ≈ (7.20)
μo Ac
The permeability mo is very small compared to lg. This means that the reluctance will increase on intro-
ducing an air gap. Figure 7.5 depicts the change in the slope of the B–H curve on the introduction of the
air gap within the core. The increase in reluctance reflects as a decrease in the effective permeability. Note
that mg is the permeability of the core with the air gap of length lg introduced. Referring to Figure 7.5, the
core without air gap having a permeability of m1 will saturate at current i1. However, on introducing an
air gap the core will saturate at current ig. Thus for a given energy-storage capability of the core, the core
with air gap will handle higher mmf or, in other words, allow higher current capability for the windings.
Alternately, one may also argue that increasing the reluctance by introducing the air gap implies that the
permeance has decreased. The energy-storing capability of the core is given by Eq. (7.16). Therefore, for a
given energy-storage requirement, the permeance should decrease to handle higher mmf.
With air gap mg

m1
0
Ni1 Nig Ni
H=
lm1 lmg lm
No air gap
Figure 7.5 Permeability change on introduction of air gap.
Chapter 07.indd 376 3/13/2009 7:08:43 PM

Area Product
An inductor consists physically of a winding and a magnetic core. The magnetic core includes the magnetic
and physical properties of the material like the flux density, core cross-sectional area, window area, magnetic
length, permeability, etc. On the other hand, the electrical side of the inductor defines the current through
the inductor, the amount of energy that the inductor requires to store, the voltage across the inductor, etc.
To design the inductor, the electrical and the magnetic parameters should be related to select the core, the
winding gauge, the number of windings and the air-gap length that needs to be incorporated.
The area product approach is a systematic magnetic component design method. In this method, two
physical parameters of the core, namely, the core cross-sectional area Ac and the window area Aw are related
to the electrical parameters. Figure7.6 depicts the core cross-sectional area Ac and the window area Aw for a
few standard core shapes.
Aw
Ac
(a)
Aw
Ac
(b)
Aw
Ac
(c)
Aw
Ac
(d)
Figure 7.6 Illustration of the window area Aw and the core cross-sectional area Ac for a few core
shapes: (a) EI core; (b) C core; (c) toroid; (d) EE core.
Chapter 07.indd 377 3/13/2009 7:08:43 PM

The Faraday’s law of Eq. (7.8) relates the Ac to the voltage across the coil. The window area Aw is a mea-
sure of the number of turns of a given wire cross-section that can be accommodated. This cross-section is
dependent on the current-carrying capacity or the rms current that flows through the wire. Thus Aw is
related to the current capacity. The product of Ac and Aw is called the area product Ap. The area product
should therefore be related to the energy that is stored in the core or the power that is transferred through
the core. For energy-storing devices like the inductor, the area product Ap is related to the energy that the
core will store. For energy-transferring devices like the transformers, the area product Ap is related to the
power that will be transferred through the core.
Window Area
If “a” is the cross-section of the wire used and N is number of turns then the window area of the core should
accommodate “N” turns of wire cross-section area “a”. Thus
I rms
K w Aw = N ⋅ a = N ⋅ (7.22)
J
where Kw is the window utilization factor and has a value less than 1; J the current density (expressed as
A/m2).
The window utilization factor Kw accounts for the loss of the available window area due to various fac-
tors like insulation, air gaps, wire enamel, creepage, etc. The window utilization factor is discussed in detail
in a later section.
The peak value Im and the rms value Irms of the current flowing through the inductor is related as
Im
Kc = (7.23)
I rms
where Kc is called the crest factor for the particular current waveshape through the inductor. Substituting
Eq. (7.23) into Eq. (7.22), one obtains
K w K c Aw J = N ⋅ I m (7.24)
Equation (7.24) relates the window area (Aw) parameter of the core to the electrical quantity which is the
peak value of the current flowing through the inductor.
Cross-Section Area
From Faraday’s law
di dφ dB
e=L =N = NAc (7.25)
dt dt dt
Integrating Eq. (7.25), one obtains
LI m = NAc Bm (7.26)
where Bm is the maximum operating flux density in the core that corresponds to the maximum current Im
through the winding. Equation (7.26) relates the core cross-sectional area Ac to the current through the
inductor and the flux density.
The energy that needs to be stored in the inductor is given as
Chapter 07.indd 378 3/13/2009 7:08:43 PM

1
EL = LI 2 (7.27)
2 m
This can be re-arranged as
1
EL = ⋅ ( LI m ) ⋅ I m (7.28)
2
In the energy equation [Eq. (7.28)], LIm is substituted with the core cross-sectional area relationship of Eq.
(7.26) and Im is substituted with the window area relationship of Eq (7.24). Thus,
2E L
Ap = A w Ac = (7.29)
K w K c JBm
Equation (7.29) is the basic area product equation for the energy-storing devices that is used for the selec-
tion of the core for a given energy-storage requirement and electrical/magnetic operating conditions.
Design of Inductors
The following steps illustrate the design for an energy-storing device, that is, the inductor.
Step 1 (Inductor Value): The value of the inductor has to be calculated. This is done based on the cir-
cuit topology where the inductor is used. For example, in the case of DC–DC converter applications,
the inductor value is calculated based on the discussions in Chapter 5. For LC resonant circuits, the
value of L is calculated based on the resonance frequency and the quality factor of the coil.
Step 2 (Area Product): From the current waveform through the inductor, the energy-storage require-
ment for the inductor is estimated. The energy that needs to be stored in the inductor is given as
1
EL = LI 2 (7.30)
2 m
where EL is the maximum energy in joules that needs to be stored in the inductor and Im is the peak
inductor current in amperes. Based on the energy to be handled by the inductor core, the area prod-
uct is given as
2E L
Ap = A w Ac = (7.31)
K w K c JBm
where KW, J and Bm are design parameters that have to be chosen by the designer for a particular appli-
cation. The starting values or default values for these parameters are: Also, Kw = 0.6 for single-winding
inductors and 0.3 to 0.4 for multiple-winding inductors like the flyback transformer discussed in
Chapter 5. Also J = 3 × 106 A/m2, Bm = 0.25 T for ferrites, 1 T for cold rolled non-grain oriented
(CRNGO) cores, 1.2 T for cold rolled grain oriented (CRGO) cores, 1.5 T for amorphous cores. The
selected core should have an area product greater than that calculated by Eq. (7.31). Appendix II gives
a representative list of commonly used core types and their physical parameters.
Step 3 (Permeance): The reluctance of a core with air gap lg is given in Eq. (7.19). The permeance is the
inverse of the reluctance and is given as
Chapter 07.indd 379 3/13/2009 7:08:44 PM

μo μr Ac
Λ= (7.32)
l m + μr l g
where Ac is the core cross-sectional area of the selected core (in m2); mr is the relative permeability of the
selected core material; lm is the magnetic path length of the selected core (in m); lg is the introduced
air-gap length (in m). The air-gap length is a design parameter that is chosen by the designer.
Step 4 (Number of Turns): From Eq. (7.14) it can be observed that the permeance is related to the
inductance and the number of turns as
Λ =L/N2
The permeance has the units of inductance per turn2 or H/turns2. Commercially, cores are available
specifically for inductor applications wherein the gap is prefixed or the effective permeance is adjusted
at the time of sintering the ferrite core material. The permeance of such cores is available in the data-
sheet as AL factor which is expressed in nH/turns2. Re-arranging the above equation, one obtains
L
N= (7.33)
Λ
Step 5 (Gauge of Wire): The cross-section area of the wire can be calculated as
I rms
a= (7.34)
J
where Irms is the rms value of the current flowing through the inductor which is equal to Im/Kc. The
gauge of the wire selected should have a cross-section area that is greater than that calculated by
Eq. (7.34). Appendix III gives a representative list of commonly used wire gauges.
Step 6 (Available Window Area Check): The inequality AwKw > aN should be satisfied or else repeat
the calculations for the number of turns and gauge of wire after choosing the next bigger core. Note that
the value of “a” used for checking the above inequality should be the actual cross-section area of the
wire used and not the calculated value as per Eq. (7.34).
Problem 7.1
Design an inductor for a buck converter configuration as discussed in Chapter 5 for the following specifications:
Output voltage, Vo = 3.3 V
Output current, Io = 5 A
Switching frequency, fs = 20 kHz
Input voltage, Vi = 10 V ± 10%
Solution
Step 1: Determine L
The inductor value for this converter is given by Eq. (5.9) as
V ⋅ (1 − Dmin )
L= o
ΔiL ⋅ f s
DiL is the current ripple in the inductor. Take DiL = 10% of Io = 0.5 A.
As discussed in Chapter 5 and from Eq. (5.8), the minimum duty ratio is given as
Chapter 07.indd 380 3/13/2009 7:08:44 PM

Vo
Dmin = = 0.3
Vimax
Substituting the appropriate values in the above equations, the value of the inductor is
L = 0.231 mH
Step 2: Area Product
The energy and area product calculations are as follows
Δi
I m = I o + L = 5.25 A
2
1
EL = LI m 2
2
1
= × 0.231× 10−3 × 5.252 = 3.1834 × 10−3 J
2
2E L
Ap = A w Ac =
K w K c JBm
As the switching frequency is 20 kHz, a core with high resistivity has to be chosen to reduce the eddy
current losses. A ferrite core material would be an appropriate choice. Take Bm = 0.25 T for ferrite, J =
3 × 106 A/m2, K c ≈ I m / I o and Kw = 0.6. Substituting the values in the area product equation, one
obtains
Ap = 1.34747 × 10–8 m4 = 13474.7 mm4
Select a core from Appendix II which has an Ap higher than the value calculated above. Let P 36/22
be selected (Ac = 201 mm2, Aw = 101 mm2, Ap = 20100 mm4, lm = 53.2 mm).
Step 3: Permeance
To calculate the permeance, the relative permeability is taken from the datasheet for the core mate-
rial that is being used. A representative value from the datasheet for CEL HP3C grade ferrite is
given as
μr = 2000 ± 25%
The relative permeability has a wide tolerance band. At the lower tolerance value of relative permea-
bility, the permeance value is minimum which results in lower value of inductance for a specified
number of turns. Therefore, for worst case design, use the lower tolerance value of relative permeabil-
ity if the inductance value should be atleast the value as specified in Step (1). Use the higher tolerance
value of relative permeability if the inductance value should be atmost the value as specified in Step
(1). For this example the lower tolerance value of relative permeability is used for calculating the
permeance.
The air-gap length is chosen as 0.5 mm. Incorporating these values in the permeance equation
gives
μμ A 4π × 10−7 × 1500 × 201 × 10−6
Λ= o r c = = 4.717 × 10−7 H/turns 2
l m + μ r l g 53.2 × 10−3 + 1500 × 0.5 × 10−3
Step 4: Number of Turns

The number of turns can now be calculated from the permeance and inductance relationship. It is
given as
Chapter 07.indd 381 3/13/2009 7:08:44 PM

L 0.231 × 10−3
N= = = 22.13
Λ 4.717 × 10−7
Rounding up to the next higher integer if the calculation does not give an integer value, the number of
turns N is 23.
Step 5: Wire Gauge Selection
The gauge of the wire is now estimated as follows: In Step (3) the value of current density J used is
J = 3 × 106 A/m2. The cross-section area of the wire is calculated as
I rms 5
a= = = 1.67 × 10−6 m 2 = 1.67 mm 2
J 3 × 10 6
Referring to Appendix III, select a wire gauge whose cross-section area is greater than that calculated
above.
SWG 16 is selected (a = 2.075 mm2)
Step 6: Available Window Area Check
The inequality AwKw > aN has to be checked.
AwKw = 101 × 0.6 = 60.60 mm2 and aN = 2.075 × 23 = 47.725 mm2
The inequality is satisfied, which means that the windings will fit into the available window area.
Multiple-Winding Inductors
In certain applications, inductors can have more than one winding. One winding may be used to store
energy into the core and another winding may be used to remove energy from the core. In some applica-
tions, the energy stored in the core may need to be shared with more than one load. In all such cases, the
inductor will have multiple windings.
One of the more popular multiple-winding inductor is the flyback transformer. The flyback transformer
is a misnomer. It is not a transformer in the sense that it is an energy-storing device by function. Therefore,
the flyback transformer that is used in the flyback DC–DC converter topology (discussed in Chapter 5) is
nothing but a multiple-winding inductor.
The design of the multiple-winding inductor is very similar to the single-winding inductor. The winding
which is connected to the source that will transfer the energy to be stored in the core, is the main winding.
The inductor design is done with respect to the main winding assuming that the main winding is the only
winding. The difference is in deciding the number of turns for the other windings and the available window
area as the window utilization factor is lower in multiple-winding inductors. The value of Kw is between 0.3
and 0.4 for multiple-winding inductors. All the design steps up to Step (3) are the same as for the single-
winding inductor. There are minor modifications from Step (4) onwards. They are:
Step 4 (Number of Turns): Permeance is related to inductance and the number of turns as
L
N= (7.35)
Λ
The number of turns N is for the main winding. The other windings called the subsidiary or the secondary
windings are calculated based on the turns ratio obtained from the output voltage requirements and/or circuit
topology. If n1, n2, …, nm are the turns ratio of the m secondary windings that share the same energy of the
core, then
Chapter 07.indd 382 3/13/2009 7:08:44 PM

N s1 = n1 ⋅ N , … , N sm = nm ⋅ N
Step 5 (Gauge of Wire): The cross-section area of the wire can be calculated as
I rms
a= (7.36)
J
where Irms is the rms value of the current flowing through the main winding of the inductor which is
equal to Im/Kc. If Irms-i is the rms value of the current flowing through the ith secondary winding, then
I rms-1 I
as1 = , … , asm = rms-m
J J
Step 6 (Available Window Area Check): The inequality

m
Aw K w > a ⋅ N + ∑ asi ⋅ N si (7.37)
i =1
should be satisfied or else repeat the calculations for the number of turns and gauge of wire after choos-
ing the next bigger core. Note that the value of “a” and “asi” used for checking the above inequality
should be the actual cross-section areas of the wires used and not the calculated values as per Step (5).
7.3 Potential Transformer
U nlike an inductor, a transformer does not store energy. It only transfers the energy from one energy port
to another. Therefore, a transformer must have at least two windings. The winding that is connected to
the energy source is called the primary winding. The other winding(s) to which the load(s) is connected is
called the secondary winding(s).
df df
dt dt
RL1
Ns1
Vp Np Ns R Vp Np
Nsm
Core
RLm
(a) (b)
Figure 7.7 (a) Single secondary transformer; (b) multiple secondary transformer.
Chapter 07.indd 383 3/13/2009 7:08:44 PM

If the power source in the primary is a voltage source then the transformer is called a PT. If the power
source in the primary is a current source then the transformer is called a CT. The principle of operation for
both the transformers is the same though the causal variables are different. Figure 7.7(a) shows a basic trans-
former consisting of the primary winding Np connected to a voltage source Vp, a secondary winding Ns
connected to a load resistor R. Both the windings are wound on a common core that shares the same flux.
Figure 7.7(b) shows a transformer with multiple secondary windings that are connected to individual loads.
All the windings share the same core.
Operating Principle
On applying an AC voltage to the primary, current starts flowing through the primary winding. This pro-
duces a magnetic field around the coil, the intensity (H ) of which is given by the Ampere’s law. The mag-
netic field around the coil results in an mmf (Ni) within the core. The mmf causes a flux change within the
core. All the windings linked to the core experience the flux change and by Faraday’s law a voltage is induced
across each secondary winding.
The operating principle of the transformer is illustrated in Figure 7.8. Consider an AC voltage source Vi
applied to the primary winding Np. A current ip flows through the primary winding to set up an mmf in the
core. The mmf due to the current flowing in the primary winding is given as
mmf p = N p ⋅ ip (7.38)
The mmf in the core sets up a flux change rate df/dt. The rate of change of flux is decided by the voltage
across the winding that is connected to the voltage source. In this case, the primary winding is connected to
a voltage source Vi. Therefore,
ep = Vi
and from Faraday’s law
df ep
=
dt Np
ip is
mmfp mmfs
ep Np ip Np Ns Ns is es RL
Vi
Figure 7.8 Operating principle of the transformer.
Chapter 07.indd 384 3/13/2009 7:08:44 PM

dφ e p V
= = i (7.39a)
dt N p N p
It can be observed from Eq. (7.39a) that the flux rate depends only on Vi and Np and therefore is fixed by
the input source voltage alone for a given number of primary turns. If a load resistor RL is connected across
the secondary winding, a current is flows through the load RL. An opposing mmf is produced in the core by
virtue of the secondary current flowing in the secondary winding. This mmf is given as
mmf s = N s ⋅ is (7.40)
If the load resistor RL decreases, the secondary current is increases. This causes an increase in the oppos-
ing mmfs. As df/dt is fixed by the input voltage source, the primary mmfp will increase to maintain df/dt
according to Eq. (7.39a) within the core. The mmfp will increase by drawing more primary current ip from
the voltage source Vi. Thus more power is drawn from the source due to mmfs acting as the loading effect on
the primary mmfp.
If there is no load on the secondary, then is is zero and therefore mmfs is zero. The current drawn from
the input source Vi is called the no-load primary current ipo. The current ipo drawn from the source now
consists of two components: (a) im to set up the magnetic field; (b) ic to supply the losses within the core like
the hysteresis and eddy current losses. In most practical transformers the current drawn from the source
under no-load condition is about 5–10% of the current at full load.
In the presence of the secondary load, the current drawn from the source consists of three components:
(a) im to set up the magnetic field; (b) ic to supply the core losses and (c) the load reflected current to account
for the mmfs.
The flux rate df/dt is caused only by the mmf developed due to the component of the primary current im
that is used to set up the flux. This does not depend on the load as can be observed from Eq. (7.39a). Equation
(7.39a) can be expressed in terms of circuit parameters like inductance and currents based on Eq. (7.13) as
dim e p V
= = i (7.39b)
dt LM LM
where LM is the magnetizing inductance and is given as Λ ⋅ N p2 from Eqs. (7.12) and (7.13).
Figure 7.8 also shows the secondary winding that is wound over the same core. As the secondary wind-
ing is wound on the same core, the same flux rate also links the secondary windings. The voltage induced
across the secondary winding according to the Faraday’s law is given as
dφ ep
es = N s ⋅ = Ns ⋅ (7.41)
dt Np
Turns Ratio
On re-arranging Eq. (7.41), one obtains
Ns
es = ⋅ e = n ⋅ ep (7.42)
Np p
Equation (7.42) relates the primary and the secondary induced voltages. The variable “n” gives the ratio of
the number turns in the secondary winding to the number of turns in the primary winding. Thus from
Eq. (7.42),
Chapter 07.indd 385 3/13/2009 7:08:44 PM

N s es
n= = (7.43)
N p ep
Here “n” is called the turns ratio. The primary mmfp has to account for the opposing load reflected mmfs and
the mmf required to set up the magnetic field and the core losses. Thus,
mmf p = N p ⋅ ip = N s ⋅ is + N p ⋅ ipo (7.44a)
⎛ N ⎞
N p ⋅ ⎜ ip − s ⋅ is ⎟ = N p ⋅ ipo
⎜ Np ⎟
⎝ ⎠
ip − n ⋅ is = ipo (7.44b)
Equation (7.44b) gives the no-load current that is used for setting up the magnetic field within the core and
supplying the core losses.
Referring to Eq. (7.44a), if the transformer is designed such that ipo << ip then
N p ⋅ ip ≈ N s ⋅ is
and Eq. (7.43) can be extended to include the current ratio also as
N s e s ip
n= = = (7.45)
N p e p is
Leakage Flux Linkage

The permeability of the core is somewhat similar to conductivity in an electric circuit. Just as no conductor
is a perfect conductor of electricity, no magnetic material is a perfect conductor of magnetic field. As the
permeability of a core is never infinity, but has a finite value, the permeance of the core and the air are like
two parallel impedance paths in an electric circuit. One part of the flux rate flows within the core and
another part of it flows in the air as indicated in Figure 7.9. The part of the flux rate that flows through the
air path is called the leakage flux linkage as it does not link with any other winding except the primary
winding.
For a current ip flowing through the primary winding of Np turns, the mmfp that is set up is Npip. This
mmfp causes a flux rate df/dt in the core of permeability μo μ r and a flux rate dfsp/dt in the air path of per-
meability μo . From Faraday’s law,
Vi dφ dφσ p
= +
N p dt dt
dφ dφσ p
Vi = N p ⋅ + Np ⋅ (7.46)
dt dt
From Eqs. (7.13) and (7.44b), the above equation can be expressed equivalently with respect to circuit
parameters like inductances and currents as
Chapter 07.indd 386 3/13/2009 7:08:44 PM

df
dt
dfσp
ip dt
Np
Vi
Figure 7.9 Illustrating leakage flux linkage.
dim dip
Vi = LM ⋅ + Lσ p ⋅ (7.47)
dt dt
where the current im is responsible for setting up the field in the core, LM is called the magnetizing induc-
tance which represents the part of flux rate that flows through the core and Lsp is called the primary-side
leakage inductance which represents the part of flux rate that flows through the air. The primary leakage
inductance Lsp is given as Λ σ p ⋅ N p2 where Λ σ p is the permeance of the primary-side mean leakage path.
If the secondary winding is connected to a load, then a current is flows through it. The opposing mmfs is
Nsis. The mmfs drives a flux rate dfss/dt to flow through the air which will not link with the primary wind-
ing. Thus, on the secondary winding,
e s dφ dφσ s
= −
N s dt dt
dφ dφ
es = N s ⋅ − N s ⋅ σs (7.48)
dt dt
From Eq. (7.13), the above equation can be expressed equivalently in terms of circuit parameters like induc-
tances and currents as
Ns dφ dφ
es = ⋅ Np − N s ⋅ σs
Np dt dt
dim di
e s = n ⋅ LM ⋅ − Lσ s ⋅ s (7.49)
dt dt
Chapter 07.indd 387 3/13/2009 7:08:45 PM

df
dt
dfσp
ip dt is
Ns es RL
Np
Vi
dfσs
dt
Figure 7.10 Illustrating leakage flux linkage in primary and secondary.
where the current im is responsible for flux rate df/dt that links with the secondary windings, LM is called
the magnetizing inductance which represents the part of flux rate that flows through the core and Lss is the
secondary-side leakage inductance which represents the part of flux rate that flows through the air as depicted
in Figure 7.10. The secondary leakage inductance Lss is given as Λ σ s ⋅ N s where Λ σ s is the permeance of
2
the secondary-side mean leakage path.
Equivalent Circuit
The transformer is a component wherein the energy flows from an electric domain through an energy port
into the magnetic domain and from the magnetic domain to another electric domain through another
energy port. It will be easier to visualize the operation of the transformer and obtain greater insights by
replacing the transformer with an equivalent circuit representation. To develop the equivalent circuit, one
can build equivalence for an ideal transformer to start with and proceed to extend the equivalence towards a
practical transformer.
An ideal transformer has the following characteristics:
1. Core with Infinite Permeance or Infinite Permeability: This implies that the B–H curve is along the
vertical axis. No magnetizing current is required to set up the magnetic field, that is, im = 0. Infinite
permeance also implies that there is no leakage flux linkage. All flux linkages are through the infinite
permeance core channel.
2. Core with No Hysteresis Loop in the B–H Curve: This implies that the hysteresis loss is zero.
3. Core with Infinite Resistivity: This implies that the eddy current loss is zero.
4. Core which does not Saturate at any Magnetic Field Intensity: This implies that the transformer can
be connected to any voltage source of any frequency.
5. Winding with Zero Resistance: This implies that there are no winding losses and therefore the copper
losses are zero.
Chapter 07.indd 388 3/13/2009 7:08:45 PM

The transformer equivalent circuit is developed starting with the ideal transformer. The equivalent circuit of
the ideal transformer is shown in Figure 7.11(a). The variables ep and es are the terminal voltages across the
transformer primary and secondary windings, respectively, and ip and is are the terminal currents. Here,
N s e s ip
n= = =
N p e p is
The magnetizing current im = 0 and the entire primary mmfp is used to compensate the secondary mmfs due
to the secondary current flowing in the secondary winding and the external load resistance RL. In an ideal
transformer, the permeance is infinite. On introducing a finite permeance into the model, im is no longer
zero. A finite im is required to set up the mmf in the core as the B–H loop has a finite slope. Figure 7.11(b)
shows the equivalent circuit consisting of LM followed by an ideal transformer of turns ratio n.
Referring to Figure 7.11(b),
dim
Vi = e p = LM
dt
where
im = ip − n ⋅ is
ip is ip nis is
im
Vi ep es RL Vi ep LM es RL
Np : Ns Np : Ns
(a) (b)
ip nis is Lsp ip nis Lss is
ipo ipo
ic im ic im
Vi ep es RL Vi ep es RL
Rc LM Rc LM
Np : Ns Np : Ns
(c) (d)
Figure 7.11 (a) Ideal transformer; (b) finite permeance non-ideality included – im ≠ 0; (c) core loss
non-ideality introduced; (d) leakage flux linkage non-ideality introduced.
Chapter 07.indd 389 3/13/2009 7:08:45 PM

Next the core loss non-ideality is introduced. This means that the core material exhibits B–H hysteresis
loop characteristics giving rise to hysteresis loss. The resistivity of the core is also finite giving rise to eddy
current loss. The core loss is depicted as a dissipative resistive element in the electrical domain as Rc in
Figure 7.11(c). Here
ip − n ⋅ is = ipo = im + ic
The no-load primary current ipo consists of the magnetizing component im and a core loss component ic.
Next the leakage flux linkage non-ideality is introduced. Once the permeance is assumed finite, leakage flux
linkage path through the air gap will exist. This results in equivalent primary and secondary-side leakage
inductances as shown in Figure 7.11(d). Here,
dim dip
e p = LM ⋅ + Lσ p ⋅
dt dt
Vi = ep
dim di
e s = n ⋅ LM ⋅ − Lσ s ⋅ s
dt dt
e s = is ⋅ RL
Finally, the winding resistance non-idealities are included. Rp is the primary winding resistance and Rs
is the secondary winding resistance. The winding resistances give rise to the copper losses. The complete
equivalent circuit of the transformer is shown in Figure 7.12. Here,
dim dip
e p = LM ⋅ + Lσ p ⋅
dt dt
Vi = e p + ip ⋅ Rp
dim di
e s = n ⋅ LM ⋅ − Lσ s ⋅ s
dt dt
e s = is ⋅ RL + is ⋅ Rs
Rp Lsp ip nis Lss is Rs
ipo
Vi ep Rc es RL
LM
ic im
Np : Ns
Figure 7.12 Complete equivalent circuit of the transformer with winding

resistance non-idealities also introduced.
Chapter 07.indd 390 3/13/2009 7:08:45 PM

Area Product
A transformer consists physically of a primary winding and secondary windings on a common magnetic
core. As discussed in the case of the inductor area product approach, here also the two physical parameters
of the core, viz. the core cross-sectional area Ac and the window area Aw are related to the electrical
parameters.
The Faraday’s law of Eq. (7.8) relates the Ac to the voltage across the coil. The window area Aw is a
measure of the number of turns of a given wire cross-section that can be accommodated. The wire cross-
section is dependent on the current-carrying capacity or the rms current that flows through the wire. Thus
Aw is related to the current capacity. The product of Ac and Aw is called the area product Ap. For energy-
transferring devices like the transformers, the area product Ap is related to the power that will be trans-
ferred through the core.
Consider that the transformer has one primary winding of Np turns and m secondary windings consist-
ing of Ns1, Ns2, …, Nsm number of turns, respectively. The voltage across the primary winding is the source
voltage Vi. From the Faraday’s law,
dφ dB
ep = N p = N p Ac
dt dt
T /2
2 2
T ∫ e p ⋅ dt = N A ΔB
T p c
(7.50a)
0
where T is the period of the flux waveform and DB is the swing in the flux density during the half-period
T/2.
T /2
2
T ∫ e p ⋅ dt
0
is the half-period average of the voltage across the primary winding. Thus, Eq. (7.50a) can be written as
E p-av = 2 ⋅ N p ⋅ Ac ⋅ ΔB ⋅ f (7.50b)
where f = 1/T is the frequency of the flux waveform. From (7.50b), the number of turns in the primary
winding is given as
E p-av
Np = (7.51)
2 ⋅ Ac ⋅ ΔB ⋅ f
where Ep-av is the half-period average value of the voltage across the primary winding. Similarly, the number
of turns in the secondary windings is
E si -av
N si = (7.52)
2 ⋅ Ac ⋅ ΔB ⋅ f
where Nsi is the number of turns in the ith secondary winding; Esi-av is the half-period average value of the
induced voltage across the ith secondary winding.
The current flow through each winding causes copper losses, that is, power loss in the winding resis-
tance. The winding resistance is a function of the wire length and the cross-sectional area of the wire. For a
Chapter 07.indd 391 3/13/2009 7:08:45 PM

given number of turns, the winding resistance is a function of the cross-sectional area of the wire. If the
cross-sectional area of the wire is increased, then the winding resistance and hence the copper loss decreases.
The cross-sectional area of the wire is determined by defining the current density J for the wire. Thus,
I rms
a= (7.53)
J
where Irms is the rms value of the current flowing through the winding and J is the current density for the
winding in A/m2.
For the primary winding, the wire cross-sectional area is given as
I p-rms
ap = (7.54)
J
and for the secondary windings, the wire cross-sectional areas are
I si -rms
asi = (7.55)
J
where Isi-rms is the rms value of the current in the ith secondary winding and asi is the wire cross-sectional
area of the ith secondary winding. All the windings, that is the primary and the m secondary windings,
should be accommodated within the core window area Aw. However, as discussed in the last section on
inductors, the full window area Aw is not available as some area is needed to account for the wire enamel,
coil former, insulation paper used between windings, etc. Kw, which has a value less than unity, denotes this
loss of window area called the window utilization factor. Therefore,
m
K w Aw = N p ap + ∑ N si asi (7.56)
i =1
Substituting for the wire cross-sectional area from Eqs. (7.54) and (7.55) into Eq. (7.56), one obtains
m
K w Aw J = N p I p-rms + ∑ N si I si -rms (7.57)
i =1
Substituting for the primary and the secondary turns from Eqs. (7.51) and (7.52) into Eq. (7.57), one
obtains
E p-av m E si -av
K w Aw J = I p-rms + ∑ I si -rms (7.58)
2 ⋅ Ac ⋅ ΔB ⋅ f i =1 2 ⋅ Ac ⋅ ΔB ⋅ f
1 ⎛ m ⎞
Ap = Aw Ac = ⎜⎜ E p-av I p-rms + ∑ E si -av I si -rms ⎟⎟ (7.59)
2 ⋅ K w ⋅ ΔB ⋅ f ⋅ J ⎝ i =1 ⎠
If the form factor Kf is defined as the ratio of the rms value to the average value, then Eq. (7.59) can be
written as
1 ⎛ m ⎞
Ap = Aw Ac = ⎜⎜ E p-rms I p-rms + ∑ E si -rms I si -rms ⎟⎟ (7.60)
2 ⋅ K w ⋅ K f ⋅ ΔB ⋅ f ⋅ J ⎝ i =1 ⎠
Chapter 07.indd 392 3/13/2009 7:08:45 PM

where Ap is the area product in m4. Equations (7.59) and (7.60) are the generic area product equations for
transformers of any given source voltage waveshape. It should be remembered that the voltages are half-cycle
average and rms values in Eqs. (7.59) and (7.60), respectively.
EXAMPLE 7.1 For a transformer that is supplied from a sinusoidal voltage source, the flux, voltage
and current waveforms are sinusoidal in nature.
Kf = 1.11
The flux density swing in a half-period is from −Bm to +Bm where Bm is the maxi-
mum operating flux density and chosen to be less than Bsat for the core material.
Therefore,
DB = 2Bm
The sum of all the secondary VA is the consolidated load power Po. Thus,
m
Po = ∑ E si -rms I si -rms
i =1
Po
= E p-rms I p-rms
η
where h is the efficiency of the transformer. Making the above substitutions into the
generic area product Eq. (7.60), one obtains
Po ⎛ 1⎞
Ap = ⎜1 + ⎟ (7.61)
4.44 ⋅ K w ⋅ Bm ⋅ f ⋅ J ⎝ η ⎠
where Po is the summation of all secondary VA; Bm the allowed maximum flux
density in tesla; J the current density in A/m2 which is commonly chosen as 3 × 106
A/m2; Kw the window utilization factor which is chosen between 0.3 and 0.4 for
practical transformers; f the frequency of operation in Hz.
Area Products for DC–DC Converters

Using the generic area product relationship given in Eq. (7.59), the area products of various DC–DC con-
verters are worked out and tabulated in Table 7.3.
Table 7.3 Area products of common DC–DC converters

Configuration Area Product
Po [1 + (1 / η)]
Forward converter Ac Aw =
2 K w JBm f s
Po [ 2 + (1 / η)]
Half-bridge converter Ac Aw =
4 K w JBm f s
(Continued )
Chapter 07.indd 393 3/13/2009 7:08:46 PM

Table 7.3 Continued

Configuration Area Product
Po [ 2 + (1 / η)]
Full-bridge converter Ac Aw =
4 K w JBm f s
2 Po [1 + (1 / η)]
Push–Pull Converter Ac Aw =
4 K w JBm f s
Transformer Design
The previous section discussed the area product and the method to obtain the area product of the trans-
former for any given topology. This sub-section discusses the steps to design a transformer. The design steps
are same whether the transformer is a low-frequency transformer or a high-frequency transformer.
Step 1 (Load Power Estimation): Before designing the transformer, it is necessary to estimate the total
load power Po that the transformer has to handle. This is calculated for a given topology as
m ⎛ T ⎞
1
Po = ∑ ⎜ ∫ e si ⋅ isi ⋅ dt ⎟ (7.62)
⎜
i =1 ⎝ T 0
⎟
⎠
Equation (7.62) gives the summation of the load powers of all the m secondary windings.
Step 2 (Area Product Calculation): The area product for the transformer in a particular circuit configu-
ration is estimated based on the generic area product given by Eq. (7.59) or Eq. (7.60). The area products
for commonly used topologies can be worked out based on Eq. (7.59) and tabulated in a manner similar
to that shown in Table 7.3 for DC–DC converters. Equation (7.61) gives the area product for sine wave
transformers that are used in low-frequency grid applications. From the estimated value of the area
product Ap, select a core from Appendix II that has an area product greater than that calculated.
Step 3 (Number of Turns): The number of turns is determined for the primary and the secondary
windings as
E p-av
Np = (7.63)
2 ⋅ Ac ⋅ ΔB ⋅ f
(where Ep-av is the half-period average value of the voltage across the primary winding)
E si -av
N si = (7.64)
2 ⋅ Ac ⋅ ΔB ⋅ f
(where Nsi is the number of turns in the ith secondary winding and Esi-av is the half-period average value
of the induced voltage across the ith secondary winding.) The turns ratio for the ith secondary winding
with respect to the primary winding is given as
N si
ni = (7.65)
Np
Chapter 07.indd 394 3/13/2009 7:08:46 PM

Step 4 (Wire Gauge Selection): The wire gauges are selected based on the rms currents flowing in a
given winding. The rms currents are first calculated based on the circuit topology on the primary and
secondary sides. For the primary winding, the wire cross-sectional area is given as
I p-rms
ap = (7.66)
J
and for the secondary windings, the wire cross-sectional areas are
I si -rms
asi = (7.67)
J
where Isi-rms is the rms value of the current in the ith secondary winding and asi is the wire cross-
sectional area of the ith secondary winding. From the values of the required estimate of the wire
cross-sectional areas, one can select the wire gauge from Appendix III wherein the gauge selected is
greater than that calculated.
Step 5 (Window Area Check): Wire gauges are available in discrete values as per the list given in
Appendix III. One has to invariably select a wire gauge whose area is greater than the calculated one.
This may necessitate higher window area requirement. When several windings are used (e.g., push–pull
with primary and secondary center-tapped, multi-output SMPS, etc.), the window utilization factor Kw
may decrease further as the various windings need more insulation layers. Using the actual selected wire
cross-sectional areas, check to see if the turns fit into the effective window area KwAw of the core by
using the inequality:
m
K w Aw > N p ap + ∑ N si asi (7.68)
i =1
If the above inequality is not satisfied, then select a core with the next higher Ap and re-do the calcula-
tions till the inequality is satisfied.
Problem 7.2
Design a transformer for the forward converter configuration with the following specifications:
Output voltage Vo = 12 V
Output current Io = 3 A
Switching frequency f = 20 kHz
Supply voltage Vi = 24 V ± 10%
Refer to the discussion on forward converter in Chapter 5 to understand the operation of the converter.
Solution
Step 1 (Load Power Estimation): The secondary of the forward converter transformer has to supply
the output load, the power dissipated in the secondary diodes and the power dissipated in the inductor
winding and the inductor core losses. If the demagnetizing winding is used, then the power loss in the
demagnetizing winding diode should also be accounted. However, as the current through the diode of
the demagnetizing winding is the magnetizing current which is very small compared to the load cur-
rent, the power loss in the demagnetizing winding can be neglected. Thus,
Chapter 07.indd 395 3/13/2009 7:08:46 PM

Po = Vo I o + VD I o + PL-cu + PL-core (7.69)
where PL-cu is the copper loss of the secondary-side inductor winding and PL-core is the core loss of the
inductor core. A starting conservative value of the inductor core and copper loss that is normally consi-
dered for the transformer design is PL-cu + PL-core = 10% of Vo I o .
The converter diodes in the secondary of the transformer will show a significant drop as they are
carrying high currents. The diode drops may be as high as 1.5 V for fast recovery diodes. It is safe to
design for the worst case of VD = 1.5 V. Equation (7.69) now becomes
Po = 1.1 × Vo I o + VD I o = 1.1 × 12 × 3 + 1.5 × 3 = 44.1 W
Step 2 (Area Product Calculation): Referring to Table 7.3, the area product for the forward converter
configuration is given by
Po [1 + (1 / η)]
Ap = (7.70)
2 K w JBm f s
At high frequencies, usually the core material choice is ferrite. It has a saturation flux density Bs of 0.3
T. The maximum operating flux density in the core is chosen as 0.2 T (Bm = 0.2 T). Another important
design parameter is the current density J. If it is chosen very low to bring down the copper loss, then for
a given current, a very large conductor cross-section is required (thereby demanding a large window
area). It should also be noted that a choice of very low current density will not have a significant reduc-
tion in the copper loss because at high frequencies the skin effect prevents decrease in the wire resis-
tance. A current density between 2 and 5 A/mm2 is found to be a good compromise between conductor
resistance and window area. A value of J = 3 × 106 A/m2 is a default choice in most cases. The window
utilization factor K = 0.4 and a conservative efficiency for the transformer is taken to be 0.8. Substituting
the values in Eq. (7.70), one obtains
Ap = 1.46172 × 10–8 m4 = 14617.2 mm4
Choose a suitable core from Appendix II that has an Ap greater than the value calculated above.
P 36/22 is selected (Ac = 201 mm2, Aw = 101 mm2, Ap = 20100 mm4, lm = 53.2 mm)
Step 3 (Number of Turns): The primary number of turns can be calculated using
E p-av
Np =
2 ⋅ Ac ⋅ ΔB ⋅ f
For the forward converter, the half-period average value across the primary winding is Vimax, and DB is
Bm. Thus
Vimax 26.4
Np = = = 16.4
2 ⋅ Ac ⋅ Bm ⋅ f 2 × 201 × 10−6 × 0.2 × 20000
Rounding to the next higher integer, Np = 17 turns.

For the secondary, the secondary voltage and the output voltage are related by the forward converter
input–output relationship as discussed in Chapter 5. Including even the non-idealities like the diode
drop and the inductor winding resistance drop, one obtains
Chapter 07.indd 396 3/13/2009 7:08:46 PM

e s ⋅ Dmin = Vo + VD + 0.1Vo
Vimin = Vi – 10% Vi = 21.6 V

Vimax = Vi – 10% Vi = 26.4 V
Dmax = 0.5
Dmin = (Vimin × Dmax)/Vimax = (21.6 × 0.5)/26.4 = 0.41
Vo + VD + 0.1Vo 12 + 1.5 + 1.2

es = = = 35.85 V
Dmin 0.41
es 35.85
Ns = = = 22.3
2 ⋅ Ac ⋅ Bm ⋅ f 2 × 201 × 10−6 × 0.2 × 20000
Rounding to the next higher integer, Ns = 23 turns. The turns ratio is given as
N s 23
n= = = 1.35
N p 17
The demagnetizing winding Nd is equal to N1 as they are wound bifilar.
Step 4 (Wire Gauge Selection): The rms values of the currents flowing in the windings are:
I s-rms = I o Dmax = 3 × 0.5 = 2.12 A
I p-rms = n ⋅ I s-rms = 1.35 × 2.12 = 2.86 A

The current through the demagnetizing winding is given as
1 − Dmax
I d-rms = I m ⋅
3
where
Vimin ⋅ Dmax
Im =
LM ⋅ f
where
μo μr Ac N p2
LM =
lm
A representative value from the datasheet for CEL HP3C grade ferrite is given as μ r = 2000 ± 25% .
The worst case Im is obtained at minimum value of relative permeability. Therefore, taking mr = 1500
and substituting in the LM equation, one obtains
4π × 10−7 × 1500 × 201 × 10−6 × 17 2

LM = = 2.06 mH
53.2 × 10−3
Chapter 07.indd 397 3/13/2009 7:08:46 PM

The magnetizing current is given as

21.6 × 0.5
Im = = 0.26 A
2.06 × 10−3 × 20000
1 − 0.5
I d-rms = 0.26 × = 0.106 A
3
From the rms values of the currents and for a current density J = 3 × 106 A/m2, the required wire cross-
sectional areas are
I p-rms I s-rms I
ap = , as = , ad = d-rms
J J J
Substituting the values, one obtains
ap = 0.953 mm2
Choose a wire gauge from Appendix III whose cross-section is greater than that calculated above: SWG
18 is a proper choice (ap = 1.167 mm2).
as = 0.706 mm2
19 is a proper choice (as = 0.8107 mm2).
ad = 0.035 mm2
35 is a proper choice (ad = 0.03575 mm2).
Step 5 (Window Area Check): Now check for the inequality
m
K w Aw > N p ap + ∑ Nsi asi
i =1
Using the actual values of the cross-section areas, we have

m
N p ap + ∑ Nsi asi = 17 × 1.167 + 23 × 0.8107 + 17 × 0.03575 = 39.1 mm 2
i =1
AwKw = 40.4 mm2

Thus the above inequality is satisfied, which means that the windings will fit in the available window
area.
7.4 Current Transformer
C Ts are mainly used to sense the currents in the power circuits. The sensed currents are used as feedback
signals for control, protection, monitoring and display. The CT is an energy-transferring device like the
PT but the input energy source is a current source instead of a voltage source as is the case for the PT. How-
ever, all the magnetic principles that were discussed for the PT are applicable for the CT too.
Chapter 07.indd 398 3/13/2009 7:08:46 PM

ip nis is
ipo
ip ep Rc LM es R
ic im
Np : Ns
Figure 7.13 Equivalent circuit of the CT.
Though the magnetic principles are same as that of the PT, the design constraints are different from that
of the PT. This is due to the input energy source being a current source and the output being a voltage that is
supposed to be proportional to the input current. Therefore the design methodology for the CTs is entirely
different. CT is actually a dual of the PT. If the secondary load resistor RL is made zero, then es is zero and
therefore there is no flux flow rate within the core. If the secondary load resistor RL were made infinity, that
is, the secondary is open circuited, then there is no secondary current and as a result there is no secondary
mmfs (= Nsis). The only mmf in the core is the primary-side mmfp given by Npip. As there is no opposing mmf,
the core may saturate.
The equivalent circuit for the CT is shown in Figure 7.13. It can be observed that the equivalent circuit
is similar to that of the PT; however, the input energy source is a current source ip. As the input energy
source is a current source the series winding resistance and the leakage inductance terms are removed from
the equivalent circuit as they will not affect the input current. Though these non-idealities are actually pres-
ent, the design of CT will not consider these power losses and voltage drops as the main focus is on main-
taining the current waveshape fidelity and not on efficiency.
In a CT, the primary number of turns is decided first. It is chosen as 1 or at most 2 in most cases irre-
spective of the size of the core. The secondary number of turns is decided by the circuit requirements and
the maximum primary current value. Ns is much greater than Np as the primary-side current is generally
required to be stepped down.
If ip is the primary current and is the required secondary current, then
Ns
n=
Np
and referring to the equivalent circuit as shown in Figure 7.13, one obtains
ip − ipo = n ⋅ is (7.71)
ip − ipo
is = (7.72)
n
If the secondary is terminated with a resistance R called the burden resistor, then
e s = is ⋅ R (7.73)
The output voltage es across the sense resistor should be a measure of the source current ip. However, from
Eqs. (7.72) and (7.73) it can be observed that the output voltage would be an accurate measure of ip if ipo
Chapter 07.indd 399 3/13/2009 7:08:46 PM

were equal to zero. Therefore in CTs it is imperative to ensure that ipo is very small compared to ip. The cur-
rent ipo consists of the magnetizing component im and the core loss component ic. Reducing ipo implies that
both the magnetizing component im and the core loss component ic should be made as small as possible. If
the rate of the flux or the magnetizing current is made small, then the absolute peak value of im will also be
reduced. Therefore, from Faraday’s law,
dim
e p = LM (7.74)
dt
Rearranging the above equation,
dim e p e
= = s (7.75)
dt LM n ⋅ LM
In the case of switched-mode DC–DC converters, the currents are pulsed and so the secondary voltage
developed by the pulsed secondary current is also pulsed. Therefore for the duration of the pulse, if the sec-
ondary voltage is assumed to be constant, then im is a linearly rising waveform. The maximum value of im
occurs at the end of the current pulse width. If the current pulse width is tw then integrating Eq. (7.75), one
obtains
es ⋅ t w
im-max = (7.76)
n ⋅ LM
From Eqs. (7.75) and (7.76), it can be observed that the magnetizing current can be reduced by
1. decreasing the voltage across the secondary winding es;
2. increasing turns ratio n;
3. increasing magnetizing inductance LM.
Ideally if es is zero then the magnetizing current im-max is zero. However, if the CT secondary were short cir-
cuited then the secondary voltage would always be zero and as a result the primary objective of having a
measure of the primary current for sensing purposes is not established. Therefore, es has to be finite but low
(about 0.1–0.5 V). Also,
dφ N s ⋅ Ac ⋅ ΔB
es = N s = (7.77)
dt Δt
Lower value of es implies that the flux density swing DB is also lower for a given core cross-sectional area and
the number of turns in the secondary winding. This has the double effect of reducing both the magnetizing
current im due to lower magnetic field intensity H requirement and also the core loss component as the B–H
loop area is reduced. Thus for CT applications, Bm is chosen as around 10% of Bsat for the material. For a sil-
icon steel CRGO core having a saturation flux density of 1.2 T, Bm is chosen as 0.12 T to keep the magnetiz-
ing component low.
Equations (7.75) and (7.76) also suggest that the magnetizing current can be reduced by increasing the
turns ratio n and the magnetizing inductance LM. The turns ratio is increased to the maximum by keeping
Np = 1. This will result in n = Ns. The magnetizing inductance is given as
μo μr Ac N p2
LM = Λ ⋅ N p2 = (7.78)
lm
Chapter 07.indd 400 3/13/2009 7:08:46 PM

From Eq. (7.78), it can be observed that the magnetizing inductance can be increased by increasing the per-
meability of the core or the permeance of the core. Alternately, an oversized core with a larger core cross-
sectional area can be used to increase the magnetizing inductance. Thus, one can conclude that for a good
CT, the magnetizing current should be low and for this one needs
1. larger core cross-sectional area;
2. higher permeability core or core with larger permeance;
3. lower secondary voltage or lower burden resistance R;
4. large turns ratio n.
Figure 7.14 shows two schematics for processing the CT outputs. In Figure 7.14(a), the voltage across the
burden resistor R is given as an input to a non-inverting operational amplifier (op-amp). In the case of
Figure 7.14(b), the CT secondary dot polarity is reversed and connected to the op-amp as shown. The virtual
ground principle of the op-amp is used in this circuit to make es = 0. The secondary current flows through the
burden R that is connected across the output of the op-amp and the “−” terminal. In the case of CTs where
the primary current is unidirectional, at the end of the duty cycle when the primary current becomes zero, the
stored energy due to the magnetizing current will cause a large negative voltage spike at the secondary. To pre-
vent such voltages reaching the load, a clamper circuit or a half-wave rectifier is used after the burden. In
order to minimize the effects of the half-wave rectifier drop, a precision half-wave rectifier may be used.
Design of Current Transformers

The CTs are used mainly to measure or sense the current and are not used to transfer power. Therefore, in
the case of CT design the issue is not one of power transfer but one of current waveshape fidelity. Therefore
the design steps are different from those followed for the PTs.
R1 R2
ip
−
es R
+ Vo
(a)
is
R
is
ip
−
es = 0
+ Vo
(b)
Figure 7.14 (a) CT output amplified by an op-amp non-inverting amplifier; (b) using virtual
ground of op-amp to process the CT output.
Chapter 07.indd 401 3/13/2009 7:08:47 PM

Step 1 (Determine Turns Ratio): To start with choose Np = 1 or at most 2. Based on the specification
requirement decide on the required maximum secondary current, is-max when the primary current is at
the maximum value. Then,
ip-max
n=
is-max
The secondary number of turns is given as
Ns = n ⋅ Np
Step 2 (CT Burden): The secondary voltage es is a design specification. It should ideally be zero but
should be chosen as a finite value. The value of es should be chosen as low as possible to obtain the best
fidelity. The type of amplification and post-processing circuitry will determine the maximum value that
can be chosen for es. As a thumb rule, choose es < 0.1 for very good fidelity and es < 1 for reasonable
quality CTs. The output of the CT should be amplified by a non-inverting op-amp stage. The value of
the burden resistor R is given as
e s-max
R=
is-max
Step 3 (Core Selection): In the case of CTs, the area product approach is not used as the objective here
is not power transfer. Instead Eq. (7.77) is used which will estimate the core cross-sectional area based
on the constraints of Steps (1) and (2). From the estimated core cross-sectional area an appropriate
choice of core is made from Appendix II. Thus,
e ⋅ Δt
Ac = s
N s ⋅ ΔB
In the case of CTs where the input current ip is bi-directional, the flux density swing in the core is
from −Bm to +Bm in a maximum time of half the period. Therefore for CTs with bi-directional pri-
mary current,
e s ⋅ (T / 2) es
Ac = =
2 ⋅ N s ⋅ Bm 4 ⋅ N s ⋅ Bm ⋅ f
In the case of CTs where the input current is uni-directional, the flux density swing in the core is
from 0 to +Bm in a maximum time of half the period. Therefore for CTs with uni-directional pri-
mary current,
e ⋅ (T / 2) es
Ac = s =
N s ⋅ Bm 2 ⋅ N s ⋅ Bm ⋅ f
Using the estimated value of the core cross-sectional area, select a core from Appendix II that has an Ac
value greater than that calculated above.
Step 4 (Wire Gauge Calculation): The wire gauges are selected based on the rms currents flowing in a
given winding. The rms currents are first calculated based on the current wave shape pattern. For the
primary winding, the wire cross-sectional area is given as
I p-rms
ap =
J
Chapter 07.indd 402 3/13/2009 7:08:47 PM

and for the secondary winding, the wire cross-sectional area is

I
as = s-rms
J
From the values of the required estimate of the wire cross-sectional areas, one can select the wire gauge
from Appendix III wherein the gauge selected is greater than that calculated.
Step 5 (Window Area Check): In a manner similar to that discussed for the inductor and the PT
design, one has to cross check whether the windings will fit into the available window area. Using the
actual selected wire cross-sectional areas, check if the turns fit into the effective window area KwAw of
the core by using the inequality
K w Aw > N p ap + N s as
If the above inequality is not satisfied, then select a core with the next higher Ac and re-do the calcula-
tions till the inequality is satisfied.
7.5 General Notes on Magnetics

Window Utilization Factor
The window utilization factor Kw, has been used in both the inductor and transformer design to account for
the non-availability of certain portion of the window area Aw. The effective window area that becomes avail-
able is KwAw as discussed in the previous sections. The factors that affect the available window area are dis-
cussed in the following sub-sections.
Coil Former/Bobbin
These windings are generally wound over a bobbin or coil former. The bobbin that houses the windings has
a certain thickness and this reduces the available window area. The unavailability of the window area due to
the bobbin or coil former is accounted for by a factor Kw1.
Space Factor
There is an increase in the wire cross-sectional area due to a coating of insulating enamel on the bare copper
wire. This results in a reduction in the winding space by a factor called the space factor Kw2 defined as
Conductor area
K w2 =
Conductor area + Insulation area
This factor depends on the wire gauge. For example, a SWG 45 wire gauge having a nominal diameter, dnom,
of 0.071 mm, and a maximum diameter with enamel insulation, dmax, of 0.086 mm.
(π /4)d nom
2
0.0712
K w2 = = = 0.68
(π /4)d max
2
0.0862
If the wire gauge is thicker, like say SWG 14, the value of Kw2 comes out to be 0.91. Thinner the gauge,
lesser is the space factor.
Air Gaps between Conductors

Another factor which reduces the available window area is that the adjacent turns are not air tight. There are
always gaps between the conductors. This is due to the wires having a circular cross-section. In practice, this
factor Kw3 is found to be in the order of 0.6–0.9.
Chapter 07.indd 403 3/13/2009 7:08:47 PM

Insulation Factor
There is one more factor called the insulation factor, Kw4. Generally, when one winds a transformer, there
are several insulation layers that are included, such as a layer between the primary winding and the second-
ary winding to meet the breakdown voltage requirements. If there are multiple secondary windings, addi-
tional layers have to be used further reducing the available window area.
Considering the above factors, the effective available window area is
Available area = Kw × Aw
where
Kw = Kw1 × Kw2 × Kw3 × Kw4
Typically, a value of Kw = 0.3 to 0.4 can be assumed for transformer design purposes.
Core Shapes
There are numerous core shapes and sizes that are available commercially from many manufacturers. A few
common core shapes will be illustrated here to give a flavor of the available commercial shapes.
Rod Core
This is normally made of ferrite and used in radios especially for tuning an inductor. The coil is wound on a
cylindrical coil former which is inserted over the ferrite rod to form a solenoid as shown in Figure 7.15. The
rod sits in the middle of the coil and adjustments of the rod position can be used to tune the inductance.
The field spreads out into air at the ends of the rod. The path through the air ensures that the inductor does
not saturate and remains linear. In this type of inductor radiation occurs at the end of the rod and electro-
magnetic interference (EMI) may occur.
“C” or “U” Core

Figure 7.16(a) shows a “U”-shaped core with sharp edges and Figure 7.16(b) shows a “C”-shaped core with
rounded edges. Two UU or two CC cores can be coupled together to form a closed magnetic circuit.
A U and I core or a C and I core can also be coupled to form a closed magnetic circuit.
“E” Core
The E-shaped cores are very popular. Some of the E core types are shown in Figure 7.17. Two EE cores of
the same type can be coupled together to form a closed symmetrical magnetic circuit. Normally, the wind-
ing is wound around the center leg whose cross-sectional area is twice that of the individual outer legs.
Figure 7.17(a) shows the classical E core that is most popular and commonly used. Figure 7.17(b) illustrates
Fernite rod
Coil
Figure 7.15 Rod core.
Chapter 07.indd 404 3/13/2009 7:08:47 PM

(a) (b)
Figure 7.16 (a) U-shaped core; (b) C-shaped core.
(a) (b) (c) (d)
Figure 7.17 (a) Classical E core; (b) EFD core; (c) ER core; (d) EP core.
the EFD core that is used for flat or low-profile inductors or transformers. Figure 7.17(c) shows the ER core
that has a cylindrical central leg. Figure 7.17(d) shows and EP core that is in between an ER core and a pot
core.
Two “E” cores will form a three-legged structure when coupled together as shown in Figure 7.18. If an
air gap is required, the center leg of the “E” is shortened so that the air gap sits in the middle of the coil to
minimize fringing and reduce EMI.
“E” and “I” Laminations

Sheets of silicon steel are stamped out in shapes like “E” and “I”. These E and I stampings are stacked to
form three-legged laminated core structures. The coils are wound on these three legs to form single-phase
low-frequency transformers or low-frequency inductors or three-phase low-frequency transformers.
Figure 7.18 Coupling two E cores.
Chapter 07.indd 405 3/13/2009 7:08:47 PM

Pot Core
This is used for inductors and transformers. The shape of a pot core is round with an internal hollow that
almost completely encloses the coil former and coil. The pot core is made in two halves which fit together
around the coil former (bobbin). This shape of core has a shielding effect, preventing radiation and reducing
EMI. Figure 7.19 shows one half of a pot core.
Toroidal Core
The toroidal core is shown in Figure 7.20. The coil is wound through the hole in the toroid and around the
outside. The coil is distributed uniformly all around the circumference of the toroid. The circular core geom-
etry will naturally keep the field constrained within the core material. It is a low-radiation transformer and
popular in amplifiers where the desirable features are high power, small volume and minimal EMI. It is also
used in CTs for current-sensing applications. However, it is not easy to wind a coil on a toroidal core.
Planar Core
A planar core consists of two flat pieces of magnetic material as shown in Figure 7.21. One-half is placed
above and another half is placed below a winding that is printed on a printed circuit board. It is typically
used for PCB mountable applications. This design is excellent for small volume, low-profile and very high
frequency transformers.
Polarity and Dot Convention

Knowledge of polarities and dot convention is very important when working with transformers. While
making transformer connections, if the winding polarities are not carefully considered, then the results can
be quite disastrous. Therefore the dot convention to mark the winding polarity must be strictly practiced
both in symbolic circuit diagrams and also while making interconnections.
Figure 7.19 Pot core. Figure 7.20 Torodial core.
Figure 7.21 Planar core and its mounting on a PCB.
Chapter 07.indd 406 3/14/2009 12:54:07 PM

A C
(1)
VAB
B D
(a) (1) VCDa
A C
(2) VCDb
(2)
B D
(b) (c)
Figure 7.22 (a) Transformer with A and C having same polarity; (b) transformer with A and D
having same polarity; (c) the waveforms for configurations shown in (a) and (b).
If two coils AB and CD are wound over a core in the same direction (clockwise or anti-clockwise), then
points A and C will have the same phase with respect to the other end of the winding. In other words, if at
any instant, point A is positive with respect to point B, then point C is also positive with respect to point D
and vice-versa. Points A and C are said to have the same polarity. Symbolically, the same polarity ends of the
windings of a core are denoted by a “dot” mark at the beginning of the winding as shown in Figure 7.22(a).
Figures 7.22(a) and 7.22(b) show two transformers with different dot polarities. Figure 7.22(c) shows the
winding waveforms for the two transformers.
In a transformer, when polarities are not known, it is possible to find out the polarities with a simple
test. The transformer can be connected as shown in Figure 7.23(a). Connect any two ends, say BD, together
and measure the input voltage V1 and the voltage between A and C, that is, V2. If the measured V2 is less
than VAB (V2 = VAB − VCD), then mark polarities as shown in Figure 7.23(b). If the measured V2 is greater
than VAB (V2 = VAB + VCD), then mark the polarities as in Figure 7.23(c).
Magnetic Losses
Hysteresis Loss
Referring to Figure 7.24, as one travels along the magnetization path of a virgin core specimen from O to B,
the flux density swings from 0 to Bsat. If the magnetic field intensity H is again brought back to O, there
exists some finite flux density (point C) or magnetic charge in the core due to the magnetic retentivity prop-
erty of the core material. A negative magnetization field has to be applied to bring the specimen back to zero
flux density (point D).
From Ampere’s law, as given by Eq. (7.2), the magnetizing field H is
Ni mmf
H= =
lm lm
and the flux density is given by
B = μH
Chapter 07.indd 407 3/13/2009 7:08:47 PM

V2
A C
A C
V1
B D
B D
(a) (b)
A C
B D
(c)
Figure 7.23 Testing for dot polarity.
Bsat
B
D O A H
E
Bsat
Figure 7.24 B–H loop for a core material showing the hysteresis loop.
The induced voltage across a winding that is wound on the core is

dφ dB
e=N = NAc
dt dt
Chapter 07.indd 408 3/13/2009 7:08:47 PM

From the above, the energy required to traverse the B–H loop in the path EFAB is
T /2
dφ
E up = ∫ mmf ⋅
dt
⋅ dt
0
Bm Bm
(7.71)
= ∫ H ⋅ l m ⋅ Ac ⋅ dB = Vc ∫ H ⋅ dB
− Bm − Bm
where Vc = l m ⋅ Ac is the core volume. Figure 7.25(a) illustrates the area of the B–H loop that represents the
Bm
energy Eup where ∫− B m
H ⋅ dB is the area under the B–H loop with reference to the B-axis.
Similarly, the energy required to traverse the down-swing of the B–H loop is given as
T
dφ
E down = ∫ mmf ⋅
dt
⋅ dt
T /2
− Bm Bm Bm
(7.72)
= ∫ H ⋅ l m ⋅ Ac ⋅ dB = − ∫ H ⋅ l m ⋅ Ac ⋅ dB = −Vc ∫ H ⋅ dB
Bm − Bm − Bm
Figure 7.25(b) illustrates the energy per unit volume that represents Edown. The total energy for one complete
traversal of the B–H loop is given as
⎛ ⎞
⎜ Bm Bm ⎟
⎜ ⎟
E hys = E up + E down = l m ⋅ Ac ⎜ ∫ H ⋅ dB − ∫ H ⋅ dB ⎟ (7.73)
⎜ −
Bm

− Bm
⎟⎟
⎜ Path
⎝ EFAB Path BCDE ⎠
Equation (7.73) gives the amount of energy required to traverse the B–H loop. It can be observed from
Figure 7.25 that the energy required to traverse the B–H loop is equal to the product of the area of the B–H
B B B
B
C B
A
F H D H H
Eup Edown
E Ehys
E
(a) (b) (c)
Figure 7.25 (a) Eup Per unit volume; (b) Edown per unit volume; (c) Ehys per unit volume.
Chapter 07.indd 409 3/13/2009 7:08:48 PM

loop and the core volume l m ⋅ Ac . This energy is lost forever and is called the hysteresis loss. The power lost
is the rate of energy required to traverse the B–H loop and is given as
⎛ ⎞
⎜ Bm Bm ⎟
⎜ ⎟
Phys = l m ⋅ Ac ⎜ ∫ H ⋅ dB − ∫ H ⋅ dB ⎟ f (7.74)
⎜ −
Bm

− Bm
⎟⎟
⎜ Up-swing
⎝ path Down-swing path ⎠
It can be observed from Eq. (7.74) that the hysteresis power loss increases with the frequency of operation.
The power loss also depends on the flux density and magnetization. However, it varies from core material to
material as it is a function of the B–H loop path. It can be observed that if the up-swing path and the down-
swing path coincide, then the hysteresis loss is zero as expected.
When the applied magnetic field H is removed (i.e., when the primary excitation is removed), the core
flux is non-zero. The flux density can be anywhere either on OC part of axis or on OF part of axis, that is,
positive or negative remnant flux density can exist. Such an uncertain residual flux density can create surge
currents during switching ON of the transformer; therefore, one must accordingly account for this issue in
the circuit design by incorporating soft-start circuits.
Eddy Current Loss

Whenever the flux in the core changes, that is df/dt is non-zero, a voltage is induced in the windings of the
core according to Faraday’s law. Observe that the core itself acts as a one turn winding (a closed turn) in
addition to the other windings wound on the core.
Therefore a voltage is induced in the core which will cause a current to flow in the core material depend-
ing on the resistance offered by the core material. The eddy current ie is a sheet of current that is distributed
throughout the magnetic path length. The voltage induced for the single turn core material is
dφ dφ
e eddy = N = (as N = 1)
dt dt
The resistance offered by the core material for the path of the eddy current is
ρl e
Reddy = (7.75)
Ae
where ρ is the resistivity of the core material, l e is the mean length of the eddy current path and Ae is the
mean cross-sectional area perpendicular to the flow of the eddy current. It should be noted that Ae is not the
same as the core cross-sectional area Ac. The eddy current is given as
e eddy
ie = (7.76)
Reddy
The power dissipated in the resistance of the core material is lost forever and is called the eddy current loss.
This loss is given as
2
e eddy (dφ / dt )2 ⋅ Ae 4 ⋅ f 2 ⋅ Ac2 ⋅ ΔB 2 ⋅ Ae
Peddy = ie2 Reddy = = = (7.77)
Reddy ρ ⋅ le ρ ⋅ le
Chapter 07.indd 410 3/13/2009 7:08:48 PM

df ie
le dt
ie
le
df
Ac dt
(a) (b)
Figure 7.26 Eddy current (a) in the core; (b) in a laminated core.
DB is the flux density swing in an interval of half a period. For a core where the flux swing is bi-directional
like in power transformers, half- and full-bridge transformers, DB = 2Bm. For cores where the flux swing is
uni-directional like in inductors and forward converter or flyback topologies, DB = Bm. Thus
8 ⋅ f 2 ⋅ Ac2 ⋅ Bm2 ⋅ Ae
Peddy = ( for bi-directional flux density swing) (7.78)
ρ ⋅ le
4 ⋅ f 2 ⋅ Ac2 ⋅ Bm2 ⋅ Ae
Peddy = ( for unidirectional flux density swing) (7.79)
ρ ⋅ le
It is difficult to estimate le and Ae for various core geometries. However, the core manufacturers provide
nomo-graphs of the eddy current loss as a function of frequency for specific core geometries.
With some core materials like the silicon steel CRGO and CRNGO cores the material resistivity is very
less. To reduce the resistance offered to the eddy current, the core cross-section is split into many small cross-
sections called laminations as shown in Figure 7.26(b). This way df/dt also splits up by the number of
laminations used. Ae is also reduced to increase Reddy. If n laminations are used then
dφl dφ / dt
=
dt m
Thus,
⎛ 4 ⋅ f 2 ⋅ A 2 ⋅ ΔB 2 ⋅ A ⎞ 4 ⋅ f 2 ⋅ A 2 ⋅ ΔB 2 ⋅ A
Peddy = m ⎜ c el
⎟= c el
(7.80)
⎜ m ⋅ ρ ⋅ l el
2 ⎟ m ⋅ ρ ⋅ l
⎝ ⎠ el
where Ael is the mean cross-sectional area in a lamination perpendicular to the flow of the eddy current, lel is
the mean path of the eddy current in a lamination and m is the number of laminations.
As can be observed, the effective resistance of the core increases and the eddy currents reduce, thus,
reducing the losses and temperature of the core. A second way of increasing the resistance of the solid core
is to have material with large resistivity, such as ferrites and amorphous glass. They are a solid mass and
laminations are not required due to their high inherent resistivity.
Chapter 07.indd 411 3/13/2009 7:08:48 PM

The hysteresis and eddy current losses together are called the core losses:
Pc = Phys + Peddy (7.81)
Skin Effect
The current density across a conductor cross-section is not uniform. It varies from the conductor surface to
the center of the conductor in an exponential manner. Figure 7.27 shows an enlarged view of a conductor
d
−
V
+
I
I E
E
d
+ 0 d x
2
(a) (b)
E/e
0 d x
2
sd
(c)
Figure 7.27 (a) E-field applied to a conductor; (b) DC E-field profile across the conductor width;
(c) AC E-field profile across the conductor width.
Chapter 07.indd 412 3/13/2009 7:08:48 PM

carrying a current I. The conductor has a diameter d. If the current flowing through the conductor is DC,
then the field E that acts as the motive force for the charges within the conductor is also DC. Figure 7.27(b)
shows the distribution of the E-field across the width of the conductor. For a DC-field the E-field distribu-
tion across the width of the conductor is uniform as shown. This results in a uniform current distribution
across the conductor cross-sectional area.
However, if the current is an AC, then the applied field in the conductor is also AC. The AC E-field has
a maximum value at the conductor surface and decreases exponential towards the center of the conductor.
The E-field distribution across the width of the conductor is shown in Figure 7.27(c) for an AC E-field. As
the field strength near the surface of the conductor is high, the current density near the surface is higher
than the central part of the conductor. This effect is called the skin effect.
The depth to which the E-field penetrates across the width of the conductor is measured by means of the
parameter called the skin depth. The skin depth is defined as the distance across the width of the conductor from
the conductor surface to the point where the current density is 1/e times the surface current density. It is denoted
as sd. The skin depth is given as
ρ
σd = m (7.82)
π ⋅ μo ⋅ μ r ⋅ f
where r is the resistivity of the conductor material; mo the permeability of air; mr the relative permeability of
the conductor; f the frequency of the current Hz.
For copper at 100°C, the resistivity is 2.3 × 10−8 Ωm and relative permeability is 1. Thus for copper
7.6
σd = cm (7.83)
f
It can be observed that the skin depth is dependent on the frequency. As the frequency increases, the skin
depth reduces implying that at sufficiently high frequency the central portion of the conductor will not con-
duct any current and all the charges are concentrated near the surface of the conductor. This will decrease
the effective cross-sectional area and increase the conductor resistance. As a consequence, the increase in the
current density near the surface will increase the copper loss and heat up the conductor.
This problem can be reduced by splitting the conductor cross-section in multiple section conductors,
that is, using multi-strand conductors wherein each strand is of a cross-section much small than the skin
depth. In such a case the E-field will penetrate across the width of the conductor and the field distribution
across the width will be more uniform. As a consequence, the conductor cross-section is better utilized
resulting in lower copper losses. For very high frequency applications in excess of 200 kHz, the transformer
windings are wound with multi-strand wires (like litz wires) to reduce the copper loss due to skin effect.
Proximity Effect
Consider two conductors wherein the currents are flowing in opposite directions. Figure 7.28(a) shows two
conductor cross-sections wherein an “×” indicates current flowing into the page and “•” indicates current
flowing out of the page. Due to the current flowing in the conductors the H-field pattern around each
conductor is as indicated. The far field around both conductors has a tendency to cancel each other and the
H-field between the conductors strengthens. If the two conductors have diameters that are greater than the
skin depth at the frequency of operation, then it would be expected that the current density be higher near
the surface as shown in Figure 7.28(b). However, the high-frequency current charges accumulate along one
side of the surface to minimize the magnetic field energy (nature’s minimum energy principle) and minimize
Chapter 07.indd 413 3/13/2009 7:08:48 PM

X
X
H
(a)
H H
(b)
(c)
Figure 7.28 (a) H-field due to two conductors carrying equal and opposite currents; (b) charge
accumulation due to skin and proximity effects; (c) proximity effect representation.
the inductive storage. Therefore, the current charges accumulate near the surface of the conductors opposite
to each other as shown in Figure 7.28(b). This effect is called the proximity effect. Proximity effect and skin
effect together makes the effective cross-sectional area of the conductor very small, leading to high copper
losses. Figure 7.28(c) shows the symbolic representation of the proximity effect.
Effect Due to Unequal Currents

The proximity effect due to equal currents was discussed above. However, the proximity effect due to unequal
currents has a much greater effect on the copper losses. The situation of unequal currents in the windings is
more common than the equal current situation discussed above. Consider two conductors: one primary wind-
ing conductor and another secondary winding conductor that are in close proximity as shown in Figure 7.29.
Let an “×” indicate a current of magnitude I flowing into the page and “•” indicate a current of magnitude
I flowing out of the page. The transformer is designed such that the primary current carries a current of I
and the secondary current carries a current of 2I. If the frequency is high and the conductors are thicker
than the skin depth at that frequency then the current charges in the primary and the secondary will accu-
mulate at the surface of the conductors facing each other as shown in Figure 7.29. There are two dark dots
shown in the secondary conductor representing a current of 2I. There is one dark “×” mark shown in the
primary conductor representing a current of I flowing in the conductor. Due to the proximity effect, the
charges are accumulated close to each other near the surface of each conductor. The far field due to one dot
of the secondary and one “×” of the primary will cancel; however, the other dot of the secondary will pro-
duce an H-field around it and induce a current of magnitude I to flow in the primary. This is marked as a
light “×” mark. This induced current will circulate through the outer periphery as shown in the figure. This
is shown as a light dot mark on the outer periphery. It should be observed that the total current through the
Chapter 07.indd 414 3/13/2009 7:08:49 PM

Primary winding conductor Secondary winding conductor
I I
−I
I I
I 2I
Figure 7.29 Proximity effect due to unequal currents in conductors.
primary is still maintained as I. However, the I2R loss is contributed due to all the current components flow-
ing in the primary. Thus the loss is (3Irms)2R as against Irms2R without proximity effect. Thus there is a nine
times increase in the copper loss for twice the current in the secondary as that in the primary. With multiple
layers, this has a further multiplying effect.
Thus proximity effect along with skin effect can be dangerously detrimental to the transformer especially
at high-frequency operation. The solution is to use multi-strand wires and inter-layering the primary and
secondary windings.
Inductors and transformers are ubiquitious in any The area product approach is widely followed in
system, especially in power electronics systems. A thor- the design of the inductors and transformers. It is a
ough understanding of the magnetic domain principles very systematic and algorithmic approach that is
is central to analysis and design of power electronics easily amenable for computer-aided design and anal-
circuits. As discussed in this chapter, the Ampere’s Law ysis. However, one should note that there are a
and the Faraday’s Law are two fundamental laws that couple of parameters that need to be decided by
lay the foundation for the understanding of the mag- engineering judgment. One of the parameters is the
netic domain aspects. These two laws form the frame- current density J. Throughout the chapter, a default
work for modeling of magnetic components. current density of 3 A/mm2 has been used in the
Chapter 07.indd 415 3/13/2009 7:08:49 PM

examples. This seems to be a good compromise The skin effect and the proximity effects are pre-
between the copper loss at high current densities dominant in circuits where the switching frequency
and the large core size requirement at low current is in excess of 100 kHz. These effects will have a
densities. However, the current densities can be drastic effect of bringing down the efficiency of the
selected in the neighborhood of the default value of system nullifying the primary objective of high-
3 A/mm2 depending on the application. In very low frequency switching which is to reduce the size of
power applications, the current densities as high as the magnetic components. Multi-strand wires are
5–6 A/mm2 have also been used with consequent probably the best solution to reduce the effects of
increase in copper loss. In high-power applications, skin and proximity effects.
one can go as low as 2 A/mm2 so that significant One more frequency-dependent effect that needs
savings in the copper loss offsets the increase in the to be borne in mind is the core loss, that is, hysteresis
core size. The other variable that is left to engineer- and eddy current losses. Both increase with frequency.
ing judgment is the window utilization factor. To reduce the hysteresis loss at high-frequency opera-
Though this is in the range of 0.3 to 0.6 based on tion, first core materials with narrow hysteresis loop
the factors discussed in this chapter, it is also depen- should be selected. Second, one may reduce the maxi-
dent on the skill with which the windings and the mum operating flux density to a much lower value to
inter-winding insulation are incorporated. reduce the B–H loop area in a cycle.
1. Consider the inductor measurement circuit Mode of implementation: The above circuit
shown in Figure 7.30. The inductor L is the can be studied by
test inductor that is connected to the drain of a a. Simulation in Spice
metal oxide semiconductor field effect transis- b. Simulation in SciLAB
tor (MOSFET) as shown. A diode–resistor c. Hardware bread-boarding
combination is used as the inductor current
Tasks for study:
freewheeling circuit.
(a) Select a ferrite core from the stores or buy
one (say E 42/21/15). From the datasheet
of the core material find out the relative
Vdd permeability, mean magnetic length and
the core cross-sectional area. Calculate the
permeance of the core. Wind 10 turns on
the core bobbin and mount the bobbin
onto the core. For the calculated perme-
L
ance and the number of turns used for the
R coil, estimate the inductance.
(b) Rig up the circuit as shown in Figure 7.30
with appropriately selected component
values.
(c) Connect a pulse generator and appropriate
drive circuit to the gate of the MOSFET. Set
the switching frequency to 20 kHz and keep
Figure 7.30 Inductor measurement circuit. the duty cycle close to 0. Increase the duty
Chapter 07.indd 416 3/13/2009 7:08:49 PM

cycle slightly and simultaneously observe the Mode of implementation: The above circuit
voltage across the MOSFET and the current can be studied by
through the inductor on an oscilloscope. a. Simulation in Spice
When the switch is ON the current through b. Simulation in SciLAB
the inductor will be a linearly rising wave- c. Hardware bread-boarding
form. Measure the slope of the inductor cur-
Tasks for study:
rent waveform when the switch is ON. Also
(a) Take an unknown and unmarked trans-
measure the voltage across the inductor.
former (low-frequency lamination type) and
From Faraday’s Law estimate the inductance
rig up the circuit as shown in Figure 7.31.
value. Compare with the calculated induc-
Before rigging up the circuit, measure the
tance value of Step (a).
physical dimensions and estimate the core
(d) Continue increasing the duty cycle till the
cross-section area and the mean magnetic
current through the inductor starts to
path length.
shoot up. Why does this happen? Observe
(b) The input voltage source is obtained from
the voltage across the inductor.
the 230 V grid. An auto-transformer is
(e) Increase the number of turns in the coil to
connected to the grid to obtain a variable
20. What is the effect on the inductance?
sinusoidal voltage from the auto-trans-
(f ) Introduce an air gap of 0.25 mm in the
former tap point to ground. Set the auto-
central limb of the E core. Calculate the
transformer tap point to ground. While
permeance and the inductance. What is the
connecting the load resistor, change the
effect of air gap on the inductance? Experi-
auto-transformer tap point slowly to obtain
mentally measure the inductance as in Step
5 V. Measure the voltages VAB and VCD on
(c) and compare with the calculated value.
the oscilloscope. Determine the dot polari-
(g) For different values of the duty cycle, the
ties of the primary and the secondary
peak inductor current values will be differ-
windings. Determine also the turns ratio.
ent. Tabulate the inductor current peak
(c) Take a very low copper gauge wire, say SWG
values at different duty cycles and estimate
45 and thread through the transformer
the magnetization field H.
about 10 turns. Apply a very low voltage to
2. Consider the transformer test circuit shown in this 10-turn winding using the auto-trans-
Figure 7.31. The transformer with terminals A, former variable source. Measure the voltage
B, C and D is the test transformer. Let AB be across the 10-turn winding and the voltages
considered as the primary and CD as the across AB and CD. Estimate the number of
secondary. Vi is an input source of an arbitrary turns in the AB winding and the number of
waveshape and RL is variable load resistor. turns in the CD winding. Remove the tem-
porary 10-turn winding.
(d) From Eq. (7.50b) estimate the voltage that
ip
A C is may be applied to AB. (As the laminations
is of either CRGO or CRNGO material,
assume Bm of 1 T for now.)
Vi Es RL (e) Without connecting the load and by vary-
ing the auto-transformer setting, apply
half the voltage to AB that is estimated in
B D
Step (d). Measure the primary voltage VAB
and the primary current through the
Figure 7.31 Transformer test circuit. winding AB. VAB/NAB gives the flux rate
Chapter 07.indd 417 3/13/2009 7:08:49 PM

within the core. Build an integrator circuit permeability of the core in the linear region.
using op-amp, sense VAB by using an isola- After estimating the relative permeability,
tion transformer or by resistive attenua- calculate the permeance of the core. Calcu-
tion and integrate the ratio VAB/NAB to late the mutual inductance of the core. Esti-
obtain the flux within the core. Divide mate the B–H loop area and calculate the
this output by the core cross-sectional area core loss at the operating frequency. Tabu-
to obtain the flux density within the core. late the core loss at various input voltage (by
(f) From the measured primary current, esti- varying the auto-transformer setting).
mate the mmf in the core by multiplying (i) Short circuit the secondary and measure
the number of turns in AB winding and the inductance as seen at AB. This will give
the current through it. Sense the current the sum of the primary leakage and the
through the winding by using a CT or by reflected secondary leakage in parallel with
measuring the drop across a 0.1 Ω resistor the mutual inductance.
connected in series with the winding. The (j) Short circuit the primary and measure
voltage equivalent of the sensed current is the inductance as seen at CD. This will
multiplied by NAB/lm to obtain the magne- give the sum of the secondary leakage
tization field H. and the primary leakage in parallel with
(g) As VAB is varying sinusoidally and the pri- the mutual inductance that is reflected to
mary current is also varying sinusoidally, the secondary.
the flux density obtained in Step (e) and (k) Measure the winding diameters (using
the magnetization field obtained in Step (f ) screw gauge) and estimate the winding
are also AC waveforms. With these two cross-section area for the primary and sec-
waveforms, plot the B–H characteristics. ondary windings. Measure the resistances
Some oscilloscopes have “integrate” facility. of the winding. From the resistivity of
In that case, the scope channel measuring copper, determine the length of the copper
VAB can be integrated in the scope itself and wire used for the primary and the second-
the current can be given to the external ary windings.
trigger to see the scaled B–H characteristic (l) Based on the above measurements, con-
shape on the oscilloscope. struct the equivalent circuit for the test
(h) Increase the auto-transformer voltage slowly transformer.
till the B–H curve just enters into saturation (m) What is the effect of variation in load resis-
region. Tabulate the maximum Bm and the tor RL on the B–H characteristic and the
magnetization field at maximum flux den- equivalent circuit parameters?
sity. From the B–H plot estimate the relative

1. The magnetics components are categorized as 4. The flux density is related to the flux as
energy- devices and energy- .
devices.
5. The power associated with the saturated core in
2. The energy- devices are called the the magnetic domain is .
transformers.
6. The volt-second product for a coil in terms of
3. The energy- devices are called induc- the physical parameters of the core is given as
tors. .
Chapter 07.indd 418 3/13/2009 7:08:49 PM

7. For a transformer of a given rated voltage, if a 18. The primary current needed to supply is core
lower frequency is applied, the will losses and to set up the magnetic flux is given
increase and may saturate the core. as .
8. The voltage across a coil should have a 19. In the equation, E p-av = 2 ⋅ N p ⋅ Ac ⋅ ΔB ⋅ f , the
average over a full cycle.
variable Ep-av is the average of the
9. In an inductor, the energy is stored by virtue of induced voltage across the primary winding.
the flowing through the inductor in
20. Current transformers are mainly used to
the electrical domain.
the currents in the power circuits.
10. The energy-storage mechanism within the
21. Short circuiting the secondary winding in the
magnetic domain is a storage similar
case of the current transformer implies
to that of storage in electric
operation.
domain.
22. As CTs are used as current sensing devices, the
11. The permeance is equivalent to a
main focus is on maintaining the current wave-
within the magnetic domain.
shape .
12. The core cross sectional area is related to the
23. Rod cores are mostly used in radios for
across the coil.
an inductor.
13. The window area of the core is related to the
24. In rod core type of inductor, radiation occurs
capacity.
at the end of the rod and will lead to genera-
14. The area product relates to the that tion of radiation.
is stored in the core or the that is
25. The pot core has a shielding effect, preventing
transferred through the core.
radiation and electromagnetic inter-
15. The flyback transformer is an energy- ference.
device. It is nothing but a multiple-winding
26. Toroidal cores are used in trans-
.
formers that are popular in amplifiers, but it is
16. If the power source in the primary of a trans- not easy to a coil on toroidal cores.
former is a voltage source, then the transformer
27. The hysteresis power loss with the
is called a .
frequency of operation.
17. If the power source in the primary of a trans-
28. The current density near the centre of the con-
former is a current source, then the transformer
ductor is than that near the surface
is called a .
of the conductor due to skin effect.
1. What are the two fundamental laws of electro- 3. What is Faraday’s Law?
magnetism that link voltage and current in the
4. To what equivalent variables do voltage, cur-
electrical domain to equivalent quantities in
rent and charge of the electric domain map to
the magnetic domain?
in the magnetic domain?
2. What is Ampere’s Law?
Chapter 07.indd 419 3/13/2009 7:08:49 PM

5. What is the equivalent of resistance in the mag- 20. Explain the differences between the current
netic domain? transformer and the potential transformer.
6. What is the equivalent of capacitance in the 21. In what ways can the magnetizing current be
magnetic domain? reduced in a CT?
7. Distinguish between resistance, capacitance, 22. What are the physical parameter requirements
permeance and reluctance. for a good quality CT?
8. How are magnetic field intensity and flux den- 23. What are the factors that affect the utilization
sity related? of the window area of a core? Discuss.
9. Explain the B–H characteristics and the hyster- 24. Discuss the various shapes of the commercial
esis concept. cores illustrating the core cross-sectional area
and the window area in the front and profile
10. Explain the volt-second balance principle.
views of a core section?
11. What is the effect on the B–H characteristic of
25. Explain how one would test the dot polarity of
a core when an air gap is introduced?
an unknown and unmarked transformer.
12. Discuss the window utilization factor.
26. Discuss the hysteresis loss in the cores. What is
13. Explain how the flux rate within the core is the effect of frequency on the loss?
maintained constant.
27. What are the factors affecting the hysteresis loss
14. Explain how the change of load in the second- in a core?
ary winding gets reflected on the primary side.
28. What is eddy current loss and how can this be
15. Under no-load operation, is there any current minimized?
drawn from the input source connected to the
29. What are the factors affecting the eddy current
primary? Why is this current needed?
loss?
16. What is turns ratio? Discuss.
30. Explain the skin-effect phenomenon. How is
17. What is leakage flux? Explain. the effect of skin effect reduced in the windings
of high-frequency transformers?
18. Explain the equivalent circuit of the transformer.
31. Explain the proximity effect. In what ways can
19. What are the features of an ideal transformer
the proximity effect be reduced?
and how does a practical transformer deviate
from ideality?
| PROBLEMS
1. A ferrite EE core having a cross-section area of core is 20 ampere-turns. What is the flux in
182 mm2, has a mean magnetic path length of the core?
97.2 mm. If the permeance is 4.7 μH/turns2,
3. For Problem 2, if the number of turns in the coil
find the reluctance and the relative permeabil-
winding is 10, what is the inductance of the coil?
ity of the core.
4. 100 turns of copper wire are wound on a cylin-
2. An inductor core has a permeance of 4.7 μH/
drical plastic bobbin in a single layer. The outer
turns2. The magneto-motive force within the
Chapter 07.indd 420 3/13/2009 7:08:49 PM

diameter of the bobbin is 16 mm and the the permeance if the relative permeability of
length of the coil is 50 mm. If no core is the core is 2000 and the introduced air-gap
inserted into the bobbin, what is the induc- length is 0.5 mm.
tance of the coil?
10. If the required inductance is 100 mH and the
5. For Problem 4, if the bobbin is inserted in a tight permeance is as calculated in Problem 9 for the
fitting CC ferrite core of relative permeability ferrite core E 42/21/9, estimate the number of
2000 having a mean magnetic path length of 150 turns required to be wound on the core.
mm, then what is the inductance of the coil?
11. For a 10 A DC current flowing through the
6. An inductor core has a permeance of 10 μH/ inductor coil, calculate the wire cross-section that
turns2. The magneto-motive force within the is required for the coil wire. Also select an appro-
core is 25 ampere-turns. What is the energy priate gauge of the wire from Appendix III.
stored in the inductor?
12. A transformer having 1000 turns in the pri-
7. For Problem 4, if a ferrite rod of length 75 mm mary is connected to voltage source whose
is tightly fitted into the bobbin, then what is the waveshape is given as 100 sinwt. What is the
inductance of the coil if the ferrite material has a waveform of the flux rate?
relative permeability of 2000? Also calculate the
13. A transformer is designed to take power from an
reluctance of the magnetic path. (Hint: The flux
input source rated at 230 V rms, 50 Hz sinusoi-
path is through the ferrite rod and returns back
dal source. The transformer comprises two sec-
through approximately an equal distance in air.)
ondary windings that supply power to a 100 V,
8. A particular DC–DC converter that is switched 2 A and 50 V, 5 A loads. Assuming CRGO sili-
at 50 kHz, uses a 100 μH inductor wherein con steel laminations for the transformer and an
10 A of DC current flows through it. The ripple efficiency of 80%, estimate the required area
of 1 A peak to peak is superimposed on the 10 product such that the maximum operating flux
A DC current. Calculate the area product density in the core does not exceed 1 T? Select an
requirement for a ferrite cored inductor. Also appropriate lamination type from Appendix II.
select an appropriate core from Appendix II.
14. For Problem 13, find the turns ratios and cal-
9. For a ferrite core E 42/21/9 whose physical culate the number of turns for the primary and
parameters are given in Appendix II, calculate the secondary windings.
| ANSWERS
Fill in the Blanks
1. transfer; storage 10. potential; capacitance 19. half-period
2. transfer 11. capacitance 20. sense
3. storage 12. voltage 21. no-load
4. B = f/Ac 13. current 22. fidelity
5. zero 14. energy; power 23. tuning
6. E avgT = 2 NAc B 15. storing; inductor 24. electromagnetic
7. EavgT 16. potential transformer 25. reducing
8. zero 17. current transformer 26. low radiation; wind
9. current 18. ipo = ip − n ⋅ is 27. increases
28. lesser
Chapter 07.indd 421 3/13/2009 7:08:49 PM

Chapter 07.indd 421 3/13/2009 7:08:49 PM
Modeling of Systems
Learning Objectives
CHAPTER
8
make a mathematical representation of a physical system.
obtain the state space model of a physical system.
obtain the transfer function model of a physical system.
compare and use various modeling approaches.
apply circuit averaging, bond graph and space-vector modeling methods for power elec-
tronics systems.
T ill now, the models used to describe the systems in terms of the input–output relationships of the
converters were steady-state models. These models reflect the state of system under equilibrium con-
ditions. However, the dynamics of the systems are important from the point of view of controller design
for regulation and reference tracking. The most important task confronting the system analyst is developing
the dynamic model of the process to be controlled. In most cases, the central problem in any design situation
is modeling. Once the modeling is performed with a reasonable degree of precision, the rest of the analysis
becomes more or less a trivial issue. Determination of a mathematical model of a given physical system is
an important problem in engineering design and analysis. The model must relate the various variables in
the system in a quantitative manner. A model may be defined as “a representation of the essential aspects
of a system which presents knowledge of that system in a usable manner”. To be useful, a model must not
be so complicated that it cannot be understood and thereby be unsuitable for analysis; at the same time, it
must not be oversimplified and trivial to the extent that predictions of the behavior of the system based on
this model are grossly inadequate.
The control system engineer is most often required to deal with systems that have many sub-systems
operating in different physical domains. The electromechanical process, for example, may comprise the DC
motor coupled to inertia load on one side and electrical source on the other. The electrical source itself will
have its associated dynamics and the mechanical inertia load would have its own dynamics that needs to be
controlled. Further, the DC motor itself includes the magnetic domain with associated dynamics. Therefore,
for the control system engineer to design a practical meaningful controller it is required to first model the
entire system with all its different domains and physical laws such that the system is represented in a
mathematical form that is amenable for analysis, computation, simulation and design.
Chapter 08.indd 423 3/14/2009 2:38:07 PM

As the control system engineer has to deal with systems having multi- and interdisciplinary domains,
it implies that one has to have a deeper understanding of the various sub-systems that go to make up the
whole system. Therefore, for a control systems engineer, knowledge of various domains is very impor-
tant. The information about the various sub-systems of the different domains should be translated into
a form that is compatible for controller synthesis. The equations and formulae that predict the dynamic
behavior of the system as a whole should be established. The behavior of the overall system can then be
analyzed and characterized so that a controller may be designed to meet the desired performance
specifications.
There are basically two modeling approaches: (a) system identification and (b) physical modeling. System
identification is generally associated with automated computer-aided modeling. The system is considered as a
black box, the output of which is characterized for various inputs. Various inputs like sinusoidal, step, etc. are
applied to the actual system. The response to these various types of inputs are characterized as transfer curves
to obtain the input–output relationships. Based on the transfer curves, a mathematical curve fit algorithm is
proposed such that it gives the best description of the system. System identification technique is generally
used in on-line estimation of system parameters like time constants, resistances, etc. for self-tuning and
adaptive controllers.
The physical modeling techniques, on the other hand, obtain the mathematical representation of the
physical system from first principles. This provides a much deeper insight into the behavior of the various
dynamics of the system as compared to the system identification methods. The system is represented as a set
of differential equations that describe the dynamic behavior of the system. For designing controllers it is
essential to obtain the mathematical representation from first principles so that it will aid in the analysis and
synthesis of the system to be controlled together with the controller. This chapter focuses on obtaining the
physical model of systems from first principles in the differential equation form.
The physical systems that are considered for analysis are dynamic in nature, and their behavior will be
described in the form of differential equations. Although these will normally be non-linear, it is customary to
linearize them about an operating point to obtain linear differential equations. This is done for the analysis to
be carried out conveniently. It should, however, be noted that such linear models, though useful for analysis
and design, are valid only over a limited operating range. Nevertheless, they are employed extensively in
engineering.
The central issue in any control system analysis or design study is to first obtain a reasonable mathe-
matical representation of the physical system. In obtaining a mathematical model of a given physical
system, one must make a compromise between simplicity of representation and accuracy of the prediction
of the system behavior. In deriving a model, it may be necessary to ignore certain physical properties of the
system such as distributed capacitances and inductance, non-linearities such as saturation effects that may
be present in the system. Only those properties that do not have a significant effect on the response should
be removed. This will lead to simpler mathematical models having reasonable agreement with the experi-
mental study.
In general, it is important to note that one should build a simplified model of the system so that an
intuitive understanding of the behavior of the system is possible. For more complete and precise analysis, a
more detailed model can later be evolved from the simplified model. It should also be borne in mind that
while one can make reasonable approximations to obtain simplified models of the system, the simplified
models may not be valid at other operating points. For example, if the distributed capacitances and inductances
are replaced by lumped parameter approximations, the model though valid at low frequencies may not be
valid at high frequencies. Therefore for every model, it is important to understand its limitations with
respect to the range of operation.
Chapter 08.indd 424 3/14/2009 7:52:53 AM

Modeling of Systems 425
8.1 Input–Output Relations
C onsider a system which is shown as a black box in Figure 8.1. It has an input u and an output y as
indicated. The aim of modeling is to predict the output y for any given input u. The output y is called
the response of the system and u is called the stimulus or simply the input to the system.
The input–output relations for the system may be given in a graphical form called input–output trans-
fer curves or input–output characteristic curves as shown in Figure 8.2. These curves could be obtained by
plotting the experimental response obtained for various input signal values. The curves are plotted with the
input u as the independent variable (x-axis) and the response y as the dependent variable ( y-axis).
System
u y
Figure 8.1 System as a black box.
y y y
0 u 0 u 0 u
(a) (b) (c)
y y y
0 u
0 u 0 u
(d) (e) (f)
Figure 8.2 Different types of input–output transfer curves: (a) Linear; (b) non-linear
saturation effect; (c) non-linear dead zone effect; (d) non-linear switch;
(e) non-linear hysteresis; (f) non-linear exponential law.
Chapter 08.indd 425 3/14/2009 7:52:53 AM

Figure 8.2(a) shows the input–output transfer curve of a linear system wherein the response is propor-
tional to the input. It should be noted here that in a linear system the most important property is that the
principle of superposition is applicable. This means that the response produced by the simultaneous applica-
tion of two different stimuli is the sum of the two individual responses. A system wherein the principle of
superposition is not applicable is called a non-linear system. Figures 8.2(b)–(f ) give some typical input–
output transfer curves for non-linear systems. In Figure 8.2(b), one can see the saturation effect wherein the
response y varies linearly with u upto a point. Beyond some value of u, the system output clamps to a specific
value. Saturation is very typical of most physical systems. Figure 8.2(c) is an example of dead zone non-linearity
wherein at very low values of the input signal u, there is no response from the system. The system will respond
only if an input above a threshold value is applied. Figure 8.2(d) is an example of a switch. Here the output is
a two-state system wherein it will be high for inputs that are greater than zero or a specified threshold value
and the output will be low for inputs that are lower than zero or a specified threshold value. Figure 8.2(e) is
an example of hysteresis non-linearity. Here the output follows one path for increasing input u and follows
another path for decreasing input u. This is typical of many dynamical systems. The B–H curves of magnetic
materials are a typical example of hysteresis non-linearity. Figure 8.2(f ) shows an example of exponential non-
linearity wherein y = e u. Similarly, the output can also follow square law, hyperbolic law, etc.
The input–output relationships can also be represented in the form of equations. Some examples are as
follows:
1. Linear System: y = ku where k is a constant.
2. Non-Linear System:
• y = u 2. This input–output relation shows square law dependency.
• y = cos u . This input–output relation shows cosine dependency.
8.2 Differential Equations and Linearization
T he input–output transfer curves do not give any information on the time evolution of the output to a
given system input. When an input u is applied to the system, the output will finally settle down to a
value as given by the input–output transfer curve. However, one is also interested to trace the time evolution
of the output to its final settling value on the application of an input. This is called the dynamic behavior
of the system and is represented by a set of differential equations. In practice, some simplifying engineering
assumptions are often made to obtain linear differential equations with constant coefficients, although in
most cases exact analysis would require the use of non-linear partial differential equations.
Consider a system with input u(t) and output y(t). Let it be a non-linear system with the input–output
transfer curve as given in Figure 8.3. The input–output relationship is given by
y = f (u ) (8.1)
Referring to Figure 8.3, the nominal input to the system is u and the resulting output is y . Therefore,
(u , y ) is the nominal operating point of the system. In a very small neighborhood about the operating
point given by ( Δu , Δy ) , the system operation can be considered to be linear. For non-linear systems, the
linear differential equations will be obtained for a very small neighborhood about the operating point such
that the system can be considered to be linear in that small neighborhood about the operating point.
Equation (8.1) can be expanded into a Taylor series about the nominal operating point (u , y ) as
df 1 d2 f
y = f (u ) = f (u ) + (u − u ) + (u − u )2 + ⋅⋅⋅ (8.2)
du 2 ! du 2
Chapter 08.indd 426 3/14/2009 7:52:53 AM

y Δu
y Δy
0 u u
Figure 8.3 Operating point and neighborhood of operating point.
In Eq. (8.2), the derivatives df /du, d2f /du2, etc. are evaluated at u . If the deviation (u − u ) is small, then
the second- and higher order terms may be neglected. Equation (8.2) may be simplified as
y = y + k (u − u ) (8.3a)
where
y = f (u ) (8.3b)
and k = df/du u = u− which may be considered to be a constant in the neighborhood about the operating
point. Equation (8.3a) may be re-written as
Δy = k ⋅ Δu (8.4)
where
Δy = y − y
Δu = u − u
Equation (8.4) shows that the deviation in the output is proportional to the deviation in the input and is the
linearized mathematical model of the non-linear system given by Eq. (8.1). Equation (8.1) is called the
large-signal model of the system. Equation (8.3b) is called the steady-state or equilibrium model of the system.
Equation (8.4) is called the small-signal model of the system which is obtained by subtracting the steady-state
model from the large-signal model.
Control system deals with components which are diverse in nature and may include electrical, mechani-
cal, thermal and hydraulic systems. The differential equations for these devices are obtained using the basic
laws of physics. For most physical systems one may classify the variables as “through” and “across” variable.
The “through” variable can also be described as “flow” variable and the “across” variable can also be described
as the “effort” variable. A list of analogous variables for different systems is given in Table 8.1.
Chapter 08.indd 427 3/14/2009 7:52:53 AM

Table 8.1 Analogous variables for physical systems
System Through or flow variable Across or effort variable
Electrical Current, I Voltage, e
Mechanical (translation) Velocity, n Force, f
Mechanical (rotational) Angular velocity, w Torque, T
Thermal Rate of flow of heat energy, q Difference in temperature, ΔT
Hydraulic Volumetric rate of flow of fluid Difference in pressure, ΔP

flow, dQ/dt
R
v1(t) C v2(t)
Figure 8.4 A simple low-pass filter.
It may be noted that the equations of equilibrium in different systems are based on energy conservation
principles. Kirchhoff ’s voltage law (KVL) for an electrical circuit, equating the algebraic sum of voltages in a loop
to zero, is analogous to D’Alembert’s principle in mechanics, which equates the algebraic sum of forces at a point
to zero. These principles are used to obtain the differential equations characterizing the dynamics of the system.
For example, consider the simple electrical circuit shown in Figure 8.4, wherein an input voltage v1(t) is
applied to an RC network. By the application of KVL to the circuit of Figure 8.4, the output and input are
related as
v1(t ) = R ⋅ i + v 2 (t )
dv 2 (8.5)
v1(t ) = R ⋅ C + v 2 (t )
dt
Equation (8.5) represents the mathematical model of the RC low-pass filter of Figure 8.4. The solution to
the differential equation given in Eq. (8.5) gives the prediction of the dynamic behavior of the RC low-pass
filter.
8.3 State Space Representation
T he notion of the state of a dynamical system is a fundamental concept which is central to the analysis
of physical systems. The future evolution of a dynamical system is entirely determined by its present
state. The state of a dynamic system may be defined as
the smallest set of physical quantities called state variables that completely determines the evolution of the
system in the absence of external excitation.
Chapter 08.indd 428 3/14/2009 7:52:54 AM

For any given physical system, there exists a set of physical quantities called state variables. For a system
where the input is removed, the system behavior is entirely decided by the values of the state variables. The
values of the state variables at any given point of time are called the state of the system. From the definition,
it also implies that the values of the state variables at a given point of time along with the values of the inputs
entirely dictate the future evolution of the state of the system.
The specific physical quantities (state variables) that define the system state are not unique. However, the
number of state variables called the order of the system is unique for a given system. In many cases the
choice of state variables is not very obvious. For a variable to qualify for consideration as a state variable, it
should have the important property of memory. This means that it is capable of storing immediate past his-
tory. The following can qualify as state variables:
1. A memory unit of a digital system.
2. A variable that is representable in integral form. This is due to the fact that an integrator has memory
effect. Inductor currents are representable in integral form as
1
L∫
i= v ⋅ dt
Therefore an inductor current can qualify for consideration as a state variable. Capacitor voltages are
representable in integral form as
1
v = ∫ i ⋅ dt
C
Therefore, capacitor voltages can also qualify for consideration as a state variable. Based on similar argu-
ment, one can state that all the “flow” or “through” variables of inductor type or inertial type elements
in any domain can qualify to be state variables. This means that velocity of a mass, angular velocity of
inertia, etc. can qualify as state variables. Likewise, all the “effort” or “across” variables of capacitor type
elements in any domain can qualify to be state variables. This means that force of a spring, temperature
of a thermal capacitance, mmf of a magnetic permeance, etc. can qualify as state variables.
Another interesting point to be noted is that as the state variables should have memory property; they
should be associated with energy-storing elements like inductors and capacitors only. Voltages across and
current through resistors cannot qualify for being considered as state variables. In inductors the energy stor-
age is due to current which is given by (1/2)Li 2 and in capacitors the energy is stored due to the voltage
across the capacitors which is given by (1/2)Cv 2. Therefore, as a rule the variable responsible for energy stor-
age in the devices can qualify for being considered as state variable.
For the RC low-pass filter circuit shown in Figure 8.4, only the voltage across the capacitor v2(t) can
qualify for being a state variable. The state variable v2(t) along with the input v1(t) completely describes the
dynamic behavior of the circuit of Figure 8.4. The differential equation [Eq. (8.5)] can be re-written as
dv 2 −1 1
= v2 + v (8.6)
dt RC RC 1
This is the standard form in which the system is represented and is called the state equation. This differential
equation is entirely a function of state variables and the input variable. If there are n state variables then the
system is said to be of order n. Such an order n system would be represented by n linear differential equa-
tions which are all functions of only the state variables and the input. Equation (8.6) can be further stan-
dardized as
x = a ⋅ x + b ⋅ u (8.7)
Chapter 08.indd 429 3/14/2009 7:52:54 AM

Equation (8.7) is called the state equation. With respect to the RC network circuit that is represented by
Eq. (8.6), the parameters of the standard form of the state equation [Eq. (8.7)] can be related as
dv 2
x =
dt
−1
a=
RC
1
b=
RC
One can now explicitly also give the output as a function of only the state variables and the input variables.
This can be given in a standard form as
y = c ⋅ x + d ⋅u (8.8)
Equation (8.8) is called the output equation. With respect to the RC network of Figure 8.4, the output
equation is as given in Eq. (8.8) with the following parameters:
y = v2
c =1
d=0
Equations (8.7) and (8.8) together represent the system in state space form. For more than one state, x becomes
a vector and a, b, c and d become the matrices. Therefore, for an nth order system, there will be n state variables
that define the state space.
Consider an RLC circuit as shown in Figure 8.5. The circuit is excited by an input vi to obtain a response
across the output vo. Here the current (i) through the inductor and the voltage across the capacitor (vo) qual-
ify to be considered as state variables. Therefore, the entire RLC system will be completely described by the
two state variables (i, vo) and the input variable vi. As there are two state variables, the order of the system is 2.
One can expect two linear first-order differential equations as a function of the two state variables and the
input variable. The two linear first-order differential equations that represent the RLC circuit are obtained
using KVL. They are
di
v i = iR + L + vo
dt
dvo (8.9)
i =C
dt
R L
vi C vo
Figure 8.5 RLC circuit.
Chapter 08.indd 430 3/14/2009 7:52:54 AM

Equation (8.9) can be re-written in the standard form as
di − R −1 1
= ⋅ i + ⋅ vo + ⋅ vi
dt L L L (8.10)
dvo 1
= ⋅ i + 0 ⋅ vo + 0 ⋅ vi
dt C
Equation (8.10) can be re-written in the matrix form as
⎡ di / dt ⎤ ⎡− R / L −1 / L ⎤ ⎡ i ⎤ ⎡1 / L ⎤
⎢ ⎥=⎢ ⎥⋅⎢ ⎥ + ⎢ ⎥ ⋅v (8.11)
⎣dvo / dt ⎦ ⎣ 1 / C 0 ⎦ ⎣v o ⎦ ⎣ 0 ⎦ i
Equation (8.11) is the state equation which represents the dynamic behavior of the RLC circuit of Figure 8.5.
The output equation is given as
⎡i ⎤
y = [0 1] ⋅ ⎢ ⎥ + ⎡⎣0 ⎤⎦ ⋅ v i (8.12)
⎣v o ⎦
Equations (8.11) and (8.12) completely represent the RLC circuit in the state space form. Equations (8.11)
and (8.12) are in the general form
x = A ⋅ x + B ⋅ u
(8.13)
y = C ⋅ x + D ⋅u
For any system the mathematical model can be represented in the standard form of Eq. (8.13). Here x is the
n × 1 state vector which comprises the state variables for a system of order n, A is the n × n system parameter
matrix which consists of the system constants. The matrix A contains the system-specific information about
the dynamic behavior like the various time constants of the system. If the elements of A are constants of the
system, then the system is called a linear time-invariant (LTI) system. If the elements of A are functions of
time, then the system is called a linear time-varying (LTV) system. The matrix B is the n × m input matrix
which weights the direct input excitation for state variable. U is the m × 1 input vector for m input excita-
tions, y is the p × 1 output vector for p outputs from the system. C is a p × n output matrix and D is a p × m
matrix that gives the direct feed through component of the input excitation in the output response.
8.4 Transfer Function Representation
L aplace transformation of the differential equations yields an algebraic equation, in terms of the complex
frequency variable s, relating the various “through” and “across” variables. These algebraic equations can
then easily be manipulated to obtain the transfer function of the system, defined as the ratio of the Laplace
transforms of the output to the input under zero initial conditions. This transfer function represents a linear
model of the system and is usually shown in the form of a block diagram as indicated in Figure 8.6. Note
that the block is “unidirectional”.
G(s)
U(s) Y(s)
Figure 8.6 Block diagram of a linear system.
Chapter 08.indd 431 3/14/2009 7:52:54 AM

The input may be regarded as the “cause” and the output as the “effect”. The block diagram is unidirectional
since the “effect” cannot produce the “cause”. The transfer function G(s) relates the Laplace transform Y(s)
of the output y(t) to the Laplace transform U(s) of the input u(t) through the relationship
Y (s) = G (s)U (s) (8.14)

For example, consider the RC electrical circuit shown in Figure 8.4. The output voltage v2(t) is related to
the input through the differential equation as given in Eq. (8.5) which is repeated here for clarity:
dv 2
v1(t ) = R ⋅ C + v 2 (t )
dt
Taking the Laplace transform on both sides of the above equation and assuming zero initial conditions, one
obtains
V1(s) = RCsV2(s) + V2(s) (8.15)
From Eq. (8.15) and the definition of the transfer function, one obtains the transfer function as
V2 ( s ) 1
G(s ) = = (8.16)
V1( s ) 1 + sRC
The transfer functions are also basically derived from the differential equations. Therefore, the transfer func-
tion representation is derivable from the state space representation.
Consider the state space representation given in Eq. (8.13) which is repeated here for clarity:
x = A ⋅ x + B ⋅ u
y = C ⋅ x + D ⋅u
Taking the Laplace transform of the state equation yields
sX ( s ) − x (0) = AX ( s ) + BU ( s )
or
( sI − A ) X ( s ) = x (0) + BU ( s )
Pre-multiplying both sides of the above equation by (sI − A)−1, one obtains
X ( s ) = ( sI − A )−1 x (0) + ( sI − A )−1 BU ( s ) (8.17)
From the definition of the transfer function, the initial conditions are zero. Equation (8.17) under the con-
dition of zero initial condition is given as
X ( s ) = ( sI − A )−1 BU ( s ) (8.18)
From the output equation of the state space representation of Eq. (8.13), one obtains
Y ( s ) = CX ( s ) + DU ( s ) (8.19)
From Eqs. (8.18) and (8.19), the transfer function from the state space representation is given as
Y (s )
G(s ) = = C ( sI − A )−1 B + D (8.20)
U (s )
For complex systems, it is easier to obtain the state space representation in the form of a set of linear first-order
differential equations. This can later be used to obtain specific transfer functions for specific outputs with
Chapter 08.indd 432 3/14/2009 7:52:54 AM

respect to specific inputs. It should be noted that transfer functions are defined between a single input and a
single output, that is, only for a single-input single-output (SISO) system. Therefore, for multiple-input and
multiple-output (MIMO) systems one can represent the input–output relationship through a transfer function
matrix. Each element of this matrix is a transfer function relating a particular output to a specific input, assum-
ing that all the other inputs are zero. This is justified through the principle of superposition for linear systems.
For engineers with electrical background, the analysis of mechanical systems is often easier when an electrical
equivalent circuit analogous to a mechanical system is obtained. It has the advantage that one can apply
Kirchhoff ’s laws to write the circuit equations and hence obtain the transfer function. It is also possible to write
these equations directly in terms of the Laplace transforms of the currents and voltages. Furthermore, network
theorems can often be applied to simplify the circuit. Electrical analogs for mechanical systems have also been
used for simulation and analysis. The rule for drawing the equivalent electrical circuit may be stated as follows:
Each junction in the mechanical system corresponds to a node in the equivalent electrical circuit, join-
ing excitation sources and passive elements. All points on a rigid mass are considered as the same junction
and one terminal of the capacitor analogous to the mass is always connected to the ground in the electrical
circuit. The reason for connecting one terminal of the capacitor to the ground is that the velocity (or dis-
placement) of a mass is considered with respect to a reference.
The following examples will illustrate the procedure.
EXAMPLE 8.1 The electrical analog of a carriage on wheels, coupled to the wall through a spring,
is shown in Figure 8.7. The differential equation for both systems is as given in
Eq. (8.21). In the case of the electrical network, the equation is obtained by apply-
ing Kirchhoff ’s current law (KCL) at the node and is seen to be identical to the
equation that would have been obtained by equating the forces on all the compo-
nents to the applied external force in the mechanical system.
v = dx / dt
x K D M f (t )
K
M f (t )
D
(a) (b)
Figure 8.7 A mechanical system with movement in one co-ordinate:

(a) Mechanical system; (b) electrical analog.
d2x dx
M +D + Kx = f (t ) (8.21)
dt 2 dt
where M is the mass, K the spring stiffness, D the damper, x the displacement and
f the applied force. Taking Laplace transforms of both sides of Eq. (8.21) and assum-
ing zero initial conditions, one obtains the transfer function as
X (s ) 1
G(s ) = = (8.22)
F ( s ) Ms 2 + Ds + K
Chapter 08.indd 433 3/14/2009 7:52:54 AM

D2 v2 = dx2 /dt v1 = dx1 /dt

M1
D1
x1
D1
f M2 D2 M1 K
M2
x2
f (t )
(a) (b)
Figure 8.8 A two co-ordinate mechanical system: (a) Mechanical system; (b) electrical analog.
A mechanical system with two-co-ordinate movement and its equivalent electrical circuit are shown in
Figure 8.8, where K represents a spring and D1 and D2 represent the dashpots.
In this case the equations, written directly in terms of the Laplace transform variables, are obtained by
applying KCL at each of the two ungrounded nodes.
( s 2 M 2 + sD2 + sD1 ) X 2 ( s ) − sD1 X 1 ( s ) = F ( s )
sD1 X 2 ( s ) + ( s 2 M1 + sD1 + K ) X 1 ( s ) = 0 (8.23)

The convenience of writing Eq. (8.23) in terms of node voltages is evident. If one needs to obtain the trans-
fer function relating X2(s) to F(s), solve Eq. (8.23) for X2(s). Hence
X 2 (s ) s 2 M1 + sD1 + K
= 2 (8.24)
F ( s ) ( s M1 + sD1 + K )( s 2 M 2 + sD2 + sD1 ) − s 2 D12
Modeling an Armature-Controlled DC Motor

Figure 8.9 shows the schematic diagram of an armature-controlled DC servomotor. It will be assumed that
the field current is maintained constant and a voltage v(t) is applied to the armature, which has a resistance
Ra and negligible inductance. The effect of the application of the input voltage v(t) will cause the armature
to rotate. Applying the KVL to the electrical side of the system, one obtains
dia
v (t ) = Ra ia (t ) + La + v b (t ) (8.25)
dt
where v b (t ) = back emf = K ⋅ ω and K is the motor back emf constant which is dependent on the motor
parameters and the field applied to the field winding. However, for a constant field, K is a constant.
The torque developed by the motor is given by
Td = K ⋅ ia
Chapter 08.indd 434 3/14/2009 7:52:55 AM

Constant field
ia(t )
if
Ra La J
B (friction)
v(t ) vb(t )
Figure 8.9 Armature-controlled DC servomotor.
This developed torque has to work against the acceleration torque needed to overcome the inertia J and the
friction B. Further if the shaft of the motor is connected to any other system, an additional load TL is
reflected onto the shaft of the motor. The developed torque should work against this also. Therefore, the
developed torque and the angular velocity w are related as
dω
Td = K ⋅ ia = J + Bω + TL (8.26)
dt
Equations (8.25) and (8.26) are the two first-order linear differential equations that define the behavior of
the constant field DC motor system. Taking ia and w as the state variables, the state equation is obtained by
re-writing Eqs. (8.25) and (8.26) as
⎡dia / dt ⎤ ⎡− R / La − K / La ⎤ ⎡ia ⎤ ⎡1 / La 0 ⎤ ⎡v ⎤
⎢ ⎥=⎢ ⎥⋅⎢ ⎥ + ⎢ ⎥⋅⎢ ⎥ (8.27)
⎣dω / dt ⎦ ⎣ K / J −B / J ⎦ ⎣ω⎦ ⎣ 0 −1 / J ⎦ ⎣TL ⎦
Equation (8.27) is the state equation for the DC motor system wherein v and TL are the external excitations
that are applied to the system.
To obtain the model in the transfer function form, one can use Eq. (8.27) or one can take the Laplace
transform of Eqs. (8.25) and (8.26). This gives
V ( s ) = Ra I a ( s ) + La sI a ( s ) + Kω ( s ) (8.28)
KI a ( s ) = Jsω ( s ) + Bω ( s ) + TL ( s ) (8.29)
Transfer functions are defined for one specific output variable and one specific input excitation making all
other excitations as zero. Therefore to obtain the transfer function for w(s)/V(s), the load torque input is
taken as zero. Substituting Eq. (8.29) into Eq. (8.28) and simplifying one obtains
ω (s) K / La J
G(s) = = (8.30)
V ( s ) s 2 + [( B / J ) + ( Ra / La )]s + [( Ra B / La J ) + ( K 2 / La J )]
Equations (8.28) and (8.29) can be visualized in the form of a block diagram as shown in Figure 8.10. This
block diagram has been obtained using the Laplace transforms of Eqs. (8.28) and (8.29).
Chapter 08.indd 435 3/14/2009 7:52:55 AM

TL(s)
V(s) 1 Ia(s) Td(s) − 1 w(s)

+ K +
− Ra + sLa B + sJ
Vb(s)
Figure 8.10 Block diagram showing the various relationships in the armature-controlled DC motor.
This simple example illustrates the use of the block diagram representation. It can be simplified through
certain rules of block diagram algebra. These will be discussed in the following section. It must be empha-
sized that the model obtained for the DC motor is based on several simplifying assumptions. The armature
reaction in the motor and the voltage drops in the brushes have been neglected. In addition, it is assumed
that the frictional torque is linear and directly proportional to the angular velocity. This is true only over
some operating range.
Poles and Zeros

The transfer functions are represented in the form of a numerator polynomial and a denominator polyno-
mial in terms of the Laplace variable. In general, the transfer function for any system is of the form
m −1
Y ( s ) bm s + bm −1s + + b1s + b0
m
= −
(8.31a)
U ( s ) an s + an −1s + + a1s + a0
n n 1
The order of the denominator polynomial “n” indicates the order of the system. It is in general greater than
or equal to the order of the numerator polynomial “m” for a causal system. However, in the case of predic-
tive systems wherein there is a differentiator element, “m” will be greater than “n”. In Eq. (8.31a), the
numerator and denominator polynomials can be represented in the factored form as
Y ( s ) ( s + n1 ) ⋅ ( s + n2 )( s + nm )
= (8.31b)
U ( s ) ( s + d1 ) ⋅ ( s + d 2 )(( s + d n )
From Eqs. (8.31a) and (8.31b), it is evident that there are “m” factors in the numerator and “n” factors in
the denominator. The Laplace variable “s” can take on values that are both real and complex. At the values
of s = −n1, −n2, …, –nm, the transfer function is zero. The values −n1, −n2, …, −nm are called the zeros of the
system. Likewise, when the values of s = −d1, −d2, …, −dn, the transfer function is infinity. The values −d1,
−d2, …, −dn are called the poles of the system.
8.5 Block Diagrams
I n the analysis of control systems it is very convenient to obtain the block diagrams of different components
and their interconnections. If the various components are non-interacting (i.e., there is no “loading” effect
of one component on another), it is possible to obtain the overall transfer function of the system through a
Chapter 08.indd 436 3/14/2009 7:52:55 AM

suitable combination of the transfer functions of the component blocks utilizing some basic rules of block
diagram transformations to reduce the original diagram. It should be understood that absence of loading
effect means that each block of the block diagram has infinite input impedance and zero output impedance.
This is a rather stringent constraint in block diagram algebra which is not true in practical systems. Most
practical systems have finite input and output impedances and therefore the transfer functions resulting
from block diagram simplifications must be used with care and awareness.
For example, consider the cascade blocks shown in Figure 8.11(a). Here,
X2 = G1X1 (8.32)
Y = G2X2 (8.33)
Substituting Eq. (8.32) in Eq. (8.33) one obtains
Y = G1G2X1 (8.34)
This is shown in the equivalent block diagram alongside in Figure 8.11(a).
(i) (ii)
X1
X G1 G2 Y X G1 G2 Y
(a)
(i) (ii)
+
E G
X G Y X Y
− 1 + GH
(b)
(i) (ii)
G1 X G 1 + G2 Y
+
X Y
+
G2
(c)
Chapter 08.indd 437 3/14/2009 7:52:55 AM

(i) (ii)
G X G Y
Y
1/G X
X
(d)
(i) (ii)
+
X1 G Y
X1 + +
G Y
+
X2 X2 G
(e)
(i)
+
X G1 G3 Y
+
X2 G2 X1
(ii)
+
X G1 G3 Y
+
+
X1
+
X2 G2
(f)
Chapter 08.indd 438 3/14/2009 7:52:56 AM

(i) +
X G1 Y
+
X1
X2
G2
(ii)
+
X G1 Y
+
+
X2 X1
G2
−
G2
(g)
Figure 8.11 Block diagram reduction rules: (a) Combining blocks in cascade; (b) elimination of
feedback loop; (c) combining blocks in parallel; (d) moving a pick-off point behind a
block; (e) moving a summing point behind a block; (f) moving a pick-off point ahead of
a summing point; (g) moving a pick-off point behind a summing point.
Similarly, for the feedback loop in Figure 8.11(b), one obtains

Y = GE (8.35)
and
E = X – HY (8.36)
Eliminating E from the two equations by substituting Eq. (8.36) into Eq. (8.35) and after simplification,
one obtains
G
Y = X (8.37)
1 + GH
The proofs for the remaining transformations can be obtained easily in a similar manner and hence are left
to the reader as an exercise. One may apply the block diagram results given in Eqs. (8.34) and (8.37) for the
DC motor block diagram of Figure 8.10 to obtain the DC motor transfer function as given in Eq. (8.30).
As an example, consider the block diagram shown in Figure 8.12(a). Based on the reduction rules as
given in Figure 8.11, one can try to simplify the block diagram. In this case, the first step is to replace the
innermost loop by its equivalent transfer function. The resulting block diagram is shown in Figure 8.12(b).
Proceeding in the same manner, one can again remove the next inner loop after combining the two blocks
in cascade. The resulting block diagram is shown in Figure 8.12(c).
Chapter 08.indd 439 3/14/2009 7:52:56 AM

+ + +
R(s) G1 G2 G3 G4 C(s)
− − −
H3
H2
H1
(a)
+ + G3
R(s) G1 G2 G4 C(s)
1 + G3H3
− −
H2
H1
(b)
+ G2 G3
R(s) G1 G4 C(s)
1 + G3H3 + G2G3H2
−
H1
(c)
Figure 8.12 (a) An example block diagram; (b) simplified block diagram; (c) further simplification
of the block diagram.
Finally, reducing the last feedback loop, one gets the overall transfer function as
C (s ) (G1G2G3G4 ) / (1 + G3 H 3 + G2G3 H 2 )
=
R ( s ) 1 + [(G1G2G3 ) / (1 + G3 H 3 + G2G3 H 2 )]H1
G1G2G3G4 (8.38)
=
1 + G3 H 3 + G2G3 H 2 + G1G2G3 H1
Mason’s Rule
The overall transfer function can be obtained directly from the block diagram by using Manson’s rule, given
as follows:
C (s )
∑Tk ( s )Δk ( s )
G(s ) = = k (8.39)
R(s ) Δ( s )
Chapter 08.indd 440 3/14/2009 7:52:57 AM

where Tk(s) is the transfer function of the kth forward path from the input to the output; Δ(s) is the determi-
nant of the block diagram; Δk(s) all terms in Δ(s) that do not have elements or paths common with an ele-
ment or path in Tk(s). Mathematically Δ(s) is represented as
Δ s = 1 − sum of all individual loop transfer functions
+ sum of the products of the transfer functions of all possible setts of two non-touching loops
− sum of the productsof the transfer functions of all possible sets of three non-touchin
ng loops
+&
Note that two loops are said to be non-touching if they do not have a common branch or node. Also, a loop
is a closed path in the direction of the arrows that does not retrace itself.
As an example, consider the block diagram shown in Figure 8.12(a). Here, there is only one forward
path from R(s) to C(s), so that k = 1, and three loops with transmittances −G3H3, −G2G3H2 and −G1G2G3H1.
Also, all the loops are touching each other. Thus
T1 = G1G2G3G4
D = 1 − (−G3H3 − G2G3H2 − G1G2G3H1)
= 1 − (−G3H3 − G2G3H2 − G1G2G3H1)
Hence
T1 G1G2G3G4
G (s ) = =
Δ 1 + G3 H 3 + G2G3 H 2 + G1G2G3 H1
which is as given in Eq. (8.38).
EXAMPLE 8.2 Consider the block diagram shown in Figure 8.13. The transfer function C(s)/R(s)
can be determined by block diagram simplification. This is done most conveniently
by replacing the inner loop by its equivalent and then adding the two parallel
branches to obtain a single loop diagram. One can apply Mason’s rule and the result
can be verified by the reader by using the procedure suggested above.
G4
+
+ +
R(s) G1 G2 G3 C(s)
+
− −
H2
H1
Figure 8.13 An example of block diagram.
Chapter 08.indd 441 3/14/2009 7:52:57 AM

Here, there are two forward paths from R(s) to C(s) and
T1(s) = G1G2G3
T2(s) = G4G3
Further, there are three loops with transmittances −G2H2, −G1G2G3H1 and
−G4G3H1. Furthermore, the first and the last loops are non-touching. This gives
Δ(s) = 1 + (G2H2 + G1G2G3H1 + G4G3H1)
Δ1(s) = 1
Δ2(s) = 1 + G2H2
Hence
C (s ) G1G2G3 + G4G3 (1 + G2 H 2 )
=
R ( s ) 1 + G2 H 2 + G1G2G3 H1 + G4G3 H1 + G2 H 2G4G3 H1
8.6 Lagrange Method
T he 18th-century French mathematician Lagrange developed a method to obtain the equations govern-
ing the motion of complex mechanical systems. The differential equations resulting from this method
are known as Lagrange’s equations. They are basically derived from Newton’s laws of motion. Even though
the method was initially applied to mechanical systems, it is generic enough to be applied to other energy
domains as it deal with scalar quantities like potential and kinetic energy rather than vector quantities like
forces and torques.
The fundamental concept of Lagrange method is the representation of the system by a set of generalized
co-ordinates, one for each independent degree of freedom of the system. After having defined the general-
ized co-ordinates, the kinetic energy EK is expressed in terms of these co-ordinates and their derivatives. The
potential energy EP is expressed in terms of the generalized co-ordinates. If the generalized co-ordinates are
ei where i = 1,2, …, r, then the Lagrangian function L is given by
L = E K (e1 ,…, er , e1 ,…, er ) − E P (e1 ,…, er ) (8.40)
Finally, the desired equations of motion called the Lagrange’s equation are derived using the Lagrangian
function as
d ⎛ ∂L ⎞ ∂L
⎜ ⎟− = Fi i = 1,2, …, r (8.41)
dt ⎝ ∂e ⎠ ∂ei
where Fi’s denote the generalized forces that are external to the system. They are obtained from the algebraic
sum of the external forces and non-energy-storing/dissipative forces acting on the ith co-ordinate.
In the case of mechanical systems, the generalized co-ordinates are the independent linear or the angular
displacements of the different masses of the systems. The constitutive relations are used to express all the forces
on the elements in terms of the displacement and its derivatives. Though the Lagrange method was developed
and used for mechanical systems to model the dynamics of motion of bodies, the same concept can be extended
to electrical networks to model the dynamics of motion of charges. In electrical networks, the charges can be
considered to form the generalized co-ordinates. A systematic procedure to select the generalized co-ordinates
for electrical systems is to take the independent loop currents to be the time derivatives of the generalized
Chapter 08.indd 442 3/14/2009 7:52:58 AM

charge co-ordinates. Using the constitutive relations for the various elements, all the voltages across all circuit
elements are expressed in terms of the charges and their derivatives. The Lagrange method can be understood
by the following two examples, one for the mechanical system and the other for the electrical network.
The kinetic energy EK of the system is the sum of the kinetic energy of each mass. The cart is confined
to move only in the horizontal direction and therefore its kinetic energy is given as
1
E K1 = Mx 2
2
The pendulum rod rotates about the hinge. Therefore, the pendulum mass has one component of motion
along the horizontal and another along the vertical. The kinetic energy of the pendulum mass is given by
1
E K2 = m( x22 + h2 )
2
EXAMPLE 8.3 Inverted Pendulum

The cart with an inverted pendulum, shown in Figure 8.14, is excited by a controlled
input force F. Assume that the pendulum does not move more than a few degrees
away from the vertical. The problem is to determine the dynamic equations of
motion for this system in the state space form.
It is seen that the motion of the system is defined by the displacement x of the cart
from a reference point and the angle q that the pendulum rod makes with respect to
the vertical. Therefore, the system has only two degrees of freedom and the dynam-
ics are expressed in terms of the corresponding generalized co-ordinates which in
this case is (x, q ).
x2
q
Length = l h
Mass = m
x
M
F
0 x co-ordinate
Figure 8.14 Inverted pendulum on a moving cart. Here M is the mass of the
cart; m the mass of the pendulum; l the length of the pendulum;
F the force applied to the cart; x the cart position; q the pendu-
lum angle from the vertical.
Chapter 08.indd 443 3/14/2009 7:52:58 AM

where
x 2 = x + l sin θ
x = x + l θ cos θ
2
h = l cos θ
h = −l θ sin θ
Now the total kinetic energy of the system is given by
1 1
E K = E K1 + E K2 = Mx 2 + m( x 2 + 2 xθl cos θ + l 2θ 2 )
2 2
In the vertical direction, if the cart is considered as the reference, then the potential energy is that stored in
the pendulum. It is given as
E P = mgh = mgl cos θ
Now the Lagrangian function is given as
1 1
L = EK − EP = Mx 2 + m( x 2 + 2 xθl cos θ + l 2θ 2 ) − mgl cos θ
2 2
The generalized co-ordinates are (x, q). The Lagrange’s equations for this system are
d ⎛ ∂L ⎞ ∂L
⎜ ⎟− =F (8.42)
dt ⎝ ∂x ⎠ ∂x
d ⎛ ∂L ⎞ ∂L
⎜ ⎟− =0 (8.43)
dt ⎝ ∂θ ⎠ ∂θ
Equations (8.42) and (8.43) are the Lagrange’s equations for this system. The partial derivatives of the
Lagrangian with respect to the generalized co-ordinates are given as
∂L
= ( M + m )x + ml θ cos θ
∂x
∂L
=0
∂x
∂L
= mlx cos θ + ml 2θ
∂θ
∂L
= mgl sin θ
∂θ
Substituting the above partial derivatives in the Lagrange’s equations [Eqs. (8.42) and (8.43)], one obtains
( M + m )x + ml θ cos θ − ml θ 2 sin θ = F

(8.44)
mlx cos θ − mlxθ sin θ + ml 2θ − mgl sin θ = 0
Equation (8.44) gives the exact equations of motion of the inverted pendulum mounted on a moving cart.
Evidently, the equations are non-linear differential equations due to the presence of sin θ , cos θ , θ 2, xθ
Chapter 08.indd 444 3/14/2009 7:52:58 AM

terms. Under equilibrium condition, the pendulum rod is in the vertical position. If one considers that the
deviation in q is kept small, then in the neighborhood about the vertical operating point, the system may be
considered as linear. Therefore, in the very small neighborhood about the vertical operating point,
cos θ ≈ 1
sin θ ≈ θ
It may be assumed that the derivatives x and θ are kept small and therefore the quadratic terms θ 2 and
xθ will also be negligible. Applying these linearizing assumptions to the non-linear equations of motion
given in Eq. (8.44), one obtains the linearized dynamic model given as
( M + m )x + ml θ = F
(8.45)
mx + ml θ − mg θ = 0
There are two second-order linear differential equations. This means that the overall order of the system is
four. This implies that there must be four state variables to represent the state space of the system. The state
vector containing the four state variables is
⎡x⎤
⎢θ ⎥
x=⎢ ⎥
⎢ x ⎥
⎢ ⎥
⎣θ ⎦
The four first-order linear differential equations representing the dynamic behavior of the system in terms of
the state variables and the external excitation are
dx
= x
dt
dθ
=θ
dt
From Eq. (8.45), the other two differential equations are
dx −mg 1
= θ+ F (8.46)
dt M M
dθ M + m −1
= gθ + F (8.47)
dt Ml Ml
From the four first-order linear differential equations listed above, the state space representation of the
system in the standard form is given as
⎡ x ⎤ ⎡0 0 1 0⎤ ⎡ x ⎤ ⎡ 0 ⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢θ ⎥ = ⎢0 0 0 1⎥ ⎢θ ⎥ ⎢ 0 ⎥
⋅ + ⋅F (8.48)
⎢ x⎥ ⎢0 −mg / M 0 0 ⎥ ⎢ x ⎥ ⎢ 1 / M ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
θ
⎢⎣ ⎥⎦ ⎢⎣ 0 M
(M + m ) g / Ml 0 0 ⎥⎦ ⎢⎣θ ⎥⎦ ⎢⎣−1 / Ml ⎥⎦
Chapter 08.indd 445 3/14/2009 7:52:58 AM

⎡x ⎤
⎢ ⎥
⎡
θ=⎣ ⎤ ⎢θ ⎥
0 1 0 0 ⎦ ⋅ ⎢ x ⎥ (8.49)
⎢ ⎥
⎢⎣θ ⎥⎦
Equations (8.48) and (8.49) give the state and the output equations for the inverted pendulum system. To
obtain the model in the transfer function form, one can use Eq. (8.20) or take the Laplace transforms of
Eq. (8.45). On simplification and accounting for the two integrator for the state variables x and q, one
obtains the fourth-order transfer function as
θ( s ) −1 / Ml
G(s ) = = 2 (8.50)
F ( s ) s − [( M + m ) g / Ml ]
X (s ) 1 s 2 − ( Mg / Ml )
G (s ) = = (8.51)
F ( s ) Ms 2 s 2 − [( M + m ) g / Ml ]
EXAMPLE 8.4 Electrical Network

Consider the electrical network shown in Figure 8.15. Each independent loop cur-
rent will be chosen as independent charge variables q1 and q2. These two charge
variables form the generalized co-ordinates for the Lagrange method. The voltages
across all the elements of the circuit are expressible in terms of the charge variables
and their derivatives. Thus,
vR = R ⋅ ( q1 − q2 ) is the voltage across the resistor R
vL1 = L1 ⋅ q1 is the voltage across the inductor L1
vL2 = L2 ⋅ q2 is the voltage across the inductor L2
vC1 = q1 / C1 is the voltage across the capacitor C1
vC2 = q 2 / C 2 is the voltage across the capacitor C2
L1 C1 C2
Vi R L2
dq1 dq2
i1 = i2 =
dt dt
Figure 8.15 Example of electrical network.
The Lagrangian function L is given as

1 1 q2 q2
L = EK − EP = ⋅ L1 ⋅ q12 + ⋅ L2 ⋅ q22 − 1 − 2 (8.52)
2 2 2C1 2C 2
Chapter 08.indd 446 3/14/2009 7:52:58 AM

The Lagrange’s equations for this system with [q1, q2] as the generalized co-ordinates are
d ⎛ ∂L ⎞ ∂L
⎜ ⎟− = ∑ Vnon-storing-devices = Vi − R ⋅ ( q1 − q2 ) (8.53)
dt ⎝ ∂q1 ⎠ ∂q1 loop1
d ⎛ ∂L ⎞ ∂L
− = ∑ Vnon-storing-devices = R ⋅ (q1 − q2 ) (8.54)
dt ⎜⎝ ∂q2 ⎟⎠ ∂q2 loop2
Equations (8.53) and (8.54) are the Lagrange’s equations for this system. The partial derivatives of the
Lagrangian with respect to the generalized co-ordinates are given as
∂L
= L1 ⋅ q1
∂q1
∂L −q1
=
∂q1 C1
∂L
= L2 ⋅ q2
∂q2
∂L −q 2
=
∂q 2 C 2
Substituting these partial derivatives in the Lagrange’s equations [Eqs. (8.53) and (8.54), one obtains
q
L1 ⋅ q1 + 1 = Vi − R ⋅ (q1 − q2 ) (8.55)
C1
q2
L2 ⋅ q2 + = R ⋅ (q1 − q2 ) (8.56)
C2
Equations (8.55) and (8.56) give the dynamic model of the electrical network of Figure 8.15. These two
equations can be re-arranged and split into four linear first-order differential equations using q1 , q1 , q2
and q2 as state variables to obtain the standard state equation representation.
8.7 Circuit Averaging

T he Lagrange method uses scalar variables like the kinetic energy and the potential energy of the components
in the system. This does not directly give information on the power flow direction. Another problem with
the Lagrange method is that it cannot easily model switched power systems and the dynamics related to such
systems like the switched-mode DC–DC converters. Circuit averaging method is a useful modeling approach
especially for switched power electric circuits. The circuit averaging method is explained in the following steps.
Step 1 (Large-Signal Model): Identify the active circuit for the different position of the switch. From
the equivalent active circuit for each switch position, write down the dynamic circuit equations using
the Kirchhoff ’s voltage and current laws. The state variables are the inductor currents and capacitor
voltages. The state equations of the active circuit for the various switch positions are written down. For
a converter circuit using a single-pole double-throw switch, there are two switch positions and therefore,
two operative circuit modes. The two state equations are of the form
Chapter 08.indd 447 3/14/2009 7:52:59 AM

x = A1 x + B1u (8.57)
y = C1 x + D1u
x = A2 x + B2u (8.58)
y = C 2 x + D2u
Equations (8.57) and (8.58) are the state equations for the two operative circuit modes of a converter
with a two-state switch.
Step 2 (Average Large-Signal Model): The state equations represent the dynamic model of the active
circuit resulting from the specific switch position for a specified interval of time. The above representation
is in the standard state space format for each of the intervals. If it is a two-state switch then there are two
sets of state equations. The converter alternates between the two switched states at high frequency. It is
required to represent the converter through a single equivalent dynamic representation valid for both
states of the switch.
Consider a converter system with two switch states. The pole of the switch is at throw 1 for a
period of time dTs and at throw 2 for a period of time (1 − d )Ts. If one considers the variation of the
state variables over a switching period, then
x = xavgTs = xdT ⋅ dTs + x(1− d )T ⋅ (1 − d )Ts

S S
where xavg is the average rate of change of state variables over a switching period. The above descrip-
tion is valid if xdT and x(1− d )T are constant during the dTs and (1 − d )Ts duration, respectively.
S S
This will be a valid assumption if the switching period is small compared to the natural time constants
of the respective circuits. Then the averaged state variables are obtained from the state equations
[Eqs. (8.57) and (8.58)] as
x = Ax + Bu
(8.59)
y = Cx + Du
where
A = A1d + A2 (1 − d )
B = B1d + B2 (1 − d )
C = C1d + C 2 (1 − d )
D = D1d + D2 (1 − d )
Equation (8.59) represents the equivalent state equation of the converter. Since the averaging process
has been done over a switching period, the equivalent model is valid for time durations much larger
compared to the switching period.
Step 3 (Steady-State Model): The steady-state solution is obtained by equating the rate of change of the
state variables to zero. Under steady-state or equilibrium conditions,
Chapter 08.indd 448 3/14/2009 7:52:59 AM

x = 0
x=X
(8.60)
u =U
y =Y
For the steady-state conditions, Eq. (8.59) can be written as
0 = AX + BU
(8.61)
Y = CX + DU
Step 4 (Small Signal Model): The averaged large-signal model of the converter given by Eq. (8.59) is
linear but not time invariant. This is because the characteristic matrix A and the input matrix B contain
the duty ratio variable d that is time-varying. Therefore it is necessary to model the system in the
neighborhood of the operating point to analyze the system and synthesize controllers for the system.
Such a model is called the small-signal model. From Eq. (8.59), the averaged dynamic equations are
x = [ A1d + A2 (1 − d )]x + [ B1d + B2 (1 − d )]u
(8.62)
y = [C1d + C 2 (1−
− d )]x + [ D1d + D2 (1 − d )]u
All the variables are considered to have small variations in the neighborhood of the steady-state or
equilibrium state operating point. Thus,
dˆ
d = D + dˆ; 1
D
uˆ
u = U + uˆ; 1
U
The small-signal variations in duty ratio d and input u result in perturbations in x and y about the
operating points. Thus,
xˆ
x = X + xˆ; 1
X
yˆ
y = Y + yˆ ; 1
Y
Therefore Eq. (8.62) can be written as
( X + xˆ ) = [ A1( D + dˆ ) + A2 (1 − D − dˆ )]( X + xˆ ) + [ B1( D + dˆ ) + B2 (1 − D − dˆ )](U + uˆ )
(8.63)
(Y + yˆ ) = [C ( D + dˆ ) + C (1−
1 2 − D − dˆ )]( X + xˆ ) + [ D ( D + dˆ ) + D (1 − D − dˆ )](U + uˆ )
1 2
The above equations can be expanded and separated into steady-state terms, linear small-signal terms
and non-linear terms. The non-linear terms contain the second- and higher order perturbation products
which can be neglected.
Terms containing xˆ ⋅ dˆ and dˆ ⋅ uˆ can be neglected and the steady-state terms
AX + BU = 0
Thus, the linear small-signal terms are
xˆ = Axˆ + Buˆ + [( A1 − A2 ) X + ( B1 − B2 )U ]dˆ
yˆ = Cxˆ + Duˆ + [(C − C ) X + ( D − D )U ]dˆ
1 2 1 2
Chapter 08.indd 449 3/14/2009 7:52:59 AM

The perturbation in duty ratio dˆ is now considered as an input for the small-signal model. Thus, the resul-
tant small-signal model is given as
xˆ = Axˆ + Buˆn
(8.63)
yˆ = Cxˆ + Duˆn
where
A = A1D + A2 (1 − D ) (8.63a)
B = ⎡⎣B1D + B2 (1 − D ) ( A1 − A2 ) X + ( B1 − B2 )U ⎤⎦ (8.63b)
⎡ uˆ ⎤
uˆn = ⎢ ⎥ (8.63c)
⎢⎣dˆ ⎥⎦
C = C1D + C 2 (1 − D ) (8.63d)
D = ⎡⎣ D1D + D2 (1 − D ) (C1 − C 2 ) X + ( D1 − D2 )U ⎤⎦ (8.63e)
X = − A −1BU (8.63f )
EXAMPLE 8.5 Buck Converter

Consider the buck converter as discussed in Chapter 5. The active circuits for the
two states of the single-pole double-throw (SPDT) switch are shown in Figure 8.16.
The operation and steady-state analysis of the converter is as discussed in Chapter 5.
Referring to Figure 8.16, as there are two energy-storage devices, there will be need
for two state variables. The dynamic inductor element stores energy by virtue of the
flow of current through it and therefore, the inductor current iL can be considered
as a state variable. The dynamic capacitor element stores energy by virtue of the
voltage across it and therefore, the capacitor voltage vC can also be considered as a
state variable. Thus the state vector is
⎡i ⎤
x=⎢ L ⎥ (8.64)
⎣v C ⎦
dT s period
T1 T1 (1 − d)Ts period
S1 iL S1
io iL io
Vi P Vi P
L ic ic
Vo S2 Vo
S2 C C
Ro Ro
T2 T2
(a) (b)
Figure 8.16 The operative circuit during: (a) Period dTs; (b) period (1 − d)Ts.
Chapter 08.indd 450 3/14/2009 7:52:59 AM

Step 1 (Large-Signal Model):

During dTs Period: Referring to Figure 8.16(a), the state equations during this
period are obtained by applying the KVL for the voltage across the inductor and
the KCL for the current through the capacitor. Thus,
diL v i − vC
=
dt L (8.65)
dvC iL − io iL − vC / Ro
= =
dt C C
The state equation is
⎡ diL ⎤ ⎡ −1 ⎤
⎥ ⎢0 ⎡1⎤
⎢
⎢ dt ⎥ = ⎢ L ⎥⎥ ⎡ iL ⎤ ⎢ ⎥
⋅ + L ⋅ ⎡v ⎤
−1 ⎥ ⎢⎣vC ⎥⎦ ⎢ ⎥ ⎣ i ⎦
(8.66a)
⎢ dvC ⎥ ⎢ 1
⎥ ⎢ ⎢⎣ 0 ⎥⎦
⎢
⎣ dt ⎦ ⎣C RoC ⎥⎦
and the output equation is
⎡i ⎤
vo = ⎡⎣0 1⎤⎦ ⋅ ⎢ L ⎥ + ⎡⎣0 ⎤⎦ ⋅ ⎡⎣v i ⎤⎦ (8.66b)
⎣v C ⎦
Equations (8.66a) and (8.66b) are of the form
x = A1 ⋅ x + B1 ⋅ u
y = C1 x + D1u
During (1 − d )Ts Period: Referring to Figure 8.16(b), the state equations during
this period are obtained by applying the KVL for the voltage across the inductor
and the KCL for the current through the capacitor. Thus,
diL 0 − vC
=
dt L (8.67)
dvC iL − io iL − vC / Ro
= =
dt C C
The state equation is
⎡ diL ⎤ ⎡ −1 ⎤
⎢ ⎥ ⎢0 L ⎥⎥ ⎡ iL ⎤ ⎡0 ⎤
⎢ dt ⎥ = ⎢ ⋅ + ⋅ ⎡v ⎤
−1 ⎥ ⎢⎣vC ⎥⎦ ⎢⎣0 ⎥⎦ ⎣ i ⎦
(8.68a)
⎢ dvC ⎥ ⎢ 1
⎢ ⎥ ⎢ RoC ⎥⎦
⎣ dt ⎦ ⎣C
and the output equation is
⎡i ⎤
vo = ⎡⎣0 1⎤⎦ ⋅ ⎢ L ⎥ + ⎡⎣0 ⎤⎦ ⋅ ⎡⎣v i ⎤⎦ (8.68b)
⎣v C ⎦
Equations (8.68a) and (8.68b) are of the form
x = A2 ⋅ x + B2 ⋅ u
y = C 2 x + D2u
Chapter 08.indd 451 3/14/2009 7:53:00 AM

Step 2 (Averaged Large-Signal Model): The averaged large-signal model is

obtained from Eq. (8.59). T he state equations [Eqs. (8.66) and (8.68)] for the two
operative intervals are combined to obtain the averaged large-signal model. Thus,
⎡ diL ⎤ ⎡ −1 ⎤
⎥ ⎢0 ⎡d ⎤
⎢
⎢ dt ⎥ = ⎢ L ⎥⎥ ⎡ iL ⎤ ⎢ ⎥
⋅ ⎢ ⎥ + L ⋅ ⎡⎣v i ⎤⎦ (8.69a)
⎢ dvC ⎥ ⎢ 1 −1 ⎥ ⎣vC ⎦ ⎢ ⎥
⎥ ⎢ ⎢⎣ 0 ⎥⎦
⎢
⎣ dt ⎦ ⎣C RoC ⎥⎦
⎡i ⎤
vo = ⎡⎣0 1⎤⎦ ⋅ ⎢ L ⎥ (8.69b)
⎣v C ⎦
Equations (8.69a) and (8.69b) give the state equation and the output equation of
the averaged large-signal model.
Step 3 (Steady-State Model): The steady-state model is obtained based on Eqs.
(8.60) and (8.61). Thus the steady-state model is given as
⎡ −1 ⎤
0 ⎡D ⎤
⎡0 ⎤ ⎢ L ⎥⎥ ⎡ I L ⎤ ⎢ ⎥
= ⎢ ⋅ + L ⋅ ⎡V ⎤
−1 ⎥ ⎢⎣VC ⎥⎦ ⎢ ⎥ ⎣ i ⎦
⎢ ⎥ (8.70a)
⎣0 ⎦ ⎢ 1 ⎢⎣ 0 ⎥⎦
⎢C R C ⎥
⎣ o ⎦
⎡I ⎤
Vo = ⎡⎣0 1⎤⎦ ⋅ ⎢ L ⎥ (8.70b)
⎣VC ⎦
Equations (8.70a) and (8.70b) are the state and output equations of the steady-
state model. The steady-state input–output relationship that is discussed in Chapter
5 is obtainable from the above equations. From Eq. (8.70) it can be seen that
Vo = DVi
Step 4 (Small Signal Model): The small-signal model is obtained from Eqs.
(8.63a)–(8.63f ). The small-signal model is given as
xˆ = Axˆ + Buˆn
(8.71)
yˆ = Cxˆ + Duˆn
where
⎡ −1 ⎤
⎢0 L ⎥⎥
A=⎢
⎢1 −1 ⎥
⎢C RoC ⎥⎦
⎣
⎡D D ⎤
V
B=⎢L L i ⎥⎥
⎢
⎢⎣ 0 0 ⎥⎦
⎡vˆ ⎤
uˆn = ⎢ i ⎥
⎢⎣ dˆ ⎥⎦
Chapter 08.indd 452 3/14/2009 7:53:00 AM

C = ⎡⎣0 1⎤⎦
D = ⎡⎣0 0 ⎤⎦
Equation (8.71) is the small-signal model of the buck converter where dˆ is another
input to the system. Normally, dˆ is the control input to the buck converter. The trans-
fer function Vo(s)/d(s) can be obtained from the small-signal model using Eq. (8.20).
8.8 Bond Graphs

B ond graph is a graphical tool for modeling physical systems. In 1959, Prof. H. M. Paynter gave the
revolutionary idea of portraying systems in terms of power bonds, connecting the elements of the
physical system to the so-called junction structures. The power exchange picture of a system is called
bond graph/energy bond graph/power bond graph that can be both power- and information-oriented.
“Bond graphs” is a system modeling technique. It has been used extensively in modeling interconnected
interacting physical systems. Bond graphs not only allow the modeling of systems for analysis and simula-
tion, but they are also a powerful tool for automatic computer programming. Bond graphs were invented
by Henry Paynter due to the inherent drawbacks of block diagrams for servocontrols and simulation
problems. Bond graphs consider both energy and signal exchanges between components of a system. For
systems where power and efficiency play a major role, bond graph modeling method is very convenient.
By this approach, a physical system can be represented by symbols and lines, identifying the power flow
paths. The lumped parameter elements of resistance, capacitance and inductance are interconnected in an
energy conserving way by bonds and junctions resulting in a network structure. From the pictorial represen-
tation of the bond graph, the derivation of system equations is so systematic that it can be algorithmized.
Bond graphs display both energy and signal exchanges between components or elements in installations
and systems by simple lines and symbols. They bridge the gap between control engineering and the parts of
engineering science where power and efficiency have greater importance including energy conversion.
Bond graphs are a representation of components, machines and systems on paper using conventional
signs and symbols. Bond graphs are set up causally (without such a distinction between cause and effect) and
transformed subsequently into a causal diagram by a systematic choice of the causalities.
Block Diagram Drawbacks
1. A return action between components must always be shown by a separate feedback connection, while
in reality the components have only one connection.
2. Block diagrams for a given physical system can have a radically different appearance because of the diffe-
rent possible causalities.
3. The loading effect of subsequent blocks is not reflected backwards.
Equivalent Circuits Drawbacks
1. Due to presence of fictitious elements the equivalent circuit loses its similarity with the real layout.
2. With each symbol, it must be indicated separately, whether it is a real component or an idealized
element.
3. Some of the representation may be in a domain in which the actual system does not exist. For example,
the thermal model of the system being represented as an electrical equivalent.
Chapter 08.indd 453 3/14/2009 7:53:00 AM

Table 8.2 Effort and flow variables for systems in few energy domains
Systems Effort (e) Flow (f )
Force (F ) Velocity (v)

Mechanical
Torque (t) Angular velocity (w)
Electrical Voltage (V ) Current (i)
Hydraulic Pressure (P ) Volume flow rate (dQ/dt)
Thermal Temperature (T ) Entropy change rate (dS/dt)
Chemical potential ( m) Mole flow rate (dN/dt)

Chemical
Enthalpy (h) Mass flow rate (dm/dt)
Magnetic Magneto-motive force (em) Rate of flux change (dF/dt)
Bond graphs avoid the above drawbacks and can represent interdisciplinary systems with one set of symbols
for all disciplines. Bond graphs are based on the splitting of systems into separate components that exchange
energy or power through identifiable connections or ports. They are called bonds in analogy to the energy
exchange between atoms in chemical bonds.
In this technique, power flow is represented by a half bond. Every bond is associated with two variables,
effort and flow and the causality indication. The various components in a system linked using the bonds
form the power bond graph. Any system can be modeled using the finite set of elements that are one-port,
two-port or multi-port depending on the number of ports for exchanging energy.
1. One-port: These include sources (effort and flow), passive elements (dissipation, kinetic and potential
storage).
2. Two-port: These include transformer, modulated transformers, gyrators and modulated gyrators.
3. Multi-port: These include 0-junction and 1-junction.
The energy flux or power in a bond is always the product of two variables – a potential variable called
“effort” and a “flow” variable or current variable, simply referred as flow. Table 8.2 gives the effort and flow
of various domains.
Standard Elements
The standard elements of bond graphs are classified according to the number of bonds as one-ports, two-ports
and multi-ports. This classification originates from electric circuit theory where each port or bond represents
two wire terminals or connections. A one-port has two connections, a two-port four connections, etc.
One-Ports
One-ports are elements exchanging energy with the system through one bond only. They include resistance
elements (called R-elements), inertia elements (called L-elements) and capacity elements (called C-elements).
In addition there are the voltage sources and the current sources.
Each element has two possible governing equations based on the causality. In the particular application,
the causality will indicate the governing equation for the bond. The causality is shown as a vertical bar either
at the beginning or at the end of the bond. Table 8.3 gives the one-port elements.
Chapter 08.indd 454 3/14/2009 7:53:00 AM

Table 8.3 One-ports
Bond graph element Governing equations Remarks
Sources
Se E Effort source
Sf F Flow source
mSe E Modulated effort source
mSf F Modulated flow source
Sinks – Dissipative and Storage Elements

e
R f = Dissipative effort causal
R
R e = f ⋅R
Dissipative flow causal
1
L f =
L ∫ e ⋅ dt
Kinetic energy storage with effort causal.
Note the integral cause. The effect which
df is flow is a STATE variable.
L e = L⋅
dt Flow causal. Differential causal.
de
C f =C ⋅
dt Effort causal. Differential causal.
1
C e=
C ∫ i ⋅ dt Potential energy storage with effort
causal. The effect which is effort is
a state variable.
Two-Ports
Two-ports have two bonds for energy exchange with the system. They are energy conserving in the sense that
the product of input flow and effort is at all times equal to the product of output flow and effort. Broadly
there are two types of two-ports based on the type of energy variable linking between the input and output
ports: (a) transformers and (b) gyrators.
In transformers, the cause-and-effect variables are both either efforts or flows. This means that there is
only input effort–output effort link or input flow–output flow linkage. Transformers, although they are not
normally distinguished in bond graph, are of the following two different kinds.
1. Impedance Transformers: Here the input and output variables are of the same energy domain or class.
They change the impedances, that is, the ratio of effort to flow in the bonds. Example: Electrical trans-
formers or gear reducers.
Chapter 08.indd 455 3/14/2009 7:53:00 AM

2. Class Transformers: Here input and output variables belong to different energy domains or classes or
disciplines. Example: Hydraulic cylinders and pumps connecting hydraulic with mechanical variables
because the impedance here belongs to different classes (have different physical dimensions), it is not
possible to compare them.
Class transformers are important because they allow an efficient representation of interdisciplinary engineering
systems by bond graphs. It should be mentioned again that both class and impedance transformers conserve
the power at all times.
In the case of gyrators, the effort of one-port is linked to the flow of the other port. Thus one has cross-
linkage of variables in the case of gyrators like input effort–output flow or input flow–output effort linkage.
Table 8.4 gives the list of two-ports elements.
Table 8.4 Two-ports
e1 m e2 e 2 = m ⋅ e1 Transformer effort causal. e1 is the cause

f1 TF f2 f1 = m ⋅ f 2 and f1 is the effect.
e1 m e2 1 Transformer flow causal. f1 is the cause

f2 = ⋅ f1 and e1 is the effect.
f1 TF f2 m
1
e1 = ⋅ e2
m
mTF Same as transformer with effort Modulated transformer effort causal

causal, but m is modulated (varying).
mTF Same as transformer with flow causal, Modulated transformer flow causal
but m is modulated (varying).
e1 m e2 1 Gyrator effort causal. e1 is the cause and

f1 GY f2
f2 = ⋅ e1 f1 is the effect.
m
1
f1 = ⋅ e2
m
e1 m e2 e 2 = m ⋅ f1 Gyrator flow causal. f1 is the cause and

f1 GY f2 e1 is the effect.
e1 = m ⋅ f 2
mGY Same as gyrator with effort causal, Modulated gyrator effort causal.
but m is modulated (varying).
mGY Same as gyrator with flow causal, but Modulated gyrator flow causal.
m is modulated (varying).
Chapter 08.indd 456 3/14/2009 7:53:03 AM

Multi-Ports (Junctions)
The elements of bond graphs are connected by either of the following two junctions:
1. The 0-junction where the efforts (voltages) on all bonds are same and the flows (currents) are additive
or, more precisely, add up to zero. This junction corresponds to the parallel circuit in electronics.
2. The 1-junction where the efforts (voltages) on all bonds add up to zero and the flows (currents) on all
bonds are same. It corresponds to the series circuit in electronics.
Junctions are power conserving at each instant and the power transports of all bonds add up to zero at all
times. Table 8.5 describes the 0-junction and the 1-junction.
Rules for the Selection of Causality

Causality is not a natural fact but selected by the engineer to facilitate computation. The choice is free in
principle but subject to the following limitations:
1. Sources always impose an effort or a flow to the system. Consequently the causality must be chosen first.
2. With one-port elements, the integral causal is the preferred causality by nature. These also decide the
state variables of the system. The dynamic elements are then chosen for integral causality.
3. The junction rules for 0-junction and 1-junction should then be applied.
4. The dissipative elements do not have any preferred causality. Causality of these bonds are chosen to
have consistent causality for the whole system.
Steps in Obtaining the System Model

System analysis using bond graph is very organized and systematic. Given any system the following simple
steps should be executed to arrive at the model of the physical system:
1. Construct the bond graph for the given system.
2. Assign causality using following order:
Table 8.5 Multi-ports

e1 e1 = e 2 = ... = en 0-junction: There can be only one bond with a

f1 causal bar at the 0-junction. This means that only
n
e2 en one bond can decide the junction effort. The algebraic
f2
0
fn
∑ fi = 0 sum of all the associated bond flows should be zero.
i =1
e3 This is called the FLOW LAW. Example: Parallel
f3 circuits in electrical domain (KCL).
e1 n 1-junction: There can be only one bond without a

f1 ∑ ei = 0 causal bar at the 1-junction. This means that only
e2 en i =1 one bond can decide the junction flow. The algebraic
1 sum of all the associated blond efforts should be
f2 fn f 1 = f 2 = ... = f n zero. This is called the EFFORT LAW. Example:
e3
f3 Series circuits in electrical domain (KVL).
Chapter 08.indd 457 3/14/2009 2:38:23 PM

• Sources.
• Dynamic elements (Integral equation is given preference over differential equation.).
• Junction rule to be applied.
• Dissipative element.
3. State Equation Extraction: The order of the system as well as the total number of equations needed to
describe the system depends upon the number of storage elements. Write down the state equations
considering the causality and the effort-flow relationship for the various elements.
Bond Graph Construction

The above steps in obtaining the state space model will be illustrated with a number of examples. However,
one must first learn to construct the bond graph for a given system. There are two ways in which the bond
graph may be constructed: (a) by inspection approach and (b) by algorithmic approach. The former is very
intuitive and lends itself very well to modeling complex physical systems depending on the amount of
knowledge and experience the modeler has with the particular system. The latter method is algorithmic and
lends itself to computer-aided model extraction. This section will discuss both the approaches.
Consider the following RLC circuit as shown in Figure 8.17. To obtain the state space model for the
RLC system, one has to first construct the bond graph. Both the methods of bond graph construction will
be illustrated for the circuit of Figure 8.17.
Inspection Approach
The RLC circuit of Figure 8.17 can be re-written as shown in Figure 8.18(a). The circuit consists of the voltage
source e1, the resistance R1, the inductance L and block B1. All these components share the same current;
therefore, the corresponding bond graph for these elements is as shown in Figure 8.18(a). The main property
of the 1-junction is that all bonds connected to it share a common flow. The flow in this case is the current.
Therefore, components having same current must be connected to a 1-junction. On studying the bond
graph of Figure 8.18(a), it is observed that there are four bonds at the 1-junction:
• An energy bond for the effort source Se1.
• An energy bond for energy flowing into the resistance R1 and dissipating.
• An energy bond for the energy flowing into the inductance L and stored.
• An energy bond for the energy flowing into the block B1.
• The arrows indicate the general direction of the flow of energy. It should be noted that the instantaneous
power can also flow in a direction opposite to that indicated in the bond arrows in which case the corres-
ponding power variable (either the voltage/effort or the current/flow) is negative. However, the bond
arrows should, in general, be placed to indicate the energy flow.
R1 L
e1 C R2
Figure 8.17 RLC circuit.
Chapter 08.indd 458 3/14/2009 7:53:05 AM

Block
R1 L B1
e1 C R2
Figure 8.18 (a) RLC circuit and bond graph; (b) block B1 and bond graph; (c) overall bond graph.
Figure 8.18(b) shows block B1 which has C and R2 in parallel. Both the components of block B1 share
a common voltage/effort. Therefore, one should use a 0-junction to connect the two components because in
the case of a 0-junction all the bonds connected to it share a common effort. Referring to the bond graph of
block B1 in Figure 8.18(b), the 0-junction consists of
1. the energy bond from the 1-junction that indicates the flow of energy to block B1.
2. an energy bond indicating the flow of energy into resistance R2.
3. another energy bond to indicate the flow of energy into capacitance C for storage.
Figure 8.18(c) shows the completed bond graph by concatenating the bond graphs of Figures 8.18(a) and
8.18(b). The bond graph of Figure 8.18(c) can also be arrived at by the algorithmic approach. This approach
will now be explained.
Algorithmic Approach
The bond graph may also be constructed in an algorithmic way that will prove useful for computer-aided
generation of bond graphs for physical systems. However, this approach lacks the intuition that one gains
from the method of inspection. The sequential steps involved in this method are as follows:
1. Replace every component by a 1-junction and connect the appropriate one-port bond to the junction.
2. Introduce 0-junctions for all nodal points in the circuit.
Chapter 08.indd 459 3/14/2009 7:53:05 AM

3. Combine all contiguous 1-junctions into a single 1-junction.

4. Combine all contiguous 0-junctions into a single 0-junction.
5. Remove any junction that has less than three bonds connected to it and again apply Steps (3) and (4) if
necessary.
6. Identify a ground node (any of the 0-junction nodes).
7. Remove all bonds connected to ground node 0-junction including the junction.
8. Apply Steps (3), (4) and/or (5) again.
Applying the above steps to the RLC circuit of Figure 8.17, the sequence of bond graph construction is
shown in Figure 8.19. Figure 8.19(a) shows the application of Step (1) to the RLC circuit. Here every com-
ponent is replaced with a 1-junction. A bond is used to connect the component to the 1-junction. The bond
direction is decided on the energy flow direction, that is, energy flows out of sources and into the sinks like
R, L and C. Figure 8.19(b) shows the application of Step (2). Here a 0-junction is introduced at every node
as shown in Figure 8.19(b). Figure 8.19(c) shows the application of Steps (3) and (4). Contiguous 1-junctions
are combined into a single 1-junction. Likewise contiguous 0-junctions are combined into a single
0-junction. As 1-junctions and 0-junctions are multi-ports, 1- and 0-junctions with less than three energy
bonds connected to them can be replaced with a single bond. This is illustrated in Figure 8.19(d). After this,
once again the process of reducing contiguous junctions is performed.
R1 L
R1 L
1 1
e1 R2 S e1 1 C 1 1 R2
C
(a) Step 1
R1 L
0 1 1 0 0
Se1 1 C 1 1 R2
0 0 0
(b) Step 2
Chapter 08.indd 460 3/14/2009 7:53:06 AM

R1 L R1 L
0 1 0 0 0 1 0
Se1 1 C 1 1 R2 Se1 1 C 1 1 R2
0 0 0 0
(c) Steps 3 and 4
R1 L R1 L
1 0 Se1 1 0
Se1 1 C 1 1 R2 C 1 1 R2
0 0
(d) Steps 5 and re-applying Step 3
R1 L R1 L
S e1 1 0 S e1 1 0
C 1 1 R2 C 1 1 R2
Choose this as All bonds connected to

ground node or reference junction are removed
reference 0-junction along with the reference junction
(e) Steps 6 and 7
R1 L
S e1 1 0
C R2
(f) Step 8
Figure 8.19 (a) Applying Step (1) to the RLC circuit; (b) applying Step (2); (c) bond reduction
by applying Steps (3) and (4); (d) applying Step (5); (e) applying Steps (6) and
(7); (f) applying Step (8).
Chapter 08.indd 461 3/14/2009 7:53:06 AM

Now a reference 0-junction has to be chosen. Any 0-junction can be chosen as the reference 0-junction.
In Figure 8.19(e), one such 0-junction is considered as the reference. After choosing the reference 0-junction,
remove that junction along with all the energy bonds connected to that 0-junction. Finally reducing the
bond graph by applying Steps (3), (4) and/or (5), the final bond graph is obtained which is identical to the
bond graph obtained by the inspection approach as given in Figure 8.18(c).
Causality Assignment
After the bond graph has been constructed, the next step is to assign causality to each bond. The assignment
of causality gives an insight into the internal working of the system at each bond level. As there are two
power variables associated with each bond, one will act as the cause variable and the other will be the effect
variable. The causality assignment will define the cause-and-effect relationship between the effort and the
flow variables across the entire system. The general rules for causality assignment have been mentioned pre-
viously. The causality assignment sequence is as follows:
1. Sources always impose an effort or a flow to the system. Therefore, the causality of the effort and/or the
flow sources must be chosen first. The causality is assigned as given in the section “Rules for Selection
of Causality”.
2. The causality of bonds connected to dynamic elements (L and C) is chosen next. Referring to the sec-
tion “Steps in Obtaining the System Model”, the causality assignment for the dynamic elements is
chosen such that the governing equation is an integration rather than differentiation. This is because
nature prefers continuum. This means that for an inductive element, the effort causal assignment is
preferred and for a capacitive element, the flow causal assignment is preferred.
3. Next, the 0-junction and the 1-junction rules are applied to propagate the causality to other bonds. It
should be noted that at the 0-junction only one bond can decide the effort at the junction, that is, only
one bond can have a causal bar at the junction end and all other bonds should have the causal bar away
from the junction end. In the case of the 1-junction, only one bond can decide the flow at the junction,
that is, only one bond can have a causal bar away from the junction end and all the other bonds should
have the causal bar towards the junction end.
4. The causality of the two ports is chosen in a manner such that there is no conflict with the above three rules.
5. The dissipative elements do not have any preferred causality. Causality of these bonds is chosen in the
end to have consistent causality for the whole system.
The bond graph which also shows the causality assignment is called the augmented bond graph. The above
causality assignment rules will be applied to the bond graph of the RLC circuit given in Figure 8.19(f ). The
causality assignment sequence is shown in Figure 8.20. Figure 8.20(a) begins the causality assignment
sequence by defining the source causality. The effort source causality is assigned such that it has the causal
bar at the arrow end.
Next the causality of the dynamic elements is considered. Without loss of generality one can start with
any of the dynamic elements. The causality for the L-bond is assigned such that the governing equation has
an integration. Thus the causal bar for the L-bond is at the arrow end which indicates effort causal, that is,
for the L-bond, the effort or voltage across the inductor is the independent variable or the cause variable and
the flow or the current through the inductor is the dependent variable or the effect variable. The flow effect
is related to the effort cause by means of an integral equation as given in Table 8.3. Likewise the C-bond is
also assigned the causality such that the governing equation has an integration. Referring to the discussion
on the one-ports and Table 8.3, the causal bar is at the non-arrow end. This means that the flow or the
current is the cause and the effort or voltage is the effect which is obtained as the output of the integral equation.
Figure 8.20(b) shows the causality assignment for the bonds connected to dynamic elements.
Chapter 08.indd 462 3/14/2009 7:53:07 AM

R1 L
Causality for
R L dynamic elements
Se1 1 0
S e1 1 0
Source C R2
causality
assignment C R2
(a) (b)
R L
R1 L
S e1 1 0
Se1 1 0
C R2
Apply junction rules
and assign causality to
C R2 other bonds
(d) (c)
Figure 8.20 (a) Assign causality to source bonds; (b) assign causality to dynamic element bonds;
(c) assign causality to other bonds based on junction rules; (d) final augmented bond
graph of RLC circuit.
This is followed by assigning causality to bonds connected to the 1- and 0-junction by applying the junc-
tion rules. At the 1-junction all the bonds except one should have the causal bar at the junction end of the
bond. As the L-bond already has the causal bar at the arrow end, all the other bonds connected to the 1-junction
should have the causal bar at the junction end. For the case of the 0-junction only one bond should have a
causal bar at the junction end. As the C-bond already has the causal bar at the junction end, all other bonds
connected to the 0-junction should have the bonds away from the 0-junction end. Without loss of generality,
one can assign causality to the bonds using the junction rules to as many bonds as possible such that there are
no causal conflicts. Figure 8.20(c) shows the causality assignment by applying the junction rules.
In the end, only the bonds connected to dissipative elements like resistances will be left. The dissipative
elements do not have any preferred causality as their governing equations are only algebraic. Thus the
remaining bonds that have not been assigned any causality will be those connected to dissipative elements.
Here any causality can be assigned such that the junction rule is not violated. The complete augmented
bond graph is shown in Figure 8.20(d).
State Equation Extraction

The final part of the modeling process is the extraction of the state equation from the augmented bond
graph. The state equation extraction sequence is very simple and straightforward. With some amount of
practice, one can write down the state equation of any system just by inspection of the augmented bond
graph for the system. The state equation extraction sequence is as follows:
Chapter 08.indd 463 3/14/2009 7:53:07 AM

1. Define the state variables and the input variables.

2. Number the junctions that are connected to dynamic elements.
3. Write the effort and flow variables of all the bonds connected to these numbered junctions in terms of
the state variables and input variables only.
4. Apply junction equations to obtain the first-order linear differential equations.
5. There will be as many equations as there are dynamic elements with integral causality. These equations
are merged together to be represented in matrix form called the state equation.
Figure 8.21 illustrates the application of the state equation extraction algorithm. First the input and
state variables are identified. Figure 8.21(a) shows the input and state variables for the RLC circuit. Here
Vin is the input variable. Current through the inductor iL and the voltage across the capacitor vC are the
state variables. There are two junctions in the system bond graph of the RLC circuit. Both these junc-
tions are associated with dynamic elements and therefore both these junctions are uniquely numbered.
Junction 1 is associated with the inductive dynamic element and junction 2 is associated with the capaci-
tive dynamic element.
Next, the efforts and the flows of all the bonds connected to these numbered junctions will have to be
identified in terms of the input and state variables. This is shown in Figure 8.21(b). Junction 1 is a 1-junction
and therefore all the bonds connected to it will share the same flow, that is, iL. Junction 2 is a 0-junction and
therefore all the bonds connected to it will share the same effort, that is, vC. The effort of the R1-bond is obtained
from its governing equation as iL R1. Likewise, the flow of R2-bond is obtained from its governing equation as
vC / R2 . All the bond efforts and flows for the two numbered junctions are shown in Figure 8.21(b).
In the next step the linear differential equations can be read out by inspection by applying the junction
rule which states that for a 1-junction, the algebraic sum of all efforts of bonds connected to the 1-junction
is zero and for the 0-junction, the algebraic sum of all flows of bonds connected to the 0-junction is zero. It
should be noted that bond arrows give an indication of the energy flow and therefore efforts and flows of
bonds where energy enters the junction are considered positive and the efforts and flows of bonds wherein
the energy leaves the junction are considered negative.
di L
L
R1 dt
R1 L
L
i LR1
iL
iL
iL
Vin 2 Vin 2
vC
S e1 1 0 S e1 1 0
iL iL
1 1 vC vC
vC d v
v C C C R2
dt
C C
R2 R2
(a) (b)
Figure 8.21 (a) Defining input and state variables; (b) efforts and flows of bonds connected to
junctions that are associated with dynamic elements.
Chapter 08.indd 464 3/14/2009 7:53:07 AM

Considering junction 1:
As it is a 1-junction, one should apply the rule that the algebraic sum of all the efforts of bonds connected
to this junction is zero. Therefore
di
Vin − L L − iL R1 − vC = 0 (8.72)
dt
The state derivative of Eq. (8.72) can be written as a function of input and state variables by re-arranging
Eq. (8.72). Thus,
diL R 1 1
= − 1 iL − vC + Vin (8.73)
dt L L L
Considering junction 2:
As this is a 0-junction, one should apply the rule that the algebraic sum of the flows of all the bonds
connected to this junction is zero. Therefore,
dv v
iL − C C − C = 0 (8.74)
dt R2
The state derivative of Eq. (8.74) can be written as a function of input and state variables by re-arranging
Eq. (8.74). Thus,
dvC 1 1
= iL − v + 0 ⋅Vin (8.75)
dt C R2C C
Equations (8.73) and (8.75) give the dynamic behavior of the system. As there are two dynamic elements in
the RLC circuit, only two first-order linear differential equations are sufficient to fully describe its dynamic
behavior. These two equations can be combined into matrix representation to obtain the state equation as
⎡ diL ⎤ ⎡ R1 1 ⎤
⎢ ⎥ ⎢− − ⎥ ⎡1⎤
L ⎥ ⎡ iL ⎤ ⎢ ⎥
⎢ dt ⎥ = ⎢ L ⋅ ⎢ ⎥ + L ⋅Vin (8.76)
⎢ dvC ⎥ ⎢ 1 1 ⎥ ⎣v C ⎦ ⎢ ⎥
⎥ ⎢ − ⎢⎣ 0 ⎥⎦
⎢
⎣ dt ⎦ ⎣ C R2C ⎥⎦
Equation (8.76) gives the state equation of the RLC circuit of Figure 8.17. The process of obtaining the state
equation for the RLC circuit has been explained in detail in this section. The same procedure can be applied
to any physical system to obtain the state space model.
DC Motor Examples
Separately Excited DC Motor Figure 8.22(a) shows the schematic of a separately excited DC motor.
Using the method discussed above, the bond graph model with causality assigned is shown in Figure 8.22(b).
Every junction has a subscripted number that indicates the junction number. This is used in order to
identify the junction during equation formation. The state equation is obtained from the bond graph model
by the method discussed in this section. Observe that the state equation obtained is same as that derived for
the armature-controlled DC motor of Figure 8.9.
⎡ dia ⎤ ⎡ Ra k ⎤ ⎡1 ⎤
⎢ ⎥ ⎢− L − ⎥ ⎢
La ⎥ ⎡ia ⎤ ⎢ La
0 ⎥
⎡ ⎤
⎢ dt ⎥ = ⎢ a ⎢ ⎥ + ⎥⎢ v ⎥ (8.77)
⎢ dω ⎥ ⎢ k B⎥ ω ⎢ 1⎥ T
⎢⎣ dt ⎥⎦ ⎢ − ⎥⎣ ⎦ ⎢ 0 − ⎥⎣ L⎦
⎣ J J ⎦ ⎣ J⎦
Chapter 08.indd 465 3/14/2009 7:53:07 AM

La J
ia w
B Va k kia
ia J kw GY2 Bw
Se 11 12 B
ia ia w w
Td
Va ia −TL w
w
Ra Se(sin k)
Electrical domain Mechanical domain
(a) (b)
Figure 8.22 (a) Separately excited DC motor; (b) bond graph with assigned causality.
DC Motor Excited by Armature Supply The schematic of a DC motor that is excited by the same armature
supply that also supplies power to the armature is shown in Figure 8.23(a). The bond graph is developed
along similar lines discussed in this section and is shown in Figure 8.23(b). Observe that there are three state
variables as there are three dynamic elements: (a) armature inductance, (b) field inductance and (c) mechanical
inertia. Further from the bond graph given in Figure 8.23(b), there is state variable multiplication in the
mGY component that multiplies the armature current and the field current to produce the torque. The back
effect of the gyrator that produces the back emf is also a product of the two state variables: angular speed and
field current. This makes the model non-linear. The set of non-linear large-signal differential equations are
given in Eqs. (8.78a), (8.78b), (8.78c). One can obtain the linearized small-signal model by considering
perturbations in the neighborhood of the operating point as discussed in the section “Differential Equations
and Linearization”.
dia R ki ω v
= − a ia − f + (8.78a)
dt La La La
La J
ia w
B J Va Va kifw kif
ia kifia Bw
Se 01 12 mGY 14 B
i ia ia w w
if Td
Va Va if ia −TL w
w if
Rf 13 Ra Se(sin k)
if
if k

Lf
(a) (b)
Figure 8.23 (a) Armature excited DC motor; (b) bond graph with assigned causality.
Chapter 08.indd 466 3/14/2009 7:53:08 AM

La Lf
J
i i
w
Va kiw ki
Se 11 ki 2 Bw
mGY 12 B
B i i w w
J i
ia
Td i i k −TL w
Va
w Ra Rf Se(sink)
(a) (b)
Figure 8.24 (a) Series DC motor; (b) bond graph with assigned causality.
dω kif ia Bω TL
= − − (8.78b)
dt J J J
dif R v
= − f if + (8.78c)
dt Lf La
DC Series Motor Figure 8.24(a) shows the schematic of a DC series motor wherein the field winding is in
series with the armature. The bond graph of the DC series motor with the causality assigned is as shown in
Figure 8.24(b). The series motor is also a non-linear system with state multiplication. The large-signal non-linear
differential equations representing the DC series motor are given in Eqs. (8.79a) and (8.79b).
di R + Rf k 1
=− a i− iω + v (8.79a)
dt La + Lf La + Lf La + Lf
dω k 2 B T
= i − ω− L (8.79b)
dt J J J
Modeling Switched Power Systems

Bond graphs have been used extensively in the modeling of physical systems. However, modeling switched
power systems using bond graph pose some difficulties. This is because there is no switch element in bond
graph. To overcome this problem, several authors have used macro models of power switches and several
others have used modulated bond graph elements as a switch. The former method leads to very stiff systems
and the latter method switches either the flow or effort variable losing control on the other variable. To overcome
the problem of modeling switched systems, the concept of switched junctions is used. Here, the dominant
junction variable is switched to emulate the practical switch. A few switched-mode power converters are
modeled and simulated to demonstrate the efficacy of using the switched junctions.
Bond graphs have been used primarily for modeling dynamical systems where there is continuous energy
or power flow through all components of the system. They do not inherently support modeling of switched
power systems like switched-mode power converters and inverters. To overcome this problem, two
approaches have generally been used: (a) one of the methods commonly used is to macro model the power
Chapter 08.indd 467 3/14/2009 7:53:08 AM

switches using inductances, capacitances and resistances; and (b) the other method is to use modulated
transformers and gyrators with modulation index either being 1 or 0 depending on the state of the power
switch. The first approach poses no problem in modeling power electronic systems wherein the components
are operating in the linear region. However, for switched power electronic systems, the time constants of the
macro models of the power switch will be in the order of microseconds and the simulation time will be in
the order of seconds. Such a system wherein there is a large difference in time constants is called a stiff
system. If such a system were to be simulated it will take days to finish. In the second approach wherein
modulated transformers or gyrators may be used to represent power switches, the very nature of the effort–flow
relationship may get altered. This approach controls either the effort or the flow of the power switch. As a
consequence the model is not a correct representation of the physical system in most cases. To overcome
the above-stated problems, the concept of switched junction is proposed which is a generalization of the
conventional bond graph junction.
Switched Junctions
The 0-junction and the 1-junction are the multi-ports used in bond graph as discussed in a previous
section. In this section the multi-ports are extended to a more generalized mathematical framework
wherein multiple effort sources can decide a 0-junction effort and likewise multiple flow sources can
decide a 1-junction flow.
The 0-junction is defined as
∑ fk = 0
k
where fk is the flow in the kth bond of the 0-junction. The bond effort is given as
ek = e0j for all k
where e0j is the 0-junction effort. There is only one effort-decider bond at the junction which will determine
the junction effort e0j. Similarly, the 1-junction is defined as
∑ ek = 0
k
where ek is the effort in the kth bond of the 1-junction. The bond flow is given as
f k = f 1j for all k
where f1j is the 1-junction flow.
There is only one flow decider bond at the junction which will determine the junction flow, f1j. The 0- and
1-junctions, as defined above are for continuous power flow through the system. To handle switched power
flow through the system, these definitions are extended for the switched junctions. The switched 0-junction
or 0s-junction is defined as follows:
∑ fk = 0
k
where fk is the flow in the kth bond of the 0-junction and

ek = e0j for all k
where e0j is the 0s-junction effort. There can be many bonds connected to the 0s-junction with the causal
bar at the junction end. This implies that the junction effort is decided by many bonds, however at mutually
Chapter 08.indd 468 3/14/2009 7:53:09 AM

exclusive instants of time. If ei , fi are the effort and flow, respectively, of the ith effort-decider bond of the
switched 0s-junction and ui is the information signal that selects the ith effort-decider bond as the active
effort decider for the junction then
⎡ e1 ⎤
⎢ ⎥
⎢ e2 ⎥
⎢ ⎥
e0 j = ui ⋅ ⎡⎣u1 u2 un un +1 ⎤⎦ ⋅ ⎢ ⎥
⎢ ei ⎥
⎢e ⎥
⎢ i +1 ⎥
⎢⎣ ⎥⎦
⎧0 i ≠ n
ui ⋅ un = ⎨
⎩1 i = n
f i = 0 if ui ⋅ un = 0
Note that this definition is a generalization of the 0-junction definition wherein there is only one effort
decider bond at the 0-junction.
The switched 1-junction or 1s-junction is defined as follows:
∑ ek = 0
k
where ek is the effort in the kth bond of the 0-junction and
f k = f 1j ∀k
where f1j is the 1s-junction flow. There can be many bonds connected to the 1s-junction with the causal bar
away from the junction end. This implies that the junction flow is decided by many bonds, however at
mutually exclusive instants of time. If ei, fi are the effort and flow, respectively, of the ith flow decider bond
of the switched 1s-junction and ui is the information signal that selects the ith flow decider bond as the
active flow decider for the junction then
⎡ f1 ⎤
⎢ ⎥
⎢ f2 ⎥
⎢ ⎥
f 1j = ui ⋅ ⎡⎣u1 u2 un un +1 ⎤⎦ ⋅ ⎢ ⎥
⎢ fi ⎥
⎢f ⎥
⎢ i +1 ⎥
⎢⎣ ⎥⎦
⎧0 i ≠ n
ui ⋅ un = ⎨
⎩1 i = n
ei = 0 if ui ⋅ un = 0
Note this definition is a generalization of the 1-junction definition wherein there is only one flow decider
bond at the 1s-junction.
Chapter 08.indd 469 3/14/2009 7:53:09 AM

Application of Switched Junctions

The basic building block in a switched junction is a switch element. An SPDT switch is considered as the
basic switch block. All other switch circuits can be built using many SPDT switches. The SPDT switch is
shown in Figure 8.25. The SPDT switch consists of a pole P and two throw positions T1 and T2. The
characteristic of this SPDT switch is that at the throw positions, the flow is being switched. When the pole
P is switched between T1 and T2, the flow through T1 and T2 gets switched. Therefore, the throw positions
of the switch will be represented by 1s-junction. The pole P of the switch will have switching in the effort
and therefore, the pole position will be represented by a 0s-junction. When the pole P is at throw position
T1, the flow through T2 is zero. The effort at the pole P is determined from the circuit connected to throw
position T1. Similarly, when the pole P is at throw position T2, the flow through T1 is zero. The effort at the
pole P is determined from the circuit connected to the throw position T2. The bond graph of the SPDT
switch component is illustrated in Figure 8.26.
From Figure 8.26(b), one sees two 1s-junctions that are numbered 1 and 2 to represent the throw
positions T1 and T2. There is one 0s-junction numbered 3 to represent the pole P. The throw positions
T1 and T2 will take the flow of the pole ( f ) whenever the pole P is connected to them. The pole will take
efforts of the throws (eT1 or eT2) depending on whether the pole is connected to T1 or T2, respectively.
During the time when u is high or asserted, the pole effort ep is taken as the effort eT1 which is from the cir-
cuit connected to throw T1. The throw flow fT1 is taken as f which is coming from the pole circuit. At this
time the throw flow fT2 is zero which is coming from the zero flow source connected to the 1s-junction 2.
During the time when u is asserted or high, the pole effort ep is taken as the effort eT2 which is from the
T1 P
T2
Figure 8.25 Schematic of an SPDT switch.
Sf
u, 0
e T1 3
u e T1 u ep
f T1 ep 1s 0s
T1 P u f T1 f
u f 1
u f T2
T2 u
u Sf 1s
e T2 2
0
f T2 u
e T2
(a) (b)
Figure 8.26 Bond graph of an SPDT switch.
Chapter 08.indd 470 3/14/2009 7:53:09 AM

circuit connected to throw T2. The throw flow fT2 is taken as f which is coming from the pole circuit. Now
fT1 is zero which is coming from the zero flow source connected to the 1s-junction 1.
Buck Converter
Consider the buck converter circuit shown in Figure 8.27(a) which is one of the primary switched-mode
power converter. The bond graph of the buck converter circuit is shown in Figure 8.27(b). The state equations
can now be obtained from the augmented bond graph based on the rules explained in the previous sections.
However, one should note that the system is now a time-varying system as the equations are dependent on
u which is a function of time.
Referring to Figure 8.27(b), the state equation can be obtained in a manner similar to that discussed in
the previous section. There are two state variables: (a) the inductor current iL and (b) the capacitor voltage
vC. Applying the “effort law” at the 1-junction to which the inductor is connected, one obtains
diL
L = u ⋅ v i − vC (8.80)
dt
Applying the “flow law” at the 0-junction to which the capacitor is connected, we have
dvC v
C = iL − C (8.81)
dt Ro
Equations (8.80) and (8.81) can be re-arranged in the standard form to obtain
⎡ diL ⎤ ⎡ −1 ⎤
⎥ ⎢0 ⎡u ⎤
⎢
⎢ dt ⎥ = ⎢ L ⎥⎥ ⎡ iL ⎤ ⎢ ⎥
⋅ ⎢ ⎥ + L ⋅ ⎡⎣v i ⎤⎦ (8.82)
⎢ dvC ⎥ ⎢ 1 −1 ⎥ ⎣vC ⎦ ⎢ ⎥
⎥ ⎢ ⎢⎣ 0 ⎥⎦
⎢
⎣ dt ⎦ ⎣C RoC ⎥⎦
Equation (8.82) is the large-signal model of the buck converter. Observe that the input matrix contains the
Boolean variable u. During interval dTs, when the switch pole is connected to the input, u = 1. During
Sf L
u, 0 iL
Vin u VC
Se 1s 0s 1 0
u, DT iL iL
L
VC VC VC/Ro
Vin Ro u
u, (1 − D)T C
Sf 0 C Ro
1s
Se
(a) (b)
Figure 8.27 (a) Buck converter circuit; (b) augmented bond graph of buck converter.
Chapter 08.indd 471 3/14/2009 7:53:10 AM

(1 – d )Ts interval, the switch pole is connected to ground and u = 0. From Eq. (8.82), the state equations
for the two operative positions of the switch are obtained by substituting u = 1 and u = 0 for the two intervals.
This will result in the same large-signal model as obtained with the circuit averaging method. Once the
large-signal model is obtained, the averaged large-signal, steady-state and the small-signal models are
obtained in the same manner as that discussed in the section on circuit averaging method.
Boost Converter
Figure 8.28 shows the circuit topology of a boost converter circuit which is operating in continuous con-
duction mode (CCM). The operation and steady-state analysis of the boost converter circuit is discussed
SPDT
u
L RL vC
u
Vi C Ro Vo
SPDT
u
L RL vC
Vi C Ro Vo
u
(a)
Sf = 0
L
iL u
u vC u
Vi 1 0s 1s 0
u iL
vC
u vC
R
Sf u C R
RL 1s
=0
Se = 0
(b)
Figure 8.28 (a) Boost converter; (b) bond graph of the boost converter.
Chapter 08.indd 472 3/14/2009 7:53:10 AM

in Chapter 5. The SPDT switch is realized with a MOSFET and a diode as shown in the figure. The bond
graph of the boost converter is shown in Figure 8.28. The SPDT switch model of Figure 8.26 is used to repre-
sent the MOSFET-diode SPDT switch combination.
The 0s-junction has two effort-decider bonds which show that there are only two physically feasible
operating modes for this configuration. Applying the “effort law” for the 1-junction to which the inductor
is connected, one obtains
diL
L = v i − iL RL − u ⋅ vC (8.83)
dt
Applying the “flow law” for the 0-junction to which the capacitor is connected, one obtains
dvC v
C = u ⋅ iL − C (8.84)
dt Ro
From Eqs. (8.83) and (8.84), the large-signal model can be obtained as
⎡ diL ⎤ ⎡ − RL −u ⎤
⎢ ⎥ ⎢ ⎥ ⎡1⎤
⎢ dt ⎥ = ⎢ L L ⎥ ⎡ iL ⎤ ⎢ ⎥
⋅ ⎢ ⎥ + L ⋅ ⎡⎣v i ⎤⎦ (8.85)
⎢ dvC ⎥ ⎢ u −1 ⎥ ⎣vC ⎦ ⎢ ⎥
⎥ ⎢ ⎢⎣ 0 ⎥⎦
⎢
⎣ dt ⎦ ⎣ C RoC ⎥⎦
The Boolean variable u = 0 during the interval dTs and u = 1 during the period (1 – d )Ts. Substituting for
u accordingly, the large-signal models for the two operative modes of the SPDT switch are obtained. Once
the large-signal model is obtained, the averaged large-signal, steady-state and the small-signal models are
obtained in the same manner as that discussed in the section on circuit averaging method.
8.9 Space-Vector Modeling
T he electrical machines are electromechanical devices wherein there are three energy domains, namely
electrical, magnetic and mechanical domains. The electrical power sources for these electrical machines
are, in general, obtained from three-phase grid systems. A rotating field is produced in almost all of these
electromechanical devices to produce the motive torque. Systems that are energized by three- or poly-phase
energy sources can be easily modeled by using space vectors. The space-vector modeling approach is a generic
method that can be used for two-, three- or poly-phase systems. However, this section will give a flavor for
space-vector modeling by considering a three-phase induction motor as a continuous illustrative example in
this section.
The operation of the three-phase induction motor is based on Faraday’s law and the Lorentz force on a
conductor. According to Faraday’s law, if the flux linking a conductive loop varies as a function of time, then
an emf which is proportional to the rate of change of flux is induced in the loop, thereby forcing a current
to flow in the conductive loop. In the case of the induction motor, the induced currents in the rotor conduc-
tors interact with the air-gap flux to produce the torque. As the rotor tries to catch up with the magnetic flux
in accordance with Lenz’s law, the rate at which the conductors are cut by the magnetic flux is reduced. As a
consequence, the induced rotor currents decrease and the Lorentz force on the rotor conductors reduces.
Evidently, if the rotor conductors were to catch up with the magnetic flux, there would be no relative
motion between the conductors and the flux, and this would result in zero rotor currents and zero torque.
Chapter 08.indd 473 3/14/2009 7:53:11 AM

Therefore, for the induction motor to operate, there must exist some relative speed between the flux and the
conductors which is called the slip speed wsl. The slip speed is given by
ωsl = ωs − ωm (8.86)
where ws is the synchronous speed and wm is the electrical rotor speed.
Rotating Magnetic Field

It is essential that a rotating magnetic field be produced for the rotation of the induction motor. Consider the
schematic of a simple two-pole induction machine as shown in Figure 8.29. Three identical windings AN,
BN, CN each composed of two coils (Aa–aN, etc.) in series are arranged to be spatially spaced at 120° to each
other. With respect to the terminals A, B and C, the windings constitute a balanced three-phase system. The
spatial representation of the three-phase system is shown in Figure 8.29(b). The three-phase system is consid-
ered to have three spatial axes “sa”, “sb” and “sc” which are displaced at 120° with respect to each other.
Note that isa, isb and isc produce the three-phase stator flux linkages ysa, ysb and ysc. When three-phase
sinusoidal currents are fed to a balanced three-phase induction motor, a rotating magnetic field is gene-
rated. It should be noted that the stator windings should be distributed in such a manner which results
in three-phase sinusoidal flux linkages on application of the three-phase sinusoidal stator currents. To handle
three-phase quantities by means of a single equivalent vector quantity, the concept of space vectors is
introduced in the next sub-section.
A
isa
is
-ax
sb
a N
N c
b 120°
b 240°
sa-axis
isc
c C
isb a
B
is
-ax
sc
N
(a) (b)
Figure 8.29 (a) Stator structure showing the three-phase windings; (b) spatial representation
of a, b and c axes.
Chapter 08.indd 474 3/14/2009 7:53:11 AM

Space Vectors
The induction motor is a three-phase machine. Therefore, all its variables like the stator currents, rotor
currents, etc. are three-phase quantities each having three components. Consider for the moment the
stator current components isa, isb and isc. Referring to Figure 8.30, one can note that the isa component of
the stator current is along an axis in space which is represented by the sa-axis. Similarly, the isb and isc
components of the stator current are represented by the sb-axis and sc-axis in space, respectively. As these
current components are represented as vectors in the spatial co-ordinates, they are called current space
vectors. Considering the sa-axis as the spatial reference, it is evident from Figure 8.30 that the sb-axis and
the sc-axis are positioned spatially at 2p/3 and 4p/3 radians, respectively, measured anti-clockwise from
the sa-axis. The b-phase and c-phase stator current space vectors can be represented with respect to the
sa-axis as
isb = isbe j ( 2π /3) ; isc = isc e j ( 4π /3) (8.87)
The three stator current components can be represented by a single equivalent vector, called the resultant
stator current space vector is, which is given by
is = isa + isbe j ( 2π /3) + isc e j ( 4π /3) (8.88)
Similarly, any three-phase variable in the induction motor can be represented as a resultant space vector.
If c represents a general three-phase variable of the induction motor, then the resultant space vector of the
variable χ is given by
is
-ax
sb
, is
hasor
ep
pac
j4π/3
isc
lta nt s isbe
R esu
2p/3
isa
sa-axis
isb
4p/3 j 2p/3
isbe
is
-ax
sc
Figure 8.30 Space-vector representation of stator currents of the induction motor.
Chapter 08.indd 475 3/14/2009 7:53:11 AM

b -axis
cβ
c
cα
α-axis
sa -axis
(Stator axis)
Figure 8.31 Representation of a general space phasor along the orthogonal axes.
χ = χa + χ b ⋅ e j ( 2 π / 3 ) + χ c ⋅ e j ( 4 π / 3 ) (8.89)
where the variable c can represent the currents is and ir; voltages, Vs,Vr; fluxes, fs, fr, fm and flux linkages ys,
yr , ym. It should be noted clearly that ca, cb and cc denote phase quantities and not line quantities.
Representation of Space Vectors in Orthogonal Co-ordinates

The resultant space vector c can be represented with respect to two orthogonal axes, a and b. The a-axis is
aligned along the sa-axis as shown in Figure 8.31. The projections of the space vector c along the a-axis and
b-axis are cα and cβ , respectively, as shown in Figure 8.31. These orthogonal space-vectors components are
simple transformation as given by
⎡χ ⎤
⎡ χα ⎤ ⎡1 −1 / 2 −1 / 2 ⎤ ⎢ a ⎥
⎢χ ⎥ = ⎢ ⎥ χb
⋅ (8.90)
⎣ β ⎦ ⎣0 3 / 2 − 3 / 2⎦ ⎢ ⎥
⎣⎢ χc ⎥⎦
Space-Vector Transformations
Referring to Figure 8.32, the a-axis and the b-axis are fixed in space. If the space vector χ rotates, then the
projections cα and cβ of χ on the two orthogonal axes will vary even if the magnitude of χ is a constant.
This is clearly shown in Figure 8.32(a), where the space vector χ shown at two instants of time t1 and t2
results in different projections along the a−b axis, which is stationary. However, if the a−b axis also rotates
along with the space vector at the same speed, then the projections cα and cβ on the rotating a−b axis will
remain constant if the magnitude of χ is a constant as shown in Figure 8.32(b).
As an illustration, the three-phase stator currents of an induction motor, under steady-state operating con-
ditions, are sinusoidal quantities. The three-phase stator currents (isa, isb and isc) are given by the expression
isa = I sinwt; isb = I sin(wt − 2p/3); isc = I sin(wt − 4p/3)
The stator current space vector is, as given by Eq. (8.88), reduces to is = (3I/2) e j(wt − p/2). The resolved com-
ponents isα and is β of is in the a−b reference frame are given by Eq. (8.90). Thus, isa = (3I/2) cos(wt − p /2)
and is β = (3I/2) sin(wt − p/2). It is evident that the magnitude of the space vector is 3I/2, which is a
constant and the space vector is rotating at speed w. If the a−b axis is fixed in space, then the projections of
Chapter 08.indd 476 3/14/2009 7:53:11 AM

b
b(t1) a(t2)
c(t2)
cβ(t2) c(t2)
b(t2)
q
cα(t2)
χβ(t1)
c(t1) c(t1)
cβ(t1)
q
cβ(t2)
cα(t2) cα(t1) a cα(t1) a(t1)
(a) (b)
Figure 8.32 (a) Stationary a−b reference frame; (b) a−b reference frame rotating
synchronously with χ.
the stator current space vector isα and isβ along the stationary a−b axis will vary sinusoidally. However, if the
a−b axis also rotates at the same speed w as that of the space vector is, then the resolved components of the
stator current space vector will be DC in nature. Thus, even if the motor sees sinusoidally varying cur-
rents, the resolved components of the currents along the rotating orthogonal axes are unchanging with time,
that is, are DC quantities.
This concept is exploited in the vector control of induction motors. (However, one should note that
non-sinusoidal quantities in the stationary reference frame will result in a varying unipolar quantity in a
reference frame rotating synchronously with the space vector − this is the case during transient conditions in
the induction motor.)
Figure 8.33 shows the graphical representation of a space vector r. This has a magnitude r and is at an
angle q from the x-axis. This is represented in the polar co-ordinates as follows:
r x -axis = r ⋅ e jθ (8.91)
y-axis
is
r ax
a-
q r
x-axis
Figure 8.33 Space-vector representation in polar co-ordinates.
Chapter 08.indd 477 3/14/2009 7:53:12 AM

The space vector r is to be represented with respect to an arbitrary reference axis (a-axis) which is at an angle
r with respect to the x-axis as shown in Figure 8.33. (It can be noted that the arbitrary axis may or may not
rotate synchronously with the space vector r.) The space vector r can now be represented as
ra-axis = r ⋅ e ja = r ⋅ e j (θ − ρ ) = rx-axis ⋅ e − j ρ (8.92)
From Eq. (8.92), it is evident that to represent a space vector r with respect to an arbitrary reference axis
(a-axis), the x-axis representation is multiplied by a factor e−jρ which is called the transformation factor.
Generalizing, one may represent any space vector with respect to a new reference axis by multiplying the
representation of the space vector with respect to the old reference axis by the transformation factor e−jρ,
where r is the relative shift of the new axis with respect to the old axis.
a and r are to be taken as positive when measured in the anti-clockwise direction and negative when
measured in the clockwise direction. The arbitrary axis (a-axis) is called the d-axis or direct axis and the axis
perpendicular to the arbitrary axis is called the q-axis or quadrature axis.
In Figure 8.34, an induction motor space vector c is shown resolved along two co-ordinate systems, one
with respect to the a−b axes and the other with respect to the d−q axes. The d−q co-ordinate system is phase
shifted with respect to the a−b co-ordinate system by an angle r. Applying the transformation as indicated
in Eq. (8.92) and splitting into real and imaginary parts, one obtains
⎡ χd ⎤ ⎡ cos ρ sin ρ ⎤ ⎡ χα ⎤
⎢χ ⎥ = ⎢ ⎥⋅⎢ ⎥ (8.93)
⎣ q ⎦ ⎣− sin ρ cos ρ ⎦ ⎣ χβ ⎦
Equation (8.93) gives the transformation for transforming the two-phase variables in the a−b co-ordinates to
the two-phase variables in the d−q co-ordinates. If the d−q co-ordinate system is rotating synchronously with
the space vector c, then the cd and cq components will be constants if the magnitude of c is a constant. It can
be noted that in the context of the transformations carried out in this section, the a−b co-ordinates form the
stator reference frame and the d−q co-ordinates form the arbitrary reference frame, as explained earlier. In the next
sub-section, modeling of the induction motor using the concepts of the space vectors will be discussed.
b d
c
cb
q
cd r
cq
ca a
Figure 8.34 Space-vector representation in stator (a−b ) and arbitrary (d−q) reference frames.
Chapter 08.indd 478 3/14/2009 7:53:12 AM

Modeling of Induction Motor

In developing the dynamic model of the induction motor, the following assumptions are made:
1. The motor has symmetrical three-phase three-wire windings.
2. The mmf wave is sinusoidally distributed in space.
3. The stator and rotor iron have infinite permeability.
4. Skin effect and core losses are negligible.
5. The motor is operating in the linear region of the B–H characteristics of the stator and the rotor materials.
The voltage which is supplied to the stator of the induction motor should
1. sustain the drop across the stator resistance;
2. develop the emf required for setting up the stator flux linkage.
As a result
dψ s
v s = is ⋅ Rs + (8.94)
dt
where vs, is and ys are space vectors as defined by Eq. (8.89). Similarly, the rotor voltage vr, expressed with
respect to the rotor reference frame, is given by
dψ r
v r = ir ⋅ Rr + (8.95)
dt
where vr, ir and yr are space vectors as defined by Eq. (8.89). The stator and the rotor flux linkages can be
expressed in terms of the stator and the rotor current space vectors as follows:
ψ s = Lss ⋅ is + M ⋅ ir ⋅ e jε (8.96)
ψ r = M ⋅ is + Lrr ⋅ ir ⋅ e jε (8.97)
where
Lss = Lss + M (8.98)
Lrr = Lsr + M (8.99)
Also Lss, Lsr are the per phase leakage inductances of the stator and rotor, respectively; M is the equivalent
three-phase magnetizing inductance. In Eqs. (8.96) and (8.97) the space vectors ys, yr and is are represented
in the stator reference frame. But the rotor current space vector ir is represented with respect to the rotor axis
which is at an angle of e relative to the stator axis (see Figure 8.35). Therefore the rotor current space vector
ir in Eqs. (8.96) and (8.97) is multiplied by the transformation factor eje (as explained in the previous
sub-section) so that all the variables in Eq. (8.96) and (8.97) are represented in the stator reference frame.
Substituting ys and yr from Eqs. (8.96) and (8.97) into Eqs. (8.94) and (8.95), respectively, the induction
motor voltage equations in the stator reference frame are given by
dis d(i ⋅ e is )
v s = Rs ⋅ is + Lss ⋅ +M⋅ r (8.100)
dt dt
dir d(i ⋅ e − js )
v r ⋅ e j ε = Rr ⋅ ir ⋅ e j ε + Lrr ⋅ e j ε ⋅ + M ⋅ e jε ⋅ s (8.101)
dt dt
Chapter 08.indd 479 3/14/2009 7:53:13 AM

The dynamics of the rotor shaft and the load is governed by
J dωm B
= Td − TL − ωm (8.102)
p dt p
where Td is the electromagnetic drive torque; TL the load torque; B the friction coefficient; J the inertia seen
on the rotor shaft; p the number of pole pairs; wm the electrical rotor speed which is “p” times the shaft
speed. Equations (8.100)–(8.102) represent the model of the induction motor. In the following sub-section,
the model is represented in the arbitrary reference frame to arrive at the d–q axes model of the induction
motor.
Representation of the Voltages in the Arbitrary Reference Frame

Consider an arbitrary reference frame as shown in Figure 8.35 that is rotating at speed wa (which is not
necessarily equal to the synchronous speed ws). Let r be the angle between the stator reference axis a and
d-axis at any given instant of time. Then dr/dt = wa. The voltage equations [Eqs. (8.100) and (8.101)] in the
stator reference frame can be transformed to the arbitrary reference frame by using the co-ordinate trans-
formation factor e–jr. Ultimately, the intention is to ensure that the arbitrary reference frame is rotating at
the synchronous speed, that is, wa = ws, so that all the variables appear as DC quantities under steady-state
operating conditions. Therefore,
dis d(ii ⋅ e js )
v s ⋅ e − j ρ = Rs ⋅ is ⋅ e − j ρ + Lss ⋅ e − j ρ ⋅ + M ⋅ e − jρ ⋅ r (8.103)
dt dt
dir d(i ⋅ e − js )
v r ⋅ e j (ε − p ) = Rr ⋅ ir ⋅ e j (ε − ρ ) + Lrr ⋅ e j (ε − ρ ) ⋅ + M ⋅ e j (ε − ρ ) ⋅ s (8.104)
dt dt
Vs
d)
xis(
is r ya
itra
Arb
isq
isd
d ir
xis
Rotor a
g1 g
x
r e
Stator axis (a)
Figure 8.35 Space-vector diagram for the induction motor.
Chapter 08.indd 480 3/14/2009 7:53:13 AM

Referring to Figure 8.35, the stator current and voltage space vectors is and vs represented with respect
to the stator reference frame are given by
is = is ⋅ e jγ (8.105)
v s = v s ⋅ e jγ 1 (8.106)
Similarly, the rotor current and the voltage space vectors ir and vr represented with respect to the rotor axis
are given by
ir = ir ⋅ e j ξ (8.107)
v r = v r ⋅ e j ξ1 (8.108)
Substitution of Eqs. (8.105)–(8.108) in Eqs. (8.103) and (8.104), results in
j (γ 1 − ρ ) d(is ⋅ e jγ ) d(i e j (ξ + s ) )
vs ⋅ e = Rs ⋅ is ⋅ e j (γ − ρ ) + Lss ⋅ e − j ρ ⋅ + M ⋅ e − jρ ⋅ r (8.109)
dt dt
d((ir ⋅ e j ξ ) d(i ⋅ e j (γ − s ) )
vr ⋅ e j ( ξ1 +ε − ρ ) = Rr ⋅ ir ⋅ e j ( ξ +ε − ρ ) + Lrr ⋅ e j (ε − ρ ) ⋅ + M ⋅ e j (ε − ρ ) ⋅ s (8.110)
dt dt
For a squirrel cage induction motor, vr = 0 as the rotor conductors are short circuited. Referring to Figure 8.34,
one can define the following relationships:
dε
= ωm (the mechanical speed of the induction motor) (8.111)
dt
dξ
= ωξ = ωsl (the slip speed of the induction motor) (8.112)
dt
dγ
= ωs (synchronous speed) (8.113)
dt
dρ
= ωa (speed of the arbitrary reference frame) (8.114)
dt
The d–q Axes Model
The voltage space-vector equations [Eqs. (8.109) and (8.110)] are split into two orthogonal components to
obtain the d–q axes model of the induction motor. As any space vector can be represented as projections of
the space-vector magnitude along any orthogonal co-ordinate system, one can define the following relation-
ships with respect to the arbitrary reference frame:
Direct axis stator voltage, Vsd = vs cos(g1 – r) (8.115a)
Quadrature axis stator voltage, Vsq = vs sin(g1 – r) (8.115b)
Direct axis stator current, isd = is cos(g – r) (8.115c)
Quadrature axis stator current, isq = is sin(g − r) (8.115d)
Chapter 08.indd 481 3/14/2009 7:53:13 AM

Direct axis rotor current, ird = ir cos(x + e − r) (8.115e)

Quadrature axis rotor current, irq = ir sin(x + e − r) (8.115f )
Using the definitions given in Eqs. (8.111)–(8.115), the stator voltage space-vector equations [Eqs. (8.109)
and (8.110)] can be represented as stator and rotor voltage equations along the direct and the quadrature
axes of an arbitrary reference frame. Thus
disd dird
Vsd = Rs ⋅ isd + Lss ⋅ − Lss ⋅ ωa ⋅ isq + M ⋅ − M ⋅ ωa ⋅ irq (8.116)
dt dt
disq dirq
Vsq = Rs ⋅ isq + Lss ⋅ + Lss ⋅ ωa ⋅ isd + M ⋅ + M ⋅ ωa ⋅ ird (8.117)
dt dt
dird disd
0 = Rr ⋅ ird + Lrr ⋅ − Lrr ⋅ (ωa − ωm ) ⋅ irq + M ⋅ − M ⋅ (ωa − ωm ) ⋅ isq (8.118)
dt dt
dirq disq
0 = Rr ⋅ irq + Lrr ⋅ + Lrr ⋅ (ωa − ωm ) ⋅ ird + M ⋅ + M ⋅ (ωa − ωm ) ⋅ isd (8.119)
dt dt
Equations (8.116)–(8.119) can be represented in matrix form as follows:
⎡Vsd ⎤ ⎡ a11 a12 a13 a14 ⎤ ⎡isd ⎤

⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢Vsq ⎥ ⎢a21 a22 a23 a24 ⎥ ⎢isq ⎥
⎢ ⎥ = ⎢a ⋅⎢ ⎥ (8.120)
a a33 a34 ⎥⎥ ⎢ird ⎥
⎢ 0 ⎥ ⎢ 31 32
⎢⎣ 0 ⎥⎦ ⎢⎣a41 a42 a43 a44 ⎥⎦ ⎢⎣irq ⎥⎦
where
d d
a11 = a22 = Rs + Lss ⋅ ; a33 = a44 = Rr + Lrr ⋅
dt dt
d
a12 = −a21 = − Lss ⋅ ωa ; a13 = a24 = a31 = a42 = M ⋅
dt
a14 = −a23 = − M ⋅ ωa ; a32 = −a41 = − M ⋅ (ωa − ωm )
a34 = −a43 = − Lrr ⋅ (ωa − ωm )
Equations (8.116)–(8.119) are called the d–q axes voltage equations of the induction motor which results in
the d–q axis equivalent circuit (in the arbitrary reference frame) as shown in Figure 8.36. The flux linkages
shown in the equivalent circuit of Figure 8.36 are given by
ysd = Lssisd + Mird (8.121)
ysq = Lssisq + Mirq (8.122)
yrd = Misd + Lrrird (8.123)
yrq = Misq + Lrrirq (8.124)
Chapter 08.indd 482 3/14/2009 7:53:14 AM

isd ird
Rs L σs Lσr Rr
+
Vsd ysd M yrd
−
+ − − +
w a⋅y sq (w a−w m)⋅y rq
(a)
isq irq
Rs Lσs Lσr Rr
+
Vsq ysq M yrq
−
− + + −
wa⋅ysd (wa−wm)⋅yrd
(b)
w m /p
+ J B +
Td a (y rd⋅isq−y rq⋅isq) TL
− −
(c)
Figure 8.36 The d–q axis equivalent circuit of the induction motor: (a) Direct axis;
(b) quadrature axis; (c) torque model w.r.t. rotor fluxes.
When wa = ws, this equivalent circuit is with respect to the synchronous reference frame, and when wa = 0,
then this equivalent circuit is with respect to the stator reference frame. Equations (8.120) and (8.102)
together form the model of the induction motor. This is generally referred to as the d–q model of the induction
motor where a three-phase induction motor is represented as an equivalent two-phase (d-phase and
q-phase) motor. In the next sub-section, the state space approach to modeling is presented, as it is preferred
for controller and flux estimator synthesis.
State Space Representation of the d–q Model of the Induction Motor

The model of the induction motor as represented in Eq. (8.120) does not give the induction motor model
in the form of the standard state equation form. Therefore the various dynamics involved with respect to the
stator currents and the flux linkages are not very clear at the outset. For the design of the controller, it is
essential to clearly understand the nature of the time constants involved with respect to the stator currents
Chapter 08.indd 483 3/14/2009 7:53:14 AM

and the flux linkages. Therefore it is necessary to represent the d–q model of the induction motor given by
Eqs. (8.120) and (8.102) in the standard state space form. In the state space representation, the various time
constants and the interaction between the various state variables can be clearly visualized. One can note that
in the state space description of the model, the A, B, C and D matrices will be free of time derivatives for a
time-invariant system. The state space representation of the d–q model of the induction motor is essential
for synthesis of flux estimators and controllers.
From the model of the induction motor discussed in the previous section, it is evident that the induction
motor system can be split into two dynamical systems. One of the systems consists of the torque dynamics
which involves Eq. (8.120) and the other system consists of the speed dynamics which involves Eq. (8.102).
These two dynamical sub-systems are shown in Figure 8.37. The state space model for the speed dynamics is
directly obtainable from Eq. (8.102) which leads to
d(ωm / p ) ⎡ B ⎤ ωm ⎡ 1 ⎤
= ⎢− ⎥ ⋅ + ⎢ ⎥ ⋅ (Td − TL ) (8.125)
dt ⎣ J⎦ p ⎣J⎦
where (wm/p) is the state variable for the speed dynamics indicated in Eq. (8.125). To obtain the state space
model for the torque dynamics, Eq. (8.120) is used in conjunction with Eqs. (8.121)–(8.124). The resulting
equations can be reduced to the conventional state space form given by
x = A ⋅ x + B ⋅ u (8.126)
y =C ⋅x (8.127)
where x is the state vector (stator currents and flux linkages); u the input vector (stator voltages) and y the
output vector (stator currents). The exact state variables depend on the flux linkage of interest.
Torque dynamics
isd
Vsd
(d −q model)
Torque isq
Y rd
dynamics of
Vsq the induction motor
Y rq
tm = J/B
Drive torque Td + wm
generation
−
TL
Speed dynamics
Figure 8.37 Induction motor system split into torque and speed dynamical sub-systems.
Chapter 08.indd 484 3/14/2009 7:53:14 AM

While considering the state space models with respect to the arbitrary reference frame the following
models can be obtained depending on the choice of the state variables:
1. rotor flux state model;
2. stator flux state model;
3. air-gap flux state model.
In addition, a concept of generalized flux state model will be introduced which can represent any of the above
three flux state models.
Rotor Flux State Model

In this model, the state vector consists of the stator currents and the rotor flux linkages as state variables. The
different vectors used in this model are defined as follows:
State vector x = [isd, isq, yrd, yrq]T
Input vector u = [Vsd, Vsq]T
Output vector y = [isd, isq]T
The system matrix A is derived from Eq. (8.120) in conjunction with Eqs. (8.121)–(8.124):
⎡ −(1 / τ σ ) (ωm − ωsl ) v r / ( Lσ ⋅ τ r ) v r ωm / Lσ ⎤

⎢ ⎥
−(ωm + ωsl ) −(1 / τ σ ) −[(v r ωm ) / Lσ ] v r / ( Lσ τ r )⎥
A=⎢ (8.128a)
⎢ R ⋅v 0 −(1 / τ r ) ωsl ⎥
⎢ r r
⎥
⎢⎣ 0 Rr ⋅ v r −(ωsl ) −(1 / τ r ) ⎥⎦
The variables used here are defined as follows:

nr = M/Lrr; Rs = Rs + nr2Rr; Ls = Lss + nrLsr = Lss – nrM; ts = Ls /Rs ; tr = Lrr/Rr = (Lsr + M )/Rr (8.128b)
The input matrix B is given by
⎡1 / Lσ 0 ⎤
⎢ ⎥
0 1 / Lσ ⎥
B=⎢ (8.129)
⎢ 0 0 ⎥
⎢ ⎥
⎢⎣ 0 0 ⎥⎦
The output matrix C is given by
⎡1 0 0 0 ⎤
C =⎢ ⎥ (8.130)
⎣0 1 0 0 ⎦
It is important to note that the above fourth-order state space model is for induction motors that are powered
from voltage source inverters. If instead current source inverter is used, then the stator current components
isd and isq will become inputs to the system and will no longer be the state variables of the system. Therefore,
the only states are the rotor flux linkages yrd and yrq. As a consequence, if current source inverters are used
to supply power to the induction motor, the induction motor torque dynamics will behave as a second-order
system.
Chapter 08.indd 485 3/14/2009 7:53:14 AM

Stator Flux State Model

In this model, the state vector consists of the stator currents and the stator flux linkages as state variables. Here,
State vector x = [isd, isq, ysd, ysq]T
The system matrix A is given by
⎡−(1 / τ ′ ) ωsl ( Rr ⋅ v r ) / ( Lσ ⋅ M ) ωm / Lσ ⎤
σ
⎢ ⎥
−ωsl −(1 / τ ′σ ) −(ωm / Lσ ) ( Rr ⋅ v r ) / ( Lσ ⋅ M )⎥
A = ⎢⎢ ⎥ (8.131)
⎢ − Rs 0 0 ωa ⎥
⎢⎣ 0 − Rs −ωa 0 ⎥⎦

⎡1 / Lσ 0 ⎤
⎢ ⎥
0 1 / Lσ ⎥
B=⎢ (8.132)
⎢ 1 0 ⎥
⎢ ⎥
⎢⎣ 0 1 ⎥⎦
The output matrix C is given by Eq. (8.130). The variables are defined as follows:
Rs′ = Rs + Rr(nrLss/M ); ts′ = Ls /Rs′
All other variables are as defined in the rotor flux state model.
Air-Gap Flux State Model

In this model, the state vector consists of the stator currents and the air-gap flux linkages as state variables.
Here,
State vector x = [isd, isq, ymd, ymq]T
where
ymd = Misd + Mird (8.133)
ymq = Misq + Mirq (8.134)
⎡ a11 a12 a13 a14 ⎤
⎢ ⎥
⎢a a a23 a24 ⎥
A = ⎢ 21 22 (8.135)
a a a33 a34 ⎥
⎢ 31 32 ⎥
⎢a a44 ⎥⎦
⎣ 41 a422 a43
Chapter 08.indd 486 3/14/2009 7:53:15 AM

where
Rs + v r .Rr v .L .R
a11 = a22 = − ; a33 = a44 = − r σ s r
Lσ Lσ . M
⎛ v ⋅L ⎞
a12 = −a21 = ωa − ⎜ r σ r ⎟ ⋅ ωm
⎝ Lσ ⎠
Rr ω
a13 = a24 = ; a14 = −a23 = m
Lσ ⋅ Lrr Lσ
⎛L ⎞ ⎛ Lσ s ⎞
a31 = a42 = −v r ⋅ ⎜ σ r ⎟ ⋅ Rs + v r ⋅ ⎜ ⎟ ⋅ Rr
⎝ Lσ ⎠ ⎝ Lσ ⎠
v r ⋅ Lσ s ⋅ Lσ r ⋅ ωm
a32 = −a41 =
Lσ
⎛L ⎞
a34 = −a43 = ωa − ⎜ σ s ⎟ ⋅ ωm
⎝ Lσ ⎠
⎡ 1 / Lσ 0 ⎤
⎢ 0 1 / Lσ ⎥
B=⎢ ⎥ (8.136)
⎢(v r Lσ r ) / Lσ 0 ⎥
⎢ ⎥
⎣ 0 (v r Lσ r ) / Lσ ⎦
The output matrix C is given by Eq. (8.130). All variables are as defined in the rotor flux state model of the
system. In the next sub-section, a generalized flux state model is conceived which enables switching between
any of the above three flux state models dynamically.
Generalized Flux State Model

In speed and torque control applications using induction motor, the stator flux state model is preferred for
flux estimation at high speeds because the error due to stator resistance uncertainty becomes negligible at
high speeds. But at low speeds, the drop across the stator resistance becomes a significant portion of the
applied voltage and therefore the stator flux state model for flux estimation becomes inaccurate. Therefore,
for low-speed applications the rotor flux state model is preferred. By appropriate choice of two constants, k1
and k2, in the generalized flux state model, the induction motor can be represented in the rotor flux state
model, stator flux state model or the air-gap flux state model.
Let a generalized flux linkage in the arbitrary reference frame be defined as follows:
ψ gd = M ⋅ (k1 ⋅ isd + k2 ⋅ ird ) (8.137)
ψ gd = M ⋅ (k1 ⋅ isq + k2 ⋅ irq ) (8.138)
If k1 = Lss/M and k2 = 1, then the flux linkages as defined in Eqs. (8.137) and (8.138) would represent the
stator flux linkages given in Eqs. (8.121) and (8.122). If k1 = 1 and k2 = Lrr/M, then the flux linkages as defined
in Eqs. (8.137) and (8.138) would represent the rotor flux linkages as given in Eqs. (8.123) and (8.124).
Chapter 08.indd 487 3/14/2009 7:53:15 AM

Likewise, if k1 = 1 and k2 = 1, then the flux linkages as defined in Eqs. (8.137) and (8.138) would represent the
air-gap flux linkages as given in Eqs. (8.133) and (8.134). Thus by a proper choice of the constants k1 and
k2, the generalized flux linkage definitions of Eqs. (8.137) and (8.138) can be made to represent any given flux
linkage in the induction motor.
The rotor currents ird and irq can now be represented in terms of the generalized flux linkages as follows:
ψ gd − M ⋅ k1 ⋅ s sd
ird = (8.139)
M ⋅ k2
ψ gd − M ⋅ k1 ⋅ isq
irq = (8.140)
M ⋅ k2
Substituting for the rotor currents from Eqs. (8.139) and (8.140) into the d–q axes motor model of the
induction motor as given by Eq. (8.120), one obtains a set of linear differential equations with the stator
currents and the generalized flux linkages as the state variables. Here,
State vector x = [isd, isq, ygd, ygq]T
where ygd and ygq are defined as per Eqs. (8.137) and (8.138).
⎡ a11 a12 a13 a14 ⎤
⎢ ⎥
⎢a a a23 a24 ⎥
A = ⎢ 21 22 (8.141)
a a a33 a34 ⎥
⎢ 31 32 ⎥
⎢a a44 ⎥⎦
⎣ 41 a422 a43
where
Rs + v r ⋅ v k ⋅ Rr
a11 = a22 = −
Lσ
Rr ⎛ L v ⎞
a33 = a44 = − ⋅ ⎜1 − σk r ⎟
Lrr ⎝ Lσ ⎠
1 ⎛ M2 ⎞
a12 = −a21 = ωa − ωm ⋅ ⋅ ⎜ M ⋅ vk − ⎟⎟
Lσ ⎜ Lrr
⎝ ⎠
Rr
a13 = a24 =
Lσ ⋅ Lrr ⋅ k2
ωm
a14 = −a23 =
Lσ ⋅ k2
Chapter 08.indd 488 3/14/2009 7:53:15 AM

Table 8.6 Values of the constants k1 and k2 for the various flux linkage representation in
the generalized flux state motor model
Flux linkage k1 k2 uk Lsk
Rotor 1 Lrr/M ur 0
Stator Lss/M 1 Lss/M Ls /ur
Air gap 1 1 1 Lsr
⎡ L ⎛ v ⋅L ⎞ ⎤
a31 = a42 = −v r ⋅ k2 ⋅ ⎢ Rs ⋅ σ k − ⎜ 1 − r σ k ⎟ Rr ⋅ v k ⎥
⎣ Lσ ⎝ Lσ ⎠ ⎦
⎡1 ⎛ M2 ⎞ ⎤
a32 = −a41 = −v r ⋅ k2 ⋅ Lσ k ⋅ ωm ⋅ ⎢ ⎜ M ⋅ v k − ⎟ − 1⎥
⎢⎣ Lσ ⎝ Lrr ⎠ ⎥⎦
⎛ v ⋅L ⎞
a34 = −a43 = ωa − ωm ⋅ ⎜ 1 − r σ k ⎟
⎝ Lσ ⎠
⎡ 1 / Lσ 0 ⎤
⎢ 0 1 / Lσ ⎥
B=⎢ ⎥ (8.142)
⎢(v r ⋅ k2 ⋅ Lσ k ) / Lσ 0 ⎥
⎢ ⎥
⎣ 0 (v r ⋅ k2 ⋅ Lσ k ) / Lσ ⎦
The output matrix C is given by Eq. (8.130). The following are the definitions of the variables involved:
nk = k1/k2; Ls k = Lrr(k1/k2) – M
All other variables are as defined in the rotor flux state model of the induction motor.
From the generalized flux state model given by Eqs. (8.141) and (8.142), any of the previous models,
that is, the rotor flux state model or the stator flux state model or the air-gap flux state model, can be
obtained by simply choosing the suitable values of k1 and k2 according to Table 8.6.
Electromagnetic Torque Equation

The developed electromagnetic torque Td in a symmetrical three-phase smooth air-gap induction motor can
be expressed in the following general form:
Td = c ⋅ [(ir ⋅ e j ε ) × is ] (8.143)
where under linear magnetic conditions, c is a constant; e is the rotor position with respect to the stator;
ire je and is are the rotor and stator current space vectors, respectively, referred to the stator reference frame.
From the definition of the generalized flux linkages as given in Eqs. (8.137) and (8.138), the space-vector
representation of the generalized flux linkage space vector is given by
ψ g = M ⋅ (k1 ⋅ is + k2 ⋅ ir ⋅ e jε ) (8.144)
Chapter 08.indd 489 3/14/2009 7:53:15 AM

Using Eq. (8.144) in Eq. (8.143) and using the fact that cross-product of is with itself (i.e., is × is) is 0, the
electromagnetic torque of an induction motor is expressed as
Td = ca ⋅ψ g × is (8.145)
The developed drive torque Td, expressed in Eq. (8.145), is the cross-product of the generalized flux linkage
space vector yg and the stator current space vector is; ca is a constant. By using Table 8.6, the developed
torque can be expressed in terms of the rotor flux linkage, stator flux linkage or the air-gap flux linkage as
follows:
Td = c1 ⋅ψ r × is (8.146)
Td = c 2 ⋅ψ s × is (8.147)
Td = c3 ⋅ψ m × is (8.148)
Thus from Eqs. (8.146)–(8.148), it is evident that the electromagnetic torque can be expressed as the
cross-product of the rotor, stator or air-gap flux linkage space vector and the stator current space vector.
From this, it is evident that the drive torque is dependent on the sine of the angle between the flux link-
age and the stator current space vectors and is independent of their absolute positions. As a consequence,
the drive torque expression is invariant with respect to co-ordinate transformations. In other words, the
evaluation of the drive torque can be performed with respect to any reference frame without loss of gener-
ality. With respect to the arbitrary reference frame, the drive torque can be expressed as a function of the
resolved components of the generalized flux linkage and stator current space vectors along the d and q
axes as follows:
2 1
Td = ⋅ p ⋅ ⋅ (ψ gd ⋅ isq − ψ gq ⋅ isd ) (8.149)
3 k2
The drive torque expressed in Eq. (8.149) is a general expression in terms of the generalized flux linkages.
By substituting the proper values of constants k1 and k2 from Table 8.6 into Eq. (8.149), the electromag-
netic or the drive torque Td can be expressed in terms of the rotor, stator or air-gap flux linkages as
follows:
2
Td = ⋅ p ⋅ v r (ψ rd ⋅ isq − ψ rq ⋅ isd ) (8.150)
3
2
Td = ⋅ p ⋅ (ψ sd ⋅ isq − ψ sq ⋅ isd ) (8.151)
3
2
Td = ⋅ p ⋅ (ψ md ⋅ isq − ψ mq ⋅ isd ) (8.152)
3
The complete generalized state space description of the induction motor is now given by Eqs. (8.125),
(8.126), (8.127), (8.130), (8.141), (8.142), (8.149).
Chapter 08.indd 490 3/14/2009 7:53:15 AM

This chapter brought forth the general principles the input. This is due to the fact that the transfer
and methods for obtaining mathematical models function and block diagram method are defined for
for physical systems. This is often the most impor- infinite input impedance and zero output imped-
tant and crucial step in understanding and control- ance. When block diagram algebra is employed with
ling the physical systems. Before attempting to these constraints, the resulting model of the system
design a controller for a physical system, the model is inaccurate and far removed from the actual
of the physical system should be obtained. The system.
quality of the dynamic control will entirely depend The method of bond graphs is by far the most
on the accuracy of the mathematical representation complete modeling tool that can handle multiple
of the physical system. The control engineer often energy domains. Further, the loading effect is inher-
has to obtain the mathematical model for a variety ently reflected upto the first stage, thus giving a
of systems with components of diverse nature, such model that is much closer to the physical system.
as electrical, mechanical, hydraulic and thermal The bond graph method also handles non-linear ele-
systems, as well as various combinations of these. ments in a natural manner. This technique gives an
In all these cases one must first identify the vari- elegant way of obtaining the state equations of phys-
ables, which may be either flow variables or effort ical systems. The circuit averaging method is conve-
variables. In most cases it is helpful to draw an nient for the switched-mode power circuits.
analogous electrical circuit for which the differen- However, when complex multi-domain systems like
tial equations are easily obtained by applying motor, pumps, hydraulics, etc. are involved, the
Kirchhoff ’s laws. bond graph method is most apt.
One starts by writing the differential equations The space-vector modeling method discussed
relating the variables in each component of the towards the end of the chapter is a very powerful
system, as well as the interconnection equations. method especially for polyphase power electronic
Although in most cases these differential equations systems. Modeling AC machines like induction
are non-linear, as a first approximation they are lin- machines, slip ring inductions machines, synchro-
earized about the operating points. In this case it is nous machines, three-phase inverter systems, etc. are
very convenient to identify one variable as the input, easily handled by the space-vector methodology.
or cause, and another variable as output, or effect. The tutorial exercises in the next section give few
This leads to the block diagram for each component, sample practice modeling exercises that should
which relates the Laplace transforms of the input enable one to internalize the modeling process for
and the output through the transfer function. The any given physical system. Many more exercises than
overall transfer function of the system is then provided in the tutorials and the question set should
obtained either by simplifying the block diagram or be practiced in order to gain insight into the power
by directly using Mason’s rule. However, it should electronics system modeling. It should be noted that
be noted that the transfer function and block dia- an accurate and precise model of the physical system
gram approach has one significant drawback and will result in a better controller that will be discussed
that is the lack of reflection of the loading effect on in later chapters.
Chapter 08.indd 491 3/14/2009 7:53:16 AM

| TUTORIAL EXERCISES
1. Consider the DC–DC converter circuit shown N2 (L2)
in Figure 8.38. During the dTs period the tran-
iin
sistor Q is ON and during (1 – d )Ts period the
N1 (L1) iD
transistor is OFF and the diode is ON.
vin C
can be studied by R vo
a. Simulation in Spice Q
b. Simulation in SciLAB
Figure 8.38 DC–DC converter circuit.

Tasks for study:
(a) Using the circuit averaging method and the bond graph method obtain the averaged large-signal
model of the above converter. Verify that the state equation is given as
⎡ diin ⎤ ⎡⎢ ⎛N
− ⎜⎜ 1
⎞⎛ 1− d ⎞⎤
⎡ d ⎛ N1 ⎞ ⎛ 1 − d ⎞⎤
⎢ ⎥
0 ⎟⎟ ⎜⎜ ⎟⎟⎥
⎢ dt ⎥ = ⎢⎢ ⎝ N2 ⎠ ⎝ L1 ⎠⎥ ⎡ iin ⎤ ⎢ + ⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟⎥
⎢ dvC ⎥ ⎢⎛ N 1 ⎞⎛ 1− d ⎞ ⎥ ⋅ ⎢v ⎥ + ⎢ L1 ⎝ N 2 ⎠ ⎝ L1 ⎠⎥ ⋅ ⎡⎣v in ⎤⎦
⎢ ⎥ ⎜ ⎟⎟ ⎜ −
1 ⎥ ⎣ C⎦ ⎢ ⎥
⎣ dt ⎦ ⎢⎣⎜⎝ N 2 ⎟ ⎥ ⎣ 0 ⎦
⎠⎝ C ⎠ RC ⎦
(b) Obtain the steady-state model from the 2. Consider the push–pull converter circuit
averaged large-signal model. From the shown in Figure 8.39. Ideal switches, diodes
steady-state model obtain V o /V in and transformer can be assumed.
relationship. Mode of implementation: The above circuit
(c) If Vin = 15 V, D = 0.3, N1:N2 is 1:2. Evaluate can be studied by
Vo. a. Simulation in Spice
(d) Obtain the small-signal model for the b. Simulation in SciLAB
above system.
Tasks for study:
(e) Simulate the DC–DC converter in either
(a) Using the circuit averaging method and the
spice or SciLab.
bond graph method obtain the averaged
(f ) Simulate the obtained large-signal and
large-signal model of the above converter.
small-signal model equations in SciLAB or
(b) Obtain the steady-state model from the aver-
MATLAB and compare the state variable
aged large-signal model. From the steady-
waveforms with the circuit waveforms of
state model obtain Vo/Vin relationship.
Step (e).
Chapter 08.indd 492 3/14/2009 7:53:16 AM

S2 L
D1 Vo
C R
vin
S1 n:1
D2
Figure 8.39 Push–pull converter circuit.
(c) Obtain the small-signal model for the above system. Verify that the small-signal model is given as
⎡ diˆL ⎤ ⎡ ⎛ 1 ⎞⎤
⎡⎛ D ⎞ ⎛ Vin ⎞⎤
⎢ ⎥ ⎢ 0 − ⎜ ⎟⎥
⎢ dt ⎥ = ⎢ ⎝ L ⎠⎥ ⎡ iˆL ⎤ ⎢⎜ ⎟ ⎜ ⎟⎥ ⎡vˆ ⎤
⋅ ⎢ ⎥ + ⎢⎝ n ⋅ L ⎠ ⎝ n ⋅ L ⎠⎥ ⋅ ⎢ in ⎥
⎢ dvˆC ⎥ ⎢⎛ 1 ⎞ 1 ⎥ ⎢⎣vˆC ⎥⎦ ⎢ ⎢⎣ dˆ ⎥⎦
⎢ ⎥ ⎢⎜ ⎟ − ⎥ ⎣ 0 0 ⎥⎦
⎣ dt ⎦ ⎢⎣⎝ C ⎠ RC ⎥⎦
(d) Simulate the DC–DC converter in either Mode of implementation: The above circuit
spice or SciLab. can be studied by
(e) Simulate the obtained large-signal and a. Simulation in Spice
small-signal model equations in SciLAB or b. Simulation in SciLAB
MATLAB and compare the state variable
Tasks for study:
waveforms with the circuit waveforms of
(a) Consider iL1, iL2, vC1 and vC2 as the state
Step (d).
variables.
3. Consider the non-isolated Cuk converter cir- (b) Using the circuit averaging method and
cuit shown in Figure 8.40. The non-idealities the bond graph method obtain the aver-
with respect to the inductor and capacitor are aged large-signal model of the above
included in the form of lumped resistors rL1, converter.
rL2, rC1 and rC2.
iL1 i L2
rL1 L1 rC1 rL2 L2

C1
S rC2
vin R Vo
dTS (1−d )Ts
C2
Figure 8.40 Non-isolated Cuk converter circuit.
Chapter 08.indd 493 3/14/2009 7:53:16 AM

(c) For L1 = 1.94 mH, L2 = 0.96 mH, rL1 = 0.17 Ω, rL2 = 0.067 Ω, C1 = 850 mF, C2 = 45 mF, rC1 = 0.05
Ω, rC2 = 0.1 Ω, R = 30 Ω, Vin = 10 V and D = 0.65, verify that the large-signal model during the DTs
period is
⎡ • ⎤
⎢i ⎥
⎡1.94e − 3 0 0 0 ⎤ ⎢ L1 ⎥ ⎡ −0.17 0 0 0 ⎤ ⎡ iL1 ⎤ ⎡1⎤
⎢ ⎥ ⎢ • ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ 0 0.96e − 3 0 0 ⎥ iL2 0 −0.2166 −1 −0.9967 ⎥ ⎢ iL2 ⎥ ⎢0 ⎥
⋅⎢ • ⎥ = ⎢ + v
⎢ 0 0 850e − 6 0 ⎥ ⎢ ⎥ ⎢ 0 1 0 0 ⎥ ⎢ vC1 ⎥ ⎢0 ⎥ in
⎢ ⎥ v ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢⎣ 0 0 0 45e − 6 ⎥⎦ ⎢ C• 1 ⎥ ⎢⎣ 0 0.9967 0 −0.0332 ⎥⎦ ⎢⎣vC2 ⎦⎥ ⎢⎣0 ⎥⎦
⎢ ⎥
⎢⎣vC2 ⎥⎦
and for the (1 – D)Ts duration verify that the large-signal model is
⎡ • ⎤
⎢i ⎥
⎡1.94e − 3 0 0 0 ⎤ ⎢ L•1 ⎥ ⎡−0.22 0 −1 0 ⎤ ⎡ iL1 ⎤ ⎡1⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ 0 0.96e − 3 0 0 ⎥ iL2 0 −0.167 0 −0.9967 ⎥ ⎢ iL2 ⎥ ⎢0 ⎥
⋅⎢ • ⎥ = ⎢ + v
⎢ 0 0 850e − 6 0 ⎥ ⎢ ⎥ ⎢ 1 0 0 0 ⎥ ⎢ vC1 ⎥ ⎢0 ⎥ in
⎢ ⎥ v ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢⎣ 0 45e − 6 ⎦⎥ ⎢ C1 0.9967 0 −0.0332 ⎥⎦ ⎢⎣vC2 ⎦⎥ ⎢⎣0 ⎥⎦
⎢ • ⎥ ⎣⎢
0 0 0
⎢⎣vC2 ⎥⎦
(d) Obtain the steady-state model from the (f ) Simulate the DC–DC converter in either
averaged large-signal model obtained spice or SciLab.
above. From the steady-state model, obtain (g) Simulate the obtained large-signal and
Vo/Vin relationship. small-signal model equations in SciLAB or
(e) Obtain the small-signal model for the MATLAB and compare the state variable
above system. waveforms with the circuit waveforms of
Step (f ).

1. System identification technique is generally 5. The “through” variable may also be described as
used in of system parameters. variable and the “across” variable can
also be described as the variable.
2. The physical modeling techniques obtain the
mathematical representation of the physical 6. The equations of equilibrium in different
system from . systems are based on principles.
3. Th e input–output relations for the system 7. The differential equations characterize the
are given in a graphical form as of the system.
characteristic.
8. The smallest set of physical quantities called
4. The input–output transfer curves do not give variables completely determines the
any information on the of the output of the system in the absence of exter-
to a given system input. nal excitation.
Chapter 08.indd 494 3/14/2009 7:53:17 AM

9. The state variables that define the system state 22. In electrical networks, the can be
are not . considered to form the generalized co-ordi-
nates.
10. For a variable to qualify for consideration as a
state variable, it should have the important 23. The Lagrange method cannot model
property of . power systems and the related to
such systems like the switched-mode DC–DC
11. The transfer function of the system is defined
converters.
as the ratio of the Laplace transforms of the
to the under 24. In circuit averaging method, the steady-state
initial conditions. solution is obtained by equating the rate of
change of the state variables to .
12. The transfer function gives a time
model of the physical system. 25. Bond graph is a tool for modeling
physical systems.
13. The block diagram representation is undirec-
tional since the cannot produce the 26. Bond graphs display both and
. exchanges between components.
14. The transfer functions are defined between 27. One-ports are elements exchanging energy
a single and a output. with the system through bond only.
15. For multiple-input and multiple-output 28. Two-ports have bonds for energy
(MIMO) systems one can represent the input– exchange with the system.
output relationship through a transfer function
29. The 0-junction is a multi-port where the
.
on all bonds are same and the add
16. The order of the denominator polynomial of up to zero.
the transfer function indicates the of
30. The 1-junction is a multi-port where the
the system.
on all bonds are same and the add
17. For causal systems, the order of the denominator up to zero.
polynomial of the transfer function is
31. There can be only one bond with a
than or the order of the numerator
at the 0-junction.
polynomial.
32. There can be only one bond a causal
18. In the block diagram approach, there is
bar at the 1-junction.
of loading effect on the preceding block.
33. Systems that are energized by three-or poly-
19. Absence of loading effect in block diagrams
phase energy sources can be modeled by using
implies that each block of the block diagram
.
has input impedance and
output impedance. 34. When three-phase sinusoidal currents are fed
to a balanced three-phase induction motor, a
20. Lagrange’s method deals with scalar quantities
space vector is generated.
like and energy.
35. The electromagnetic torque of an induction
21. In the case of mechanical systems, the general-
motor is to frame transformation.
ized co-ordinates are the independent
or the displacements of the different
masses of the systems.
Chapter 08.indd 495 3/14/2009 7:53:17 AM

1. Write short notes on the following: (a) system 15. Explain the circuit averaging method of mod-
identification; (b) physical modeling. eling switched power circuits.
2. Distinguish between linear and non-linear 16. What are the merits of the circuit averaging
input–output relationships of a physical system. modeling method?
3. In plotting the input–output transfer charac- 17. Write short notes on the following: (a) large-
teristics of a system, which variable is consid- signal model; (b) averaged large-signal model;
ered independent and which is considered the (c) steady-state model; (d) small-signal model
dependent variable? for switched power circuits.
4. Explain large-signal model, steady-state model 18. What are the main features of the bond graphs
and small-signal model of a physical system. that make it an attractive multi-energy domain
modeling method?
5. Write short notes on the following: (a) system
state; (b) state variables; (c) state space; 19. How are the bond graph element classified?
(d) order of the system.
20. What is one-port? Discuss the one-port bond
6. What variables qualify to be considered as state graph elements and their governing equations.
variables? Discuss.
21. What is two-port? Discuss the two-port bond
7. Discuss on the state equation and output equa- graph elements and their governing equations.
tion of a state space representation of a physical
22. What is multi-port? Discuss the multi-port
system.
bond graph elements and their governing
8. Write short notes on the following: (a) linear equations.
time-invariant system; (b) linear time-varying
23. Write short notes on the following: (a) effort
system.
variable; (b) flow variable in bond graph.
9. How does the transfer function method of rep-
24. What are the power variables for the different
resenting a physical system differ from the state
energy domains? What is the criterion that
space representation?
needs to be considered to qualify a variable as
10. What are poles and zeros of the systems? Give a effort or flow in a particular energy domain?
physical interpretation for the poles and zeros
25. Distinguish between impedance transformer
of a system.
and class transformer.
11. Discuss the block diagram representation of
26. What is causality? List the governing equations
systems.
of the bond graph elements for flow causality.
12. Explain the block diagram algebra or block
27. List the governing equations of the bond graph
diagram reduction process.
elements for effort causality.
13. Explain Lagrange’s method of modeling physi-
28. What are the rules for the selection and assign-
cal systems.
ment of causality for a bond graph?
14. What are the limitations of the Lagrange’s
29. What are the steps in the bond graph method-
method?
ology to obtain the model of a physical system?
Chapter 08.indd 496 3/14/2009 7:53:17 AM

30. Why are switched junctions used in bond 35. Discuss the modeling of the induction motor
graph? to obtain the d–q axis model.
31. What are switched 0 and switched 1 junctions? 36. Explain the method of obtaining the state space
Why are they considered to be generalizations representation of the induction motor from the
of the 0 and 1 junctions? d–q axis model?
32. Discuss the representation of space vector 37. What are the state variables of the induction
wherein currents are fed to three axes that are motor? What is the order of the system?
place 120° apart in a co-planar space.
38. What is generalized flux state model of the
33. How are the three-axis components converted induction motor? What are its features?
to two-axis components? What is the transfor-
39. What are the different representations of the
mation matrix?
electromagnetic torque in an induction
34. How are the variables in a stationary reference motor?
shifted to an arbitrary rotating reference?
40. Why is electromagnetic torque of an induction
Discuss.
motor invariant with respect to frame transfor-
mation?
| PROBLEMS
1. Are the following input–output relationships 5. Boost converters are used as front-end convert-
linear or non-linear? ers wherein the input or inductor current is
controlled to be in-phase with the input volt-
a. y = au + b
age waveshape. This is discussed in Chapter 10.
b. y = au + by To achieve input current control, from the
small-signal model of the boost converter
c. y = au 2 + b
obtain the transfer function, (iîn / dˆ )( s ). (Hint:
d. y = a ⋅ (u ⋅ y ) + b Consider iin as one more output of the output
equation. The output matrix of C matrix will
e. y = a ⋅ arctan(u ) + b
be a two row, two column matrix).
2. From the small-signal model of the buck con-
6. Derive the large-signal, averaged large-signal,
verter, obtain the transfer functions for the
steady-state and small-signal models for a
small-signal deviations of output with respect
buck–boost converter.
to the input and duty cycle, that is, (vô / vîn )( s )
and (vô / dˆ )( s ). 7. From the small-signal model of the buck–boost
converter, obtain the transfer functions for the
3. Derive the large-signal, averaged large-signal,
steady-state and small-signal models for a boost
converter.
and (vô / dˆ )( s ).
4. From the small-signal model of the boost con-
8. Include the non-idealities like the inductor
verter, obtain the transfer functions for the
winding resistances, diode drops, switch satu-
ration drops and capacitor equivalent series
resistance (ESR) into the dual switch forward
and (vô / dˆ )( s ) .
Chapter 08.indd 497 3/14/2009 7:53:17 AM

converter circuit. Derive the large-signal, aver- vectors that are represented in a two orthogo-
aged large-signal, steady-state and small-signal nal axes space. What are the current frequen-
models. cies in the two orthogonal axes if the orthogonal
axes are rotating at a frequency of (a) 0 Hz,
9. Include the non-idealities like the winding
(b) 25 Hz and (c) 50 Hz?
resistances, diode drops, switch saturation
drops and capacitor ESR into the flyback con- 12. Derive relationships for the following frame
verter circuit. Derive the large-signal, averaged transformations:
large-signal, steady-state and small-signal
a. Transform variables from three-phase sta-
models.
tionary reference frame to two-phase sta-
10. Given a space vector, how can it be resolved tionary reference frame.
into (a) three-axis components wherein the b. Transform variables from two-phase sta-
three axes are 120o apart and (b) two-axis com- tionary reference frame to two-phase rotat-
ponents wherein the two axes are orthogonal to ing reference frame rotating at speed wa.
each other. c. Transform variables from two-phase rotat-
ing reference frame that is rotating at speed
11. The currents that are applied to the three axes
wa to two-phase stationary reference frame.
that are 120o apart in space are isa = I sinwt;
d. Transform variables from two-phase sta-
isb = I sin(wt – 2p/3); isc = I sin(wt – 4p/3)
tionary reference frame to three-phase sta-
where the frequency of the current waveforms
tionary reference frame.
is 50 Hz. Transform these currents into space
| ANSWERS
Fill in the Blanks
1. on-line estimation 13. “effect”; “cause” 25. graphical
2. first principles 14. input; single 26. energy; signal
3. input–output transfer 15. matrix 27. one
4. time evolution 16. order 28. two
5. “flow”; “effort” 17. greater; equal to 29. efforts; flows
6. energy conservation 18. absence 30. flows; efforts
7. dynamics 19. infinite; zero 31. causal bar
8. state; evolution 20. potential; kinetic 32. without
9. unique 21. linear; angular 33. space vectors
10. memory 22. charges 34. rotating
11. output; input; zero 23. switched; dynamics 35. invariant
12. linear; invariant 24. zero
Chapter 08.indd 498 3/14/2009 7:53:17 AM

Control System Essentials
Learning Objectives
CHAPTER
9
represent a controller in the digital domain.
convert analog domain models to discrete domain representations.
specify performance objectives for a system.
T here are some generic issues with respect to control system engineering that are very useful in both
analysis and design. Many such aspects like sampling, representation in discrete domain, continuous to
discrete domain conversion will be used in many contexts in this and other chapters as well. However, it will
not be easy to discuss these issues every time they crop up during the analysis and design in the various
chapters. This is therefore a utility chapter that discusses the important general tools and concepts and hence
this chapter is titled control system essentials.
Figure 9.1 shows the block diagram of a typical control system. It consists of three major functional
blocks, viz. the controller, the plant to be controlled and the feedback block which mainly performs the
signal processing on the sensed signal so that it is converted to a compatible form that can be fed back to the
controller. Until recently, all the three functional blocks were designed and implemented in the continuous
or analog domain. However, in today’s systems there is a vertical divide between the controller and the plant
to be controlled. Majority of the controllers exist in the digital domain whereas the plant itself exists in the
continuous or the analog domain. This means that the feedback signal processing block is functionally sub-
divided into an analog signal processing (ASP) part and a discrete signal processing (DSP) part.
One may wonder as why such a shift from analog controller to digital controller has occurred even
though analog controllers have many advantages like better accuracy, better precision, lesser developmental
infrastructure, more compact, better noise immunity, etc. as compared to their digital counterparts. The
digital controllers score over the analog controllers in a few significant issues which make them more preva-
lent in today’s control systems. They are as follows
1. Computation: Complex control structures involving complex equations can be handled easily.
2. Repeatability: Analog circuits have a major problem of parameter drift with temperature which is not
an issue in the case of digital circuits. Therefore, in the case of large-scale productions, digital controllers
will produce repeatable results.
3. Reusability: The same hardware can be used for other applications by re-programming the embedded
software for a different application.
Chapter 09.indd 499 3/14/2009 2:54:06 PM

Reference Output
+ Controller Plant
−
Feedback
Signal
processing
DSP ASP
Digital domain Analog domain
Figure 9.1 General block schematic of a typical control system.
Therefore, most controllers in today’s systems will exist in the digital domain. This means that there will be
a sampler in the feedback signal processing block that will convert the continuous or analog signals into dis-
crete signals. The transaction between the ASP in the analog domain and the DSP in the digital domain of
the signal processing block is through an analog-to-digital converter (ADC). There is one more transaction
that will take place between the digital and the analog domains and that is between the controller and the
plant. This depends on the nature of the control input that may be given to the plant. If the plant expects a
continuous control input, then a digital-to-analog conversion (DAC) is necessary. However, many plants
can accept control inputs in the discrete form like pulse-width modulated (PWM) type of signals. In such
cases the digital domain output of the controller may be directly interfaced with the discrete control input
of the plant without the need of a DAC.
To analyze and design the controller, it is essential that all the three major functional blocks be in a
common domain, that is, either all should be equivalently viewed in the digital domain or all should be
equivalently viewed in the analog domain. Based on this, one can envisage two approaches to the design of
digital controllers:
1. Let the plant and the signal processing functions be expressed in the continuous or the analog domain.
Design the controller in the analog domain. Then convert the continuous domain controller to the
discrete domain controller.
2. Let the continuous domain plant and the continuous domain signal processing function be converted
to the discrete domain. Then design the controller directly in the discrete domain itself.
Now the question arises as to which of the above two approaches is better. In the first approach, the con-
troller is designed in the analog domain and then converted to the discrete domain. This means that the
information about the sampling time is absent at the time of controller design. This would imply that
the design would work better only when the sampling time tends towards zero. However, low sampling
time would put a heavy burden on the digital processor where the controller resides. On the other hand, in
the case of the second approach, all the analog domain blocks like the plant and the ASP portion of the
signal processing block are first translated into the digital domain models by incorporating the sampling
time information. Now the digital controller is designed for the converted discrete plant with the specific
sampling time. Therefore, the digital processor can work at higher loop times implying lower overheads
and hence a lower cost.
Chapter 09.indd 500 3/14/2009 2:54:07 PM

Control System Essentials 501
Based on the above discussion, the following control system tools are essential:
1. representing signals and systems in the discrete domain;
2. methods for converting the analog domain models into the discrete domain models.
Apart from the above, the last part of the chapter will discuss few control system basics which will aid the
reader in understanding the various control concepts.
9.1 Representation of System in Digital Domain
A system always exists in the time domain. However, the system can be represented in different domains
to get more information on the nature and characteristic of the system. Therefore, the physical system
can have a mathematical model in every domain in which it can be viewed. In general, systems can be
represented in various domains as shown in Table 9.1.
Referring to Table 9.1, the independent variable is “t” in the continuous time domain. In the discrete
time domain, the independent variable “t” is discretized at regular intervals of width “T ” called the sampling
time. In the discrete time domain, the discretized independent variable is called “n”. The continuous and the
discrete time lines are illustrated in Figure 9.2.
Therefore, the discrete time line, n is given as
n = [0T, 1T, 2T, 3T, 4T, …, kT ], k → ∞
where T is the sampling time. To make the notation simpler, the discrete time line is written as
n = [0, 1, 2, 3, 4, …, k]
Table 9.1 Various domains for the signals and systems

Domains Continuous Discrete Used for
Time t n Responses, wave shapes
Frequency w w Spectral analysis
Pole-zero s z Analysis, design
Continuous
0 t time line
T
Discrete
0 1T 2T 3T 4T ··· kT n time line
Figure 9.2 Discretization of the time line.
Chapter 09.indd 501 3/14/2009 2:54:07 PM

where it is assumed that the sample intervals are at regular time period T. Thus if one writes the discrete
time n = k, then it means n = kT units of time.
The frequency domain variable is w. It is the same for both the continuous and discrete domains. How-
ever, in the pole-zero domain, the continuous plane variable is the Laplace variable “s” and the discrete plane
variable is the z-transform variable “z”. In the pole-zero domain, the continuous plane is called the s-plane
and the discrete plane is called the z-plane.
Although a system always exists in the time domain whether it be continuous or discrete, it is important
to transit between the various system representations to gather important characteristic information about
the system. Further, a difficult analysis in one domain becomes easy in another domain. Therefore, it is
essential to transit between domains to increase one’s understanding of a system. The system time responses,
steady-state accuracies, overshoots, waveshape and their analysis are most appropriately viewed in the time
domain. Information like bandwidth, harmonic contents, spectrum, energy spectrum and power spectral
density are better understood when viewed in the frequency domain. The pole-zero domain provides infor-
mation about damping, natural frequencies, time constants and dominant time constants for analysis and
design (e.g., about transfer function and location of poles and zeros).
9.2 The z-Transform
N ow consider a continuous signal x(t) which is a function of “t” in the continuous time domain. In any
other domain, the same signal is a function of the domain’s local variable. The representation of the
continuous time signal x(t) in the various domains is indicated in the Table 9.2.
While transiting from an analog domain to a digital domain, two key components are very
crucial. One is the analog-to-digital converter itself and the other is the sample-and-hold circuit. The
sample-and-hold circuit consists of two important functions: (a) sampling the signal at pre-determined
times and (b) holding the value till the next sample so that the ADC can convert the held analog value to
the corresponding digital value. The sampling logic signal in practice is of the order of few hundred
nanoseconds to a few microseconds to allow for the acquisition of the analog signal by the hold circuit.
However, for mathematical representation, the acquisition will be assumed to be instantaneous. Evi-
dently, the impulse signal can be used to represent the sampling instants. Therefore, it is essential to first
understand the mathematical representation of a train of impulse which is elucidated by Figures 9.3(a)–(d).
Figure 9.3(a) is a single unit impulse occurring at 0T point on the discrete time line. This impulse is
represented as d(n − 0T ). If the sampling time T is assumed to exist then it can be represented as d(n − 0)
or in this case as simply d(n).
Figure 9.3(b) shows an impulse occurring at the discrete time point of 1T. This may be viewed as a delayed
impulse by 1T or it may be viewed as an impulse occurring at 0T on a discrete time line which is shifted left
Table 9.2 Continuous and discrete variables in various domains

Domains Continuous Discrete
Time x(t) x(n), x(kT )

T – sampling time
Frequency x(w) x(w)
Pole-zero x(s) (Laplace transform) x(z) (z-transform)
Chapter 09.indd 502 3/14/2009 2:54:08 PM

1.5
0.5
0
−5 0 5 10
kT n
(a)
1.5
0.5
0
−5 0 5 10
kT n
(b)
Chapter 09.indd 503 3/14/2009 2:54:08 PM

1.5
0.5
0
−5 0 5 10
kT n
(c)
1.5
0.5
0
−5 0 5 10
kT n
(d)
Figure 9.3 (a) Impulse function δ(n − 0T ) or δ(n − 0) or simply δ(n); (b) delayed impulse δ(n − 1T ) or
∑ k =0 δ(n − kT) or simply

5
simply δ(n − 1); (c) two impulses − δ(n) + δ(n − 1); (d) impulse train
∑ k =0 δ(n − kT ) .
5
Chapter 09.indd 504 3/14/2009 2:54:08 PM

by 1T, that is, (n − 1T ) time line instead of n time line. In that case, the discrete impulse is represented at
d(n − 1T ) or simply d(n − 1). Figure 9.3(c) shows two impulses one occurring at 0T and the other occurring
at 1T on the discrete time line. Both these impulses together are represented as d(n) + d(n − 1) or ∑ k =0 δ (n − k ).
1
Likewise, Figure 9.3(d) is a train of six unit impulses which is represented as ∑ k =0 δ (n − k ) .

5
Now, consider the signal x(t) shown in Figure 9.4. For simplicity, the signal x(t) considered is a linear
ramp. The signal is sampled with sample interval T. The sample interval is the inverse of the sampling fre-
quency. The sampling frequency is chosen such that it is at least twice the maximum frequency component
of the band-limited signal x(t). This means that one has to have a practical knowledge of the frequency spec-
trum of x(t) to decide the sampling frequency. To discretize the signal, all one needs to do is multiply x(t) or
modulate x(t) with a train of unit impulses. This sampled signal is passed through a hold device which holds
the value of the impulse sampled value till the next sampling instant. This discretized signal x(n) is shown in
Figure 9.4. The discrete domain representation is as follows:
x(n ) = x (0)δ (n ) + x (1)δ (n − 1) +
∞
= ∑ x (k ) ⋅ δ (n − k ) (9.1)
k =0
where x(k) is the value of x(t) at the sample instant kT. Equation (9.1) is the discrete representation of
the continuous domain signal x(t). However, the representation as given in Eq. (9.1) is not very amenable
to mathematical analysis. Therefore, to convert the representation of Eq. (9.1) to a more useful formula-
tion that is amenable to mathematical analysis and computer manipulations, the following definition is
made: Define
x
x(t)
x(n)
t
2 3 4 5 6 7 n
T
0 1
Figure 9.4 A continuous signal x(t) and its discretized version x(n).
Chapter 09.indd 505 3/14/2009 2:54:08 PM

∞
X ( z ) ∑ x (k ) ⋅ z − k (9.2)
k =0
Comparing Eqs. (9.1) and (9.2), one can observe that the impulse component of Eq. (9.1) has been replaced
by z−k. X(z) is called the z-transform of x(n). By making a definition as given in Eq. (9.2), a very major advan-
tage has been gained. The discrete representation of Eq. (9.1) is now converted into a polynomial in z–1. This
significant accomplishment is achieved just by re-defining the discrete representation x(n) as X(z). Further,
one may note that the coefficients of the n-domain representation of the signal as given in Eq. (9.1) and that
of the z-transformed representation as given in Eq. (9.2) are the same. Hence there exists a one-to-one corre-
spondence between the signal representation in the n-domain and the z-domain. This means that one may
shift between n-domain and z-domain just by inspection as the coefficients are identical in both the
domains.
Physical Significance of z–1

If the z-domain representation of x(n) [i.e., Eq. (9.2)] is multiplied by z−1, then
∞
X ( z ) ⋅ z −1 = ∑ x (k ) ⋅ z −(k +1) (9.3)
k =0
This when reconstructed will give a signal that is delayed with respect to the original signal x(t) by one unit
time step (i.e., 1T ). This is illustrated by the following example.
EXAMPLE 9.1 Consider the signal x1(t) shown in Figure 9.5. The signal is sampled at regular intervals
of time T as indicated in the figure.
1.0
0.8 0.8
0.5 0.5
0 1 2 3 4 5 6 7 t, n
Figure 9.5 An example signal x1(t).
Based on Eq. (9.2), the z-transform of x 1(t) can be written by inspecting

Figure 9.5 as
X 1( z ) = 0 z − 0 + 0.5z −1 + 0.8z −2 + 1z −3 + 0.8z −4 + 0.5z −5 + 0 z −6 (9.4)
Chapter 09.indd 506 3/14/2009 2:54:08 PM

Now, multiply both sides of Eq. (9.4) by z–1. One obtains

X 2 ( z ) = z −1 ⋅ X 1( z ) = 0 z −1 + 0.5z −2 + 0.8z −3 + 1z −4 + 0.8z −5 + 0.5z −6 + 0 z −7 (9.5)
Referring to Eq. (9.5), and by inspecting the coefficients of the power of z–1, one
can re-construct the n-domain signal x2(t). This is shown in Figure 9.6.
x1
1.0
0.8 0.8
0.5 0.5
0 1 2 3 4 5 6 7 t, n
0 1 2 3 4 5 6 7
x2
1.0
0.8 0.8
0.5 0.5
0 0
0 1 2 3 4 5 6 7 t, n
Figure 9.6 Re-construction of x2(t) from X2(z).
It is clear from Figure 9.6 that x2(t) is a unit time delayed version of x1(t). This implies that passing a signal
through a z−1 block introduces a unit time delay. If one were to pass a signal through a z−m block, then the
input signal will be time-shifted by m time delays or by mT. While implementing a unit delay in hardware,
one will have to pass the input signal in a memory element for a period of 1T. Then after 1T duration has
elapsed, the value in the memory is released as output and the current input is now stored in the memory.
This way the values that are stored in the memory element and released are delayed by 1T. If one needs to
delay the input signal by 2T, then the output of the memory element is passed through another memory
element to obtain a further 1T delay, so that the over all delay is 2T. This means that for every delay an
Chapter 09.indd 507 3/14/2009 2:54:09 PM

additional memory element is required. Therefore, in a physical sense, z−1 is nothing by one memory element
in the digital processor and z−m means that there are m memory elements in the digital processor. This
implies that if there are m data variables, then m memory locations are needed and the order of the system
is m as each memory location is a state variable. As a memory element is the basic building block of a digital
system, z−1 is the basic building block for modeling a discrete system. With just the knowledge of the
concept of z−1, any discrete system can be modeled.
9.3 Digital Filter
A ll systems in the digital domain are considered as digital filters as they modify the waveshape of the
signal and hence the harmonic amplitude. Therefore, any design in the digital domain is a digital filter
design. There are broadly two distinct types of filters: (a) finite impulse response (FIR) filter and (b) infinite
impulse response (IIR) filter.
Consider the discrete system shown in Figure 9.7. This is built by using only z–1 blocks and gains blocks.
It is called a finite impulse response (FIR) filter.
In Figure 9.7, x(n) is the n-domain input signal and y(n) is the n-domain output signal. The input signal
x(n) is passed through a series of unit delays and some gains. The values b0, b1, …, bm are the scalar gains.
The output is given as
Y (z) = (b0 + b1z–1 + b2z–2 + … + bmz−m)X(z)

H(z) = Y(z)/X(z) = b0 + b1z–1 + b2z–2 + … + bmz−m (9.6)
Equation (9.6) gives the transfer function for the FIR filter. H(z) can also be written as
b0 z m + b1z m −1 + + bm
H (z ) = (9.7)
zm
Equation (9.7) is the transfer function of an FIR filter of order m. The order of filter is the number of unit
delay element needed or the number of poles. From Eq. (9.7) it is evident that for FIR filter all the poles are
at z = 0. This implies that only the location of the zeros can be designed by the choice of the scalar gains
x(n) z −1 z −1 z −1
b0 b1 b2 bm
Σ y(n)
Figure 9.7 FIR filter.
Chapter 09.indd 508 3/14/2009 2:54:09 PM

b0, b1, …, bm. Furthermore, it may be seen from Figure 9.7 that an FIR filter is an open-loop filter. Consider
a first-order FIR filter. Referring to Eq. (9.6), the transfer function is given as
H(z) = b0 + b1z–1
If the input is a unit impulse occurring at 0T, then the output is as shown in Figure 9.8. At time 0T, a unit
impulse is applied to the input of the FIR filter. The output at 0T is the unit impulse scaled by b0. At time
1T the unit impulse has shifted to the b1 line. The output at 1T is b1. Beyond 1T, the output is zero. There
are two important points to be noted. One is that the output response will have the same coefficients as the
transfer function because H(z) which is in the z-domain will become h(n) in the n-domain which is nothing
but the unit impulse response. Second, the output response will exist upto mT time where m is the order of
the FIR filter. This is the reason this type of filter is called finite impulse response filter.
Now, include feedback to the FIR filter shown in Figure 9.7. This is shown in Figure 9.9. The output
y(n) is fed back to a series of unit delays. The outputs of each unit delays are tapped and scaled by scalar
gains a1, a2, …, an as indicated in Figure 9.9. After introducing the feedback, the input–output relation is
Y(z) = (b0 + b1z–1 + b2z–2 + … + bmz−m)X (z) + (a1z–1 + a2z–2 + … + anz−n)Y(z)
Now, the transfer function can be written as
−1 −m
Y ( z ) b0 + b1z + + bm z
H (z ) = = (9.8)
X ( z ) 1 − a1z −1 − − an z − n
Equation (9.8) gives the transfer function of the filter shown in Figure 9.9. This filter is called an IIR filter
as the output will asymptotically die down to zero for a unit input impulse at time 0T. It should be noted
that an FIR filter is a special case of an IIR filter wherein the feedback scalar gains a0 to an are all zero. The
number of poles is always greater than or equal to the number of zeros for a causal system. It is also evident
from Eq. (9.8) that in the case of an IIR filter both the poles and the zeros can be designed leading to much
better cut-off filters. The output response for a unit input impulse is shown in Figure 9.10 which shows that
the output response dies down asymptotically to zero as time tends to infinity.
The order of the filter in the case of the FIR filter was pretty straightforward. The number of unit delay
elements indicated the order of the filter. However, in the case of an IIR filter the question arises as to what
is the order of the filter. Is it the total number of unit delay elements as indicated in Figure 9.9? Or is it the
x
1
0 n
b1
b0 T
0 1 2 3 n
Figure 9.8 Unit impulse response for FIR filter.
Chapter 09.indd 509 3/14/2009 2:54:09 PM

x(n) z −1 z −1 z −1
b0 b1 b2 bm
Σ y(n)
an a2 a1
z −1 z −1 z −1
Figure 9.9 IIR filter.
number of unit delay elements in the forward path? Or is it the number of unit delay elements in the feed-
back path? To answer this question, let us consider a simple IIR filter with one unit delay element in the
forward path and one delay element in the feedback path as shown in Figure 9.11.
Figure 9.11 is drawn in a manner to explicitly indicate the forward and the feedback paths of the filter.
The IIR filter shown in Figure 9.11 is divided into two parts. The portion between nodes 1 and 2 is the
feed-forward portion and is basically an FIR filter. It has a transfer function
V (z )
= b + b z −1
X (z ) 0 1
0 n
y
b1 + b0 a1
b0 (b1 + b0 a1) + (b1 + b0 a1) a2
0 n
Figure 9.10 Impulse response of an IIR filter.
Chapter 09.indd 510 3/14/2009 2:54:09 PM

1
x(n) z −1
Y(z) 1
=
V(z) 1 − a1z −1
b0
b1
Feedback portion
v (n) 3
Σ Σ y(n)
V(z) 2
= b0 + b1 z −1
X(z)
a1
Feed-forward
portion z −1
Y(z) b0 + b1 z −1 1 b0 + b1 z −1
= x =
X(z) 1 1 − a1z −1 1 − a1z −1
Figure 9.11 IIR filter with one delay each in feed-forward and feedback paths.
where v(n) is the signal at node 2. The portion between nodes 2 and 3 is the feedback portion of the IIR
filter. It has a transfer function given by
Y (z ) 1
=
V ( z ) 1 − a1z −1
The overall transfer function is given as
Y ( z ) V ( z ) Y ( z ) ⎛ b0 + b1z ⎞ ⎛ ⎞
−1
1
= ⋅ =⎜ ⎟⋅⎜ ⎟
X ( z ) X ( z ) V ( z ) ⎜⎝ ⎟ ⎜ −1 ⎟
1 ⎠ ⎝ 1 − a1z ⎠
Without loss of generality, the sub-transfer functions could be interchanged as
Y (z ) Y (z ) V (z ) ⎛ 1 ⎞ ⎛ b + b z −1 ⎞
= ⋅ =⎜ ⎟⋅⎜ 0 1 ⎟
X ( z ) V ( z ) X ( z ) ⎜⎝ 1 − a1z −1 ⎟⎠ ⎜⎝ 1 ⎟
⎠
On interchanging the sub-transfer functions, the overall transfer function is unchanged. Thereby in the filter
block also, the two parts could be interchanged without affecting the overall transfer function. This means
the portion of the filter between nodes 2 and 3 will be drawn first and the portion of the filter between
nodes 1 and 2 will be drawn next as shown in Figure 9.12. Here nodes 3 and 1 are joined together and the
signal is denoted by w(n). Here also it is evident that the overall transfer function is unchanged.
The filter shown in Figure 9.12 can be further simplified as shown in Figure 9.13. In Figure 9.13, it is
evident that one of the unit delay elements is redundant because the unit delayed w(n) signal is fed to both
the a1 scalar gain and also to the b1 scalar gain. Therefore, the unit delay element can be made common to
both the arms as indicated in the further simplified diagram Figure 9.13. Thus one sees that there is need for
Chapter 09.indd 511 3/14/2009 2:54:10 PM

3 w (n) 1
x(n) ∑ z −1
b0 b1
a1
z −1 ∑ y(n)
W(z) 1 Y(z)
= = b0 + b1 z −1
X(z) 1− a1z −1 W(z)
Figure 9.12 IIR filter with feed-forward and feedback portions interchanged.
only one delay element and therefore the order of the system is one. This simplified form of the IIR filter is
called the standard form or the canonical form. It can be extended for any order by appending to the first-
order filter by the dotted portions. For a causal system the number of poles is always greater than or equal to
the number of zeros. Therefore, the number of delay elements required to construct the filter without any
redundant unit delay elements is the same as the number of poles. Hence one can say that in the case of an
IIR filter, the order of the system is the number of poles of the IIR filter.
Few Common Digital Filters

Integrator
Let b0 = 0
b1 = 1
a1 = 1
All other scalar gains of Figure 9.9 are zero. With these scalar gain values, the IIR filter reduces to an Euler’s
forward integrator topology, the block schematic of which is given in Figure 9.14. From Eq. (9.8), the
transfer function is given as
Y (z ) z −1 1
H (z ) = = −
=
X (z ) 1 − z 1 z −1
Low-Pass Filter
Let b0 = 0
b1 = 1 − e−aT
a1 = e−aT
where T is the sampling time. All other scalar gains of IIR filter as shown in Figure 9.9 are zero. With the
above values of the scalar gains, one obtains a low-pass filter that is equivalent to 1/(s + a). The block sche-
matic of the low-pass filter is shown in Figure 9.15.
Chapter 09.indd 512 3/14/2009 2:54:10 PM

3 w (n) 1
x(n) ∑ b0
a1
z −1 z −1 b1 ∑ y(n)
Unit delayed w(n)
Further simplifying
w (n)
x(n) ∑ b0 ∑ y(n)
z −1
a1 b1
a2 z −1 b2
Figure 9.13 Canonical form of the IIR filter.
From Eq. (9.8), the transfer function is given as

Y ( z ) (1 − e − aT )z −1 1 − e − aT
H (z ) = = =
X ( z ) 1 − e − aT z −1 z − e − aT
Differentiator
To implement a first-order differentiator, set
b0 = 1/T
b1 = −1/T
Chapter 09.indd 513 3/14/2009 2:54:10 PM

x(n) z −1
+
y(n)
+
z −1
Figure 9.14 Block schematic of the integrator.
x(n) z −1 1−e −aT
+ y(n)
+
e−aT
z −1
Figure 9.15 Block schematic of low-pass filter.
x(n) z −1
1/T −1/T
+ +
y(n)
Figure 9.16 Block schematic of first-order differentiator.
Chapter 09.indd 514 3/14/2009 2:54:11 PM

All other scalar gains of the FIR filter of Figure 9.7 are zero. It should be noted that the differentiator is an
FIR filter that does not have any feedback terms. The schematic of the first-order differentiator is shown in
Figure 9.16.
From Eq. (9.6), the transfer function is given as
Y ( z ) 1 1 −1 1 ⎛ z + 1 ⎞
H (z ) = = + z = ⎜ ⎟
X (z ) T T T⎝ z ⎠
9.4 Mapping between s-Plane and z-Plane

T he physical system exists in the continuous or analog domain. When expressed in the digital domain
with the sampling information incorporated, it would give a better understanding if one is able to relate
critical s-domain parameter in the z-domain. The graphical visualization of the relation between s-plane and
z-plane is discussed in this section.
The z-plane variable “z”, the s-plane variable “s” and the sampling time T are related by the following
exponential relation:
z = e sT (9.9)
For s = s + jw, from Eq. (9.9) one obtains
z = e (σ + jω)T = e σ e jωT (9.10)
Equation (9.10) is of the form re jθ in polar co-ordinates wherein

r = e σ and q = wT
It should be noted that the Cartesian co-ordinate system is used in s-plane and the polar coordinate system
is used in z-plane on account of the mapping rule.
s = 0 + j0 → z = 1
Consider the origin in the s-plane wherein s = 0 + j0. Substituting for the s-variable in Eq. (9.10) one
obtains
z = e 0e j 0T = 1
This means that the origin of the s-plane maps to z = 1 point in the z-plane as shown in Figure 9.17.
w s-plane z-plane
0,0 s 0 1
Figure 9.17 Mapping of s = 0 + j0 point in z-plane.
Chapter 09.indd 515 3/14/2009 2:54:11 PM

s = 0 + jπ/ T → z = –1
Consider the point s = 0 + jp/T in the s-plane. By using Eq. (9.10), one obtains
z = e 0e jπ = 1 ⋅ e jπ = −1
This mapping is shown in Figure 9.18(a).
Consider the points all along the w-axis in the s-plane upto p /T. The s-plane variable is
s = 0 + jap/T, a≤1
This maps in the z-plane as
z = 1 ⋅ e jaπ
This is the locus of points whose vectors have an amplitude of one and angle of ap. For values of a
between 0 and 1, one would trace a semicircle (in anti-clockwise direction) in the z-plane as shown in
Figure 9.18(b).
s = 0 − jπ/ T → z = –1
Consider the point s = 0 − jp /T in the s-plane. By using Eq. (9.10), one obtains
z = e 0e − jπ = 1 ⋅ e − jπ = −1
This also maps to the –1 point as indicated in Figure 9.19(a).
Consider the points all along the w-axis in the s-plane upto –p/T. The s-plane variable is
s = 0 − jap/T, a≤1
This maps in the z-plane as
z = 1 ⋅ e − jaπ
w s-plane z-plane
p
T
0 s −1 0
(a)
w s-plane z-plane
p
T
0 s −1 0 1
(b)
Figure 9.18 Mapping of (a) s = 0 + jp/T; (b) s = 0 + jap/T.
Chapter 09.indd 516 3/14/2009 2:54:11 PM

w s-plane z-plane
0 s −1 0
-p
T
(a)
w s-plane z-plane
0 s −1 0
-p
T
(b)
Figure 9.19 Mapping of (a) s = 0 − jp/T; (b) s = 0 − jap/T.
This is the locus of points whose vectors have an amplitude of 1 and angle of −ap. For values of a between 0
and 1, one would trace a semicircle (in clockwise direction) in the z-plane as shown in Figure 9.19(b). It should
be noted that every 2p/T maps onto the same unit circle in the z-plane. This is the consequence of sampling.
s = –s + j0 → z = Real Axis between 0 and 1

For s = negative real axis, the z-plane mapping is given by
1
z = e −σ e j 0 =
eσ
From the above equation it is evident that σ = 0 maps to z = 1 and σ → ∞ maps to z = 0. Therefore, the
whole of the negative real axis maps to the real axis between 0 and 1 in the z-plane.
Consider the points s = −s + jp/T. The z-plane mapping is given by
−1
z = e −σ e jπ = σ
e
Here, s = 0 + jp/T maps to z = −1 and s = −∞ + jπ / T maps to 0. It can be seen from Figure 9.20 that it
maps to the real axis between 0 and −1.
The entire left half of the s-plane between w = −p/T and w = p/T maps within the unit circle in the z-
plane. This region is called the stable region. This implies that the stable region in the s-plane maps to the area
within the unit circle implying that the points within the unit circle in the z-plane are the stable region.
s = s + j0 → z = Real Axis between 1 and Infinity

For s = positive real axis, the z-plane mapping is given by
z = eσ e j0 = eσ
From the above equation it is evident that σ = 0 maps to z = 1 and σ → ∞ maps to z → ∞ . Therefore,
the whole of the positive real axis maps to the real axis between 1 and infinity in the z-plane. The right half
Chapter 09.indd 517 3/14/2009 2:54:12 PM

w s-plane z-plane
p
T
Stable Region
Stable
0 s −1 1
Region
-p
T
Figure 9.20 Stable region mapping.
w s-plane z-plane
Unstable
p
T
Unstable
0 s 0
Region
-p
T
Region
Figure 9.21 Unstable region mapping.
of the s-plane is called the unstable region. By similar arguments it can be shown that the area outside the
unit circle in the z-plane is the unstable region as indicated in Figure 9.21.
Thus for a discrete system transfer function if the poles and zeros lie inside the unit circle, the system is
stable whereas if they are outside the unit circle, the system is unstable.
9.5 Effect of Sampling
A continuous time signal can be converted into a discrete time signal by sampling at a proper sampling
frequency. The main question that emerges is with regard to the choice of the sampling frequency and
the effect that the sampling will have on the signal.
Chapter 09.indd 518 3/14/2009 2:54:13 PM

According to the Nyquist–Shannon sampling theorem, when sampling an analog signal the sampling
frequency must be greater than twice the highest frequency component of the analog signal so that one is able
to reconstruct the original signal from the sampled version. It should be noted that reconstruction implies
that the original waveshape is recoverable. The original amplitude and the phase may not be recoverable.
Consider a continuous signal x(t). Let this signal be sampled at regular intervals of time T called the
sampling time. The effect on the various view points of the signal in the various domains is illustrated in
Figures 9.22–9.25.
Consider a signal x(t) that has the frequency spectrum as shown in Figure 9.26.
x x
t-domain n-domain
0 t 0 n
Figure 9.22 Effect of sampling in time domain.
⎮X(w)⎮ ⎮X(w)⎮
0 wm w 0 ws 2w s w
wm
Before sampling After sampling
Figure 9.23 Effect of sampling as seen in the frequency domain.
w w s-plane
s-plane 3p/T
2p/T
p /T
2p/T 0
0 s s
− p /T
2p/T
− 3p/T
Figure 9.24 Effect of sampling in the pole-zero domain in the s-plane.
Chapter 09.indd 519 3/14/2009 2:54:13 PM

s-plane z-plane
w
3p/T
p/T
0 s −1 1
− p/T
− 3p /T
Each strip maps into unit circle stacked one upon the other
Figure 9.25 Sampled s-plane mapped on to z-plane.
0 wm w
Figure 9.26 Frequency spectrum of an arbitrary signal x(t).
w
wm wm ws 2w s
w s > 2w m
Figure 9.27 Frequency spectrum of sampled signal x(t) that is sampled with ws > 2wm.
If the Nyquist or Shannon sampling criteria is satisfied during sampling then the frequency spectrum of
the sampled signal is as shown in Figure 9.27.
However, if the Nyquist criteria are not satisfied during sampling then the frequency spectrum of the under-
sampled signal may look like that shown in Figure 9.28. It shows that the sampled spectrums overlap. As a
consequence, at the frequencies where there is an overlap one does not know whether one has to take the ampli-
tude corresponding to the high-frequency component of the lower sampled spectrum or the low-frequency
component of the higher sampled spectrum. This is called aliasing. Once aliasing occurs it is not possible to
reconstruct the signal at all as the amplitude values are not unique in the overlapping frequency regions.
If the sampling is done according to the Nyquist sampling theorem, then the frequency spectrum is
obtained as shown already in Figure 9.27. This does not have the aliasing problem. Therefore if one designs a
low-pass filter called reconstruction filter as shown in Figure 9.29, then all the higher sampled spectrums will
be filtered out and only the fundamental spectrum will remain which is the spectrum of the original signal.
Chapter 09.indd 520 3/14/2009 2:54:13 PM

wm ws w
w s < 2w m
Figure 9.28 Frequency spectrum of sampled signal x(t) that is sampled with ws < 2wm.
Low-pass
A filter
Figure 9.29 Low-pass filter used for extracting or reconstructing the original signal x(t).
In practice, one does not come across a continuous domain signal x(t) that is so well band-limited as
shown in Figure 9.26. Normally due to the presence of noise and unmodeled high-frequency dynamics the
actual band-limit may be much higher. Therefore it will become difficult to know the actual value of wm
and as a consequence it may not be easy to choose the sampling frequency, ws. To overcome this problem,
the continuous domain signal is passed through a filter, the characteristics of which will be such that it will
remove all frequency components beyond a defined frequency value. Thus the output of the filter will have
a well-defined band-limit. This band-limited signal when sampled at a higher rate than the Nyquist rate
will result in proper sampled signal that can be reconstructed. Such a filter that forces a band-limit on the
continuous signal is called the anti-alias filter. Therefore it is recommended that all continuous signals be
passed through an anti-alias filter so that the original continuous signal becomes band-limited before pass-
ing through the sampler.
9.6 Continuous to Discrete Domain Conversion
I t was already discussed in the introduction section of this chapter that one needs to convert the continu-
ous domain model of the system to the discrete domain. In doing so, the sampling information is also
carried along into the discrete domain model. Therefore, if this discrete domain model which contains the
sampling information is used for the analysis and synthesis of the digital controller, the results obtained will
be closer to reality. This section will discuss the various methods for converting the continuous domain
models to discrete domain models. The conversion methods are:
1. pole–zero map;
2. numerical integration methods;
3. hold equivalence;
4. discretizing state equation.
Chapter 09.indd 521 3/14/2009 2:54:14 PM

Pole–Zero Map
Consider a system that has a continuous domain model expressed as a transfer function H(s). It is now
desired that H(s) be converted into an equivalent model in the discrete domain as H(z). The conversion is
done by applying Eq. (9.9) which is repeated here for clarity.
z = e sT
The following procedure is used to obtain the discrete transfer function from the continuous transfer func-
tion by using the pole-zero map given by Eq. (9.9).
Consider a system that has a continuous domain transfer function model given as
( s + a1 )( s + a2 ) ⋅⋅⋅
H (s ) = K (9.10)
( s + b1 )( s + b2 ) ⋅⋅⋅
Then
1. All the poles of H(s) are mapped according to z = e sT . A pole at s = −a1 in the s-plane is mapped to the
−a T
pole z = e 1 in the z-plane.
2. All finite zeros of H(s) are also mapped according to z = e sT . A zero at s = −b1 in the s-plane is mapped
−bT
to the zero z = e 1 in the z-plane.
3. If the transfer function H(s) contains less number of finite zeros than the poles, that is, the order of the
numerator is less than the order of the denominator, then it implies that the number of zeros amount-
ing to the difference in the order between the denominator and the numerator are at ∞ . The zeros at
s = −∞ in the s-plane are mapped to z = e −∞ e − j (π /T )T = −1 .
4. If a unit delay in the discrete domain transfer function is desired, then one zero of H(s) at s = −∞ is not
mapped to the z-plane. This implies that the numerator of H(z) will have one order less as compared to
the denominator of H(z).
5. The gain of H(z) is selected to match the gain of H(s) at a critical point like
H (s ) = H (z )
s =0 z =1
Problem 9.1
Consider the following continuous domain transfer function
a
H (s ) =
s +b
Convert H(s) into the discrete domain.
Solution
Applying the rules 1 and 3 above, one obtains
z +1
H (z ) = K z
z − e −bT
Applying rule 5, one obtains
a 2
H (s ) = = H (z ) = K z
s =0 b z =1 1 − e −bT
Chapter 09.indd 522 3/14/2009 2:54:14 PM

From the above equation,

a ⎛ 1 − e −bT ⎞
Kz = ⎜ ⎟
b⎝ 2 ⎠
Now, substituting for Kz in H(z), one obtains
a ⎛ 1 − e −bT ⎞ ⎛ z + 1 ⎞
H (z ) = ⎜ ⎟⎜ ⎟
b ⎝ 2 ⎠ ⎝ z − e −bT ⎠
Problem 9.2
Consider a system whose continuous domain transfer function is given by
10( s + 1)
H (s ) =
( s + 2)( s + 3)
Solution
Applying rules 1, 2 and 3, one obtains
( z − e −T )( z + 1)
H (z ) = K z
( z − e −2T )( z − e −3T )
Applying rule 5, one obtains
10 (1 − e −T )2
H (s ) = = H (z ) = K z
s =0 6 z =1 (1 − e −2T )(1 − e −3T )
From the above equation,
5 (1 − e −2T )(1 − e −3T )
Kz =
6 (1 − e −T )
Substituting for Kz in H(z), one obtains
5 (1 − e −2T )(1 − e −3T ) ( z − e −T )( z + 1)

H (z ) =
6 (1 − e −T ) ( z − e −2T )( z − e −3T )
Numerical Integration Methods

Integration is a central idea in physical systems. The state equation models and the transfer function models
all basically indicate the integration concept. The state variables are obtained by integration and the differen-
tial equations can be solved and simulated by using only the integrator block. This implies that any physical
system may be modeled with just integrator blocks. Therefore, one needs to convert the continuous integra-
tor model into the discrete domain. By applying the conversion to all the integrators in the system model,
the discrete domain model is obtained. Thus, 1/s which is the transfer function of the integrator in the
s-domain should be equivalently represented in the z-domain. Then the inverse of the discrete domain inte-
grator is used to replace the “s” variable in the s-domain transfer function to obtain the z-domain transfer.
Chapter 09.indd 523 3/14/2009 2:54:14 PM

Consider a continuous signal x(t) as shown in Figure 9.30. Passing this signal through an integrator
means the area under the signal x(t) is the output of the integrator.
Figure 9.31 gives the continuous signal x(t) and the concept for finding the integration value of x(t) by
finding the area under the curve in the discrete domain.
Let y(n) be the variable associated with the integrator output and x(n) be the sampled version of the
continuous signal x(t) that is applied to the integrator input. Then at any instant k, y(kT ) or y(k) contains
the area under the curve. Note that y(k − 1) is the area under the curve till time instant (k − 1)T. Therefore,
the area under the curve may be written as
Area under curve till kT = Area under curve till (k − 1)T + Area under curve between time kT and (k − 1)T
This can be written as
y (k ) = y (k − 1) + ( Area update ) (9.11)
Equation (9.11) is the central theme that is used in all integration algorithms in the digital domain. This is
illustrated in the block schematic of Figure 9.32. The block schematic of the discrete integrator shown in
x(t )
0 t
Figure 9.30 A continuous signal x(t) and the shaded region is the area under the curve.
Previous area Area update
x(k − 1)
x(k)
0 k−1 k t, n
T
Figure 9.31 Finding the area under the curve in the discrete domain.
Chapter 09.indd 524 3/14/2009 2:54:15 PM

Area
update + y(k)
x(k) +
calculator
y (k −1) Previous
area
z −1
Figure 9.32 Generic discrete integration block schematic.
Figure 9.32 is a very generic algorithm. The various integration algorithms in fact differ in the area update
calculation method. Based on the generic block schematic of Figure 9.32, three algorithms, namely, Euler’s
forward rectangular rule, Euler’s backward rectangular rule and the Tustin’s rule, will be discussed in this
section.
Euler’s Forward Rectangular Rule

In this method, the update area is calculated by approximating the area of a rectangle of width T and height
equal to the signal value at the previous instant of time, that is, x(k − 1). The height of the update rectangle
is taken as x(k − 1) which is brought forward to the instant kT, hence the name forward rectangular rule.
The update area is given by x(k − 1) × T. The method is illustrated in Figure 9.33.
Referring to the generic block schematic of Figure 9.32 and the Euler’s forward rectangular illustration
shown in Figure 9.33, one obtains
y (k ) = y (k − 1) + x (k − 1)T
This can be expressed in the z-domain as
Y ( z ) = z −1Y ( z ) + z −1TX ( z )
Update area = x (k − 1) T
x (k − 1)
Previous area
0 k −1 k t, n
T
Figure 9.33 Illustration of Euler’s forward rectangular rule.
Chapter 09.indd 525 3/14/2009 2:54:16 PM

z −1 T +
x(k) x(k −1) + y(k)
z −1
y (k −1)
Figure 9.34 Block schematic of Euler’s forward rectangular rule.
From the above equation, the transfer function of the Euler’s forward rectangular integrator is given as
Y (z ) T
H (z ) = = (9.12)
X ( z ) z −1
The block schematic of the Euler’s forward rectangular rule is shown in Figure 9.34. It is an IIR filter with
b0 = 0, b1 = T and a1 = 1.
Referring to Figure 9.17, it was shown that s = 0 in the s-domain maps to z = 1 in the z-domain. In the
s-domain the integrator transfer function is 1/s which implies a pole at s = 0 in the s-domain. Note that Eq.
(9.12) has a pole at z = 1 in the z-domain which is expected as per the s-domain to z-domain mapping. It is
important to note that all forms of integrator in the discrete domain will have the pole at z = 1. It is only the
scalar gain and the zero location that will vary for different algorithms.
For the Euler’s forward rectangular rule, the s-domain and z-domain integration equivalence is
1 T z −1
↔ or s ↔ (9.13)
s z −1 T
From the above equivalence, the “s” variable is replace by (z − 1)/T in the continuous domain transfer func-
tion H(s) to obtain the discrete domain transfer function H(z). Therefore,
H (z ) = H (s ) (9.14)
z −1
s=
T
Equation (9.14) is read as: H(z) equals H(s) wherein the variable “s” is replaced by (z − 1)/T in the continu-
ous domain transfer function.
Problem 9.3
Consider a continuous domain transfer function H(s) given as
a
H (s ) =
s +b
Solution
Applying Eq. (9.14), one obtains
a aT
H (z ) = =
⎛ z −1⎞ z − 1 + bT
⎜ ⎟+b
⎝ T ⎠
Chapter 09.indd 526 3/14/2009 2:54:17 PM

Euler’s Backward Rectangular Rule

In this method, the update area is calculated by approximating the area of a rectangle of width T and height
equal to the signal value at the present instant of time, that is, x(k). The height of the update rectangle is
taken as x(k) which is taken back to the instant (k − 1)T, hence the name backward rectangular rule. The
update area is given by x(k)T. The method is illustrated in Figure 9.35.
Referring to the generic block schematic of Figure 9.32 and the Euler’s backward rectangular illustration
shown in Figure 9.35, one obtains
y (k ) = y (k − 1) + x (k )T
Y ( z ) = z −1Y ( z ) + TX ( z )
From the above equation, the transfer function of the Euler’s backward rectangular integrator is given as
Y ( z ) Tz
H (z ) = = (9.15)
X ( z ) z −1
The block schematic of the Euler’s backward rectangular rule is shown in Figure 9.36. It is an IIR filter with
b0 = T, b1 = 0 and a1 = 1.
x Update area = x(k)T
x(k)
Previous
area
0 k −1 k t, n
T
Figure 9.35 Illustration of Euler’s backward rectangular rule.
T +
x(k) + y(k)
y(k−1)
z−1
Figure 9.36 Block schematic of Euler’s backward rectangular rule.
Chapter 09.indd 527 3/14/2009 2:54:17 PM

Here also, note that Eq. (9.15) has a pole at z = 1 in the z-domain which is expected as per the s-domain to
z-domain mapping. For the Euler’s backward rectangular rule, the s-domain and z-domain integration
equivalence is
1 Tz z −1
↔ or s ↔ (9.16)
s z −1 Tz
From the above equivalence, the “s” variable is replace by (z − 1)/Tz in the continuous domain transfer func-
tion H(s) to obtain the discrete domain transfer function H(z). Therefore,
H (z ) = H (s ) (9.17)
z −1
s=
Tz
Problem 9.4
Consider the continuous domain transfer H(s) given as
a
H (s ) =
s +b
Solution
⎛ ⎞
a aTz ⎛ aT ⎞ ⎜ z ⎟
H (z ) = = =⎜ ⎟⎜ ⎟
⎛ z −1⎞ z − 1 + bTz ⎝ 1 + bT ⎠ ⎜ z − 1 ⎟
⎜ ⎟+b ⎝ 1 + bT ⎠
⎝ Tz ⎠
Tustin’s Trapezoidal Rule

In this method, the update area is calculated by approximating the area of a trapezium as illustrated in
Figure 9.37. Referring again to the generic block schematic of Figure 9.32 and the Tustin’s trapezoidal rule
illustrated in Figure 9.37, one obtains
⎛ x (k − 1) + x (k ) ⎞
y (k ) = y (k − 1) + ⎜ ⎟T
⎝ 2 ⎠
T T
Y ( z ) = z −1Y ( z ) + z −1
X (z ) + X (z )
2 2
From the above equation, the transfer function of the Tustin’s trapezoidal integrator is given as
Y (z ) T ⎛ z + 1 ⎞
H (z ) = = ⎜ ⎟ (9.18)
X (z ) 2 ⎝ z − 1 ⎠
Equation (9.18) is also called the Tustin’s form or the trapezoidal integrator or bilinear transformation in the
literature.
Chapter 09.indd 528 3/14/2009 2:54:17 PM

x x(k − 1) + x(k)
Update area = T
x (k −1) 2
Previous
area
0 k−1 k t, n
T
Figure 9.37 Illustration of Tustin’s trapezoidal rule.
The block schematic of the Tustin’s trapezoidal rule is shown in Figure 9.38. It is an IIR filter with b0 =
T/2, b1 = T/2 and a1 = 1.
Here again note that Eq. (9.18) has a pole at z = 1 in the z-domain which is expected as per the
s-domain to z-domain mapping. For the Tustin’s trapezoidal rule, the s-domain and z-domain integration
equivalence is
1 T ( z + 1) 2 ( z − 1)
↔ or s ↔ (9.19)
s 2 ( z − 1) T ( z + 1)
From the above equivalence, the “s” variable is replaced by 2(z − 1)/[T(z + 1)] in the continuous domain
transfer function H(s) to obtain the discrete domain transfer function H(z).
x(k) x(k−1)
z−1
T/2 T/2
+ y (k)
+
+
y(k−1)
z−1
Figure 9.38 Block schematic of Tustin’s trapezoidal rule.
Chapter 09.indd 529 3/14/2009 2:54:18 PM

Therefore,
H (z ) = H (s ) (9.20)
2 ( z −1)
s=
T ( z +1)
Problem 9.5
Consider the continuous domain transfer H(s) given as
a
H (s ) =
s +b
Solution
a ⎛ aT ⎞ z +1
H (z ) = =⎜ ⎟
2 ⎛ z −1⎞ ⎝ 2 + bT ⎠ z − ⎛ 2 − bT ⎞
⎜ ⎟+b ⎜ ⎟
T ⎝ z +1⎠ ⎝ 2 + bT ⎠
Hold Equivalence
This is one of the most popular methods to convert continuous domain transfer function to the discrete
domain equivalent as it is very similar to the practical scheme. Consider a system whose continuous domain
transfer function is H(s). It receives as input a continuous signal x(t) and delivers as its output a continuous
signal y(t) as shown in Figure 9.39(a). The same system, when represented in the discrete domain as a transfer
function H(z), should receive a discrete signal x(k) which is the sampled version of x(t) and deliver as its
output a discrete signal y(k) which should be the sampled version of y(t) as indicated in Figure 9.39(b). Then
one can say that H(z) is the discrete equivalent of H(s). To convert the continuous system to a discrete system,
one must include a DAC at the input of H(s) and an ADC at the output of H(s) as shown in Figure 9.39. This
way the digital input x(k) is converted to analog signal which is applied as input to the continuous system.
The output of the continuous system is a continuous domain signal which is discretized by passing it through
an ADC to deliver the discrete output signal y(k).
The DAC in Figure 9.39(b) is called the hold device and the ADC is called the sampler. Therefore, the
discrete equivalent model of the system can be re-drawn as shown in Figure 9.40. The hold, the continuous
system and the sampler together form the discrete equivalence of the continuous system. This is called the hold
equivalence. The DAC output is held constant till the next sampled value. This type of hold is called the zero-
order hold. Hence this type of equivalence is also called the zero-order hold equivalence.
The hold device accepts a sample x(k) at t = kT and holds its output constant until the next sample is received
at t = (k + 1)T. The piece-wise constant output of the hold device is the signal that is applied to the system.
Consider a unit impulse being applied to the input of the hold device at instant kT. The output of the
hold device is a pulse of amplitude 1 and width T as shown in Figure 9.41. The output of the hold device can
be considered to be the difference between a step at kT and a delayed step at (k + 1)T. The Laplace transform
of a unit step is given as 1/s. Laplace transform of the step response of the system is H(s)/s and the Laplace transform
of the delayed step response of the system is [H(s)/s]e–st as a unit delay z–1 in the discrete domain is e–sT in the
s-domain from Eq. (9.9).
Chapter 09.indd 530 3/14/2009 2:54:18 PM

H(s)
x(t ) y(t )
(a)
H(z)
x(k) y(k)
x(k) DAC H(s) A/D y(k)
(b)
Figure 9.39 (a) Continuous domain transfer function; (b) discrete domain transfer
function equivalence.
As the system is linear, superposing both these responses, the response of the system to a unit impulse at
the input of the hold is given by
H ( s ) H ( s ) − sT H (s )
Y (s ) = − e = (1 − e − sT ) (9.21)
s s s
The response of the system y(t) is the Laplace inverse of Y(s). The y(t) is passed through the sampler to
be delivered as y(k) which is the discrete impulse response of the “hold equivalence” system shown in
Figure 9.40.
Hold H(s) Sampler

x(k) y(t ) y(k)
H(z)
Figure 9.40 Discrete transfer function by hold equivalence.
Chapter 09.indd 531 3/14/2009 2:54:18 PM

0 k k +1 t, n
T
Hold
output
1
0 t, n
This is Unit step at k T = u (n )

equivalent
to
Above unit step delayed = u(n − 1)

u(n ) − u (n − 1)
Figure 9.41 Unit impulse response of the hold device.

Thus,
H ( z ) = {Y ( s )} (9.22)
where is the z-transform operator. Therefore, from Eqs. (9.21) and (9.22), one obtains
{
H ( z ) = (1 − e − sT )
H (s )
s }
= (1 − z −1 ) ⋅ { }
H (s )
s
(9.23)
Equation (9.23) gives the zero-order hold discrete transfer function equivalence of the continuous system
H(s). This is also simply referred to as ZOH equivalence. The {H(s)/s} portion of H(z) is found by refer-
ring to the z-transform tables given in Appendix IV. However, the ZOH equivalence is difficult to convert
manually. It is better done using a computer. MATLABTM and SciLAB provide functions to perform the
continuous to discrete domain conversions.
Problem 9.6
Consider the continuous transfer function
a
H (s ) =
s +b
Solution
Given
a
H (s ) =
s +b
Chapter 09.indd 532 3/14/2009 2:54:18 PM

Therefore
H (s ) a a ⎛1 1 ⎞
= = ⎜ − ⎟
s s(s + b ) b ⎝ s s + b ⎠
Referring to the z-transform tables in Appendix IV,
{ }
H (s )
s
a⎛ z
= ⎜ −
z
b ⎝ z −1 z − e
⎞
−bT ⎟
=
az (1 − e −bT )
⎠ b( z − 1)( z − e −bT )
Applying the ZOH equivalence of Eq. (9.23), one obtains

a ⎛ 1 − e −bT ⎞
H (z ) = ⎜ ⎟⎟
b ⎜⎝ z − e −bT ⎠
Discretizing the State Equation

As discussed in Chapter 8, a continuous system can be represented by a set of first-order linear differential
equations. This set can be expressed in a standard matrix form called the state equation as given in Eq. (9.24)
and the output equation as given in Eq. (9.25):
x = A ⋅ x + B ⋅ u (9.24)
y = C ⋅ x + D ⋅u (9.25)
where x is the vector of state variables, u is the vector of input variables and y is the vector of output vari-
ables. A is called the state matrix, B the input weighting matrix, C the output matrix and D the input–output
direct feed-forward matrix.
The continuous state equation representation as given in Eqs. (9.24) and (9.25) needs to be converted
to the discrete state equation representation which is of the form
x (k + 1) = Ad x (k ) + Bd u(k ) (9.26)
y (k ) = C d x (k ) + Dd u(k ) (9.27)
Equation (9.26) is the discrete state equation representation and Eq. (9.27) is the discrete output equation.
The continuous domain state equation is a set of linear differential equations whereas the discrete domain
state equation is a set of difference equations. Equation (9.24) gives the slope or the gradient of the state
variables based on which the future evolution of the state variables can be predicted whereas the discrete
state equation [Eq. (9.26)] gives the prediction of the next value of the state variables. The output equations
[Eqs. (9.25) and (9.27)] are identical in both the cases as the output is the linear combination of the state
and input variables which are the same whether in continuous domain or in discrete domain. Therefore,
C d = C and Dd = D (9.28)
It is now required to obtain a conversion relation for Ad and Bd in terms of the continuous domain matrices
and sampling time T. Consider the set of linear differential equation [Eq. (9.24)]. The solution of this gives
the state vector. The solution of Eq. (9.24) has two parts: (a) the complementary function which is the gen-
eral solution and (b) the particular integral which is the particular solution to a specific input.
Chapter 09.indd 533 3/14/2009 2:54:19 PM

The complementary function solution of Eq. (9.24) is
e A(t −to ) x (t o )
where to is the starting time and x(to) is the initial state value and the particular integral or the particular
solution is given by
t
A (t −τ )
∫e Bu(τ )dτ
to
The total solution is the sum of the complementary function and the particular solution. Thus,
t
x (t ) = e A(t −to ) x (t o ) + ∫ e A(t −τ ) Bu(τ )dτ (9.29)
to
Let x(t) be passed through a sampler of sampling period T. Consider two samples of x(t), one at kT which is
the starting time representing to in Eq. (9.29) and the other at t = (k + 1)T = kT + T, which is the value one
sample period later. Thus, the sampled version of Eq. (9.29) becomes
kT +T
x (kT + T ) = e AT x (kT ) + ∫ e A( kT +T −τ ) Bu(τ )dτ (9.30)
kT
If a zero-order hold is used for sampling and holding the input signal u(t), then
u(τ ) = u(kT ), kT ≤ τ < kT + T
Here u(kT ) will be a constant during the period kT to (k + 1)T due to the use of zero-order hold. Now
change the integration variable in Eq. (9.30) from t to t such that
t = kT + T − t
The discrete solution given in Eq. (9.30) becomes
⎛T ⎞
x (kT + T ) = e AT x (kT ) + ⎜ ∫ e At dtB ⎟ u(kT ) (9.31)
⎜ ⎟
⎝ 0 ⎠
Comparing Eq. (9.31) with the discrete state equation standard form given in Eq. (9.26), one obtains
Ad = e AT (9.32)
T
Bd = ∫ e At dtB (9.33)
0
and from Eq. (9.28), one has

Cd = C (9.34)
Dd = D (9.35)
Equations (9.32)–(9.35) give the relationship to convert the continuous state space model to the discrete
state space model. These conversion equations are tedious to do by hand. They are best done with the help
of computer using software tools like MATLABTM or SciLAB.
Chapter 09.indd 534 3/14/2009 2:54:19 PM

9.7 Control System Basics
T his section discusses a few common concepts and terminologies that will be useful in understanding the
control concepts that will be used in the succeeding chapters of this book.
Performance Specifications
Specifications are essential to quantify the performance of the control system. These are called the perfor-
mance specifications. Analysis and design are made with respect to these performance specifications. A con-
trol system may be qualified with the following three performance parameters:
1. Stability;
2. Steady-state accuracy;
3. Speed of response.
These are called the SSS performance parameters. The control system should be specified with respect to the
above SSS performance parameters. The quantification of these parameters is called the performance
specifications.
Stability
Consider a system which is applied by a step signal at the input. Due to the applied step disturbance at
the input, three possibilities can happen to the output response of the system. (a) The step input will
disturb the output response in such way that the output signal will grow and go unbounded as shown in
Figure 9.42(a). Such a system is called unstable. (b) The step input will disturb the output response and
output will oscillate without decaying as shown in Figure 9.42(b). Such a system is called marginally
stable. (c) The step input will disturb the output response; however, the output disturbance will asymp-
totically decay to zero as shown in Figure 9.42(c). Such a system is called stable or asymptotically
stable.
Figure 9.42 gives the time domain view point about the stability of the system. The pole-zero domain
view point is given in Figure 9.43. Consider the pole-zero domain or the s-domain shown in Figure 9.43. In
Figure 9.43(a), a pole of a system is on the real positive axis at “a”. This has a transfer function 1/(s − a). The
time response of this transfer function is e at. As a > 0 and t > 0, the response of the system continuously
increases with time, that is, the response goes unbounded. Such a system is called an unstable system. All the
A
A Output A
Output Output
Input Input Input
0 t 0 t 0 t
(a) (b) (c)
Figure 9.42 Time domain view point of stability: (a) Unstable system; (b) marginally stable
system; (c) asymptotically stable system.
Chapter 09.indd 535 3/14/2009 2:54:19 PM

w w
w
0 a s 0 s −a 0 s
1
1 s 1
s −a s +a
1 1 1
L−1 s − a = e at L−1 s =1 L−1 s + a = e −at
(a) (b) (c)
Figure 9.43 Pole-zero domain view point of stability: (a) Unstable system; (b) marginally stable
system; (c) stable system.
poles that are on the right half of the s-plane will give a system response that goes unbounded. Therefore, all
the systems that have pole or poles located on the right half of the s-plane are considered as unstable systems.
By Eq. (9.9), this implies that all discrete systems having poles located outside the unit circle in the z-plane
are unstable systems.
In Figure 9.43(b), the pole of the system is located at s = 0. This is an integrator. Here the response is not
a function of time. Here it neither increases with time nor decreases with time. Even in the integrator it is
evident that on giving an input pulse and removing the pulse, the integrator output will neither decay nor
increase with time. Such a system is called marginally stable. All systems that have poles only on the w -axis of
the s-plane will have a response that will neither increase nor decrease with time. They may oscillate at the
frequency where the pole is located on the w -axis of the s-plane, but the amplitude of the oscillations will
neither increase nor decrease. By Eq. (9.9), this implies that all discrete systems having poles located on the
unit circle in the z-plane are marginally stable systems.
In Figure 9.43(c), the pole of the system is located at s = –a. Here the response is given as e–at. This
implies that the system response decays with time which implies that if any disturbance occurs, the distur-
bance in the response will exponentially or asymptotically decay to zero with time. Such a system is a stable
system. All systems that have their poles located in the left half of the s-plane only will possess a decaying
mechanism and hence are called the stable systems. By Eq. (9.9), this implies that all discrete systems having
poles located within the unit circle in the z-plane are stable systems.
Steady-State Accuracy
The system response when steady state has been reached is an important performance criterion. The system
will in fact spend a major portion of its operating time in the steady state; therefore the error that is present
in this state will give a measure of the quality of the system. Consider the closed-loop system shown in
Figure 9.44. The system has a continuous transfer function in the s-domain, G(s). The output is fed back
with unity gain. The reference is compared with the fed back signal, that is, the output to obtain the error
signal which is applied to the input of the system.
Chapter 09.indd 536 3/14/2009 2:54:19 PM

System
R(s ) E(s ) Y(s )
+ G (s )
Reference − Error Output
Figure 9.44 Closed-loop system for steady-state error analysis.
Let G(s) be expressed in a generalized manner in the pole-zero form as

K (1 + Ta s )(1 + Tb s )...
G(s ) = (9.36)
s m (1 + T1s )(1 + T2 s )...
The error in the s-domain E(s) is given by
R(s )
E (s ) = (9.37)
1 + G(s )
The number “m” of pure integrations determines the type of the system. If m is zero, the system is a type 0
system; if m is 1 then the system is a type 1 system and so forth.
The standard test inputs for a system are steps, ramps and parabolic-shaped signals. Steady-state errors
due to these standard test inputs are determined in terms of constants called the error constants. These error
constants give a measure of the quality of the system’s steady-state performance. For any system with a gen-
eralized transfer function G(s) as given in Eq. (9.36), the error constants are defined as follows:
Positional error constant
K p lim G ( s ) (9.38)
s →0
For the discrete domain system it is
K p lim G ( z )
z →1
Velocity error constant
K v lim[ sG ( s )] (9.39)
s →0
K v lim[(1 − z −1 )G ( z )]
z →1
Acceleration error constant
K a lim[ s 2G ( s )] (9.40)
s →0
K a lim[(1 − z −1 )2 G ( z )]
z →1
Chapter 09.indd 537 3/14/2009 2:54:20 PM

Table 9.3 Steady-state errors for various system types
Unit step, R(s) = 1/s Ramp, Parabolic, R(s) = 1/s 3

R(s) = 1/s 2
1 1 1
System type e ss = e ss = e ss =
1+ K p Kv Ka
0 1/(1 + K ) ∞ ∞
1 0 1/K ∞
2 0 0 1/K
m≥3 0 0 0
By using the final value theorem on E(s) as given in Eq. (9.37), the steady-state error ess is given by
⎡ sR ( s ) ⎤
e ss = lim e (t ) = lim[ sE ( s )] = lim ⎢ ⎥ (9.41)
t →∞ s →0 s →0 ⎣1 + G ( s ) ⎦
For the discrete domain system
− z −1 )R ( z ) ⎤
⎡ (1−
e ss = lim e (t ) = lim[(1 − z −1 )E ( z )] = lim ⎢ ⎥
t →∞ z →1 z →1 ⎢ 1 + G ( z ) ⎥⎦
⎣
Table 9.3 summarizes the steady-state errors for systems of various types and various standard inputs.
From Table 9.3 it may be seen that for a type 0 system, the steady-state error is finite for a unit step
input. For a type 1 and above system, the steady-state error for a unit step input is zero whatever may be
the value of the gain K. The steady-state error constants have similar meaning for discrete domain sys-
tems too.
Speed of Response
The speed of response of a system is the quickness with which it can reach the steady state on the occurrence
of disturbances or input changes. This is generally quantified in terms of the rise time of the output signal.
When a unit step is given to the input of the system, the output varies from the initial to the final settling
value. The time taken for the output to change from 10% of the final steady-state value to 90% of the final
steady-state value is a measure of the speed of response in the time domain. In the frequency domain, the
bandwidth of the system gives a measure of the speed of response of the system. If the speed of response is
high then the bandwidth in the frequency domain is high and vice versa.
Measures for Performance Parameters

Consider a system with reference input r(t) and output response y(t). The input is a unit step and the typical
output signal response to a unit step input is shown in Figure 9.45. Many performance parameters are iden-
tified in the output response shown in Figure 9.45.
Chapter 09.indd 538 3/14/2009 2:54:20 PM

1. Stability Measure:
y max − y final
Po = × 100
y final
where Po is the peak overshoot. This parameter gives a measure of the relative stability. If the value of Po
is low then the relative stability is high and vice versa.
2. Steady-State Accuracy Measures: eB is called the error band when the system has settled down or
reached its steady state. eB gives a measure of the steady-state accuracy.
3. Speed of Response Measures: Figure 9.45 shows many parameters that are measures of speed of response.
The time taken by the output response to stay within the error band eB is called the settling time (ts). The time
taken by the output response to rise from 10% to 90% of the final value is called the rise time (tr) of the
output response. The time taken by the output response to reach 50% of its final value is called the delay
time (td). Note that ts, tr and td are measures of speed of response. Also, tr and td give the speed of the initial
dynamics. However, if one is interested in the dynamics upto steady state then ts is a better measure.
Consider a system having the s-domain closed-loop transfer function given by G(s) or a z-domain closed-
loop transfer function given by G(z). The transfer function in the w -domain is denoted as G( jw). The
closed-loop frequency domain viewpoint of the output response is shown in Figure 9.46. Here again, one
can visualize the SSS performance measures using frequency domain parameters.
1. Stability Measure: The maximum value of the magnitude of the closed-loop transfer function in the
w -domain Mp is a measure of the relative stability of the system. Higher the value of Mp, lower is the
relative stability of the system.
2. Steady-State Accuracy Measure: The magnitude value of the w -domain transfer function G ( jw) at
w = 0 gives the closed-loop gain at steady state. If steady-state error needs to be zero, the magnitude of
G( jw) at w = 0 should be unity. Only then will the output be same as the input. The deviation of G ( jw)
from unity at w = 0 gives a measure of the steady-state accuracy of the closed-loop response.
r, y
Po y (t )
eB
r (t )
1.0
0.9
0.5
0.1
0 t
tr
td
ts
Figure 9.45 Time domain performance measures for unit step standard input.
Chapter 09.indd 539 3/14/2009 2:54:20 PM

Y( jw )
G ( jw) = Steady-state error
R( jw )
Mp
1.0
Closed-loop
frequency response
0.707
wp wb w
Figure 9.46 Frequency domain visualization of performance measures of a closed-loop system.
3. Speed of Response Measure: w b gives the bandwidth of the system. At w = w b, the magnitude of
G( jw) has a value of 0.707. This gives upper bound on the high-frequency components in the time
domain signal and hence a measure of the rise time. Therefore w b can be considered to be a measure of
the speed of response of the system.
9.8 Control Principles
T his section will give a brief discussion on the basic control principles that will enable one to understand
the concepts to be discussed in the forthcoming chapters. Consider the closed-loop system shown in
Figure 9.47.
In Figure 9.47, the plant is controlled by a controller that is represented as a gain K(w) which is a func-
tion of frequency w. The core function of this controller K is to ensure that the error e is always zero. By
always, it is meant that both during steady-state conditions and during transient conditions. The controller
K generates a signal Vc such that error e is always kept at 0. Now the question to be answered is: At what
value of K is e always 0?
Vref V0
Plant to be
+ K (w)
− e Vc controlled
Controller
Vfb
Figure 9.47 Block schematic of a closed-loop system.
Chapter 09.indd 540 3/14/2009 2:54:20 PM

It is evident from Figure 9.47 that

Vc
e= (9.42)
K
From Eq. (9.42) it is evident that e is always 0 when
1. Vc is always zero or
2. K is always infinity ( ∞ ).
Condition (1) above implies open-loop operation. As Vc is always zero there is no control input to the plant.
As a consequence, one loses control of the plant and the plant operates in open-loop condition. Condition (2)
is the acceptable condition. If K is made ∞ , then Vc is adjusted in such a manner that error e is always 0
under all conditions of disturbances and plant parameter variations. For any given system just choose K as
infinite; then the error is always zero. K can be made infinite practically by using an op-amp as a comparator.
This would be an ideal situation. However, this is not possible due to the following two main reasons:
1. The value of Vc is limited.
2. The noise in the system also gets amplified by K.
As all controllers and controlled systems need electronics, they need to be powered by power supplies that
are either 3.3 V or 5 V or ±15 V. Therefore Vc cannot increase more than the power supply voltage. Hence,
even if K is made infinite, Vc can take only finite values and therefore, if a value of the product of e and K is
greater than the power supply voltage, then Vc can at most clamp to power supply voltage. This implies that
Vc will not change for any change in e and therefore there will be loss of control or in other words, the
system will be operating in open loop. The error band within which there is a change in Vc for a change in
e is called the control band. Close-loop operation exists only if the operation is within the control band.
Figure 9.48 shows the control band. From Eq. (9.42), it follows that
Vc
K = (9.43)
e
The plot of error e versus control voltage Vc is shown in Figure 9.48. The slope of the e −Vc curve is the con-
troller gain K at that point. Consider that the gain K is constant over the error range. Then e and Vc will have
a linear relationship with slope equal to K. Referring to Figure 9.48, consider that Vc can swing only between
−15 V and +15 V. When the gain is K1, the error e could swing between –1 and 1. If the error goes beyond
±1, then Vc will saturate to ±15. This range of error values where Vc does not saturate is called the control
band. At higher gains, the control band decreases. Evidently, at K = ∞ , the control band is 0. This means
that if there is even a finite error due to disturbance or parameter changes, the control signal Vc will saturate.
If Vc saturates, then the loop is broken and behaves like open loop due to loss of control on plant. The error
would swing + ve and Vc goes to + ve saturation. This will lead to –ve error due to plant output swinging to
a larger positive value. When the error goes negative, Vc will swing to –ve saturation and so on. This phe-
nomenon is called limit cycle oscillation. This occurs due to practical limitations on the value of Vc. If Vc is
capable of taking on values from − ∞ to + ∞ then one can achieve ideally zero error condition always.
The second practical problem is that of noise. In any electronic circuit, every component or device will
generate noise like thermal noise, shot noise and other noises associated with semiconductor components.
Apart from these there will also be noise and disturbance that will be induced from neighboring equipments
and also through the power supply. Therefore, in any practical circuit, noise is prevalent. This noise will also get
multiplied by K and appear at Vc. If K is infinite then Vc primarily contains only noise. As noise is predominant
at higher frequencies, K should be made infinite at lower frequencies and very low at higher frequencies so that
Chapter 09.indd 541 3/14/2009 2:54:20 PM

Vc
K1
K2
+15 K1 > K2
−1 0 1
e
−15
Open-loop Open-loop
operation Control operation
band
Close-loop
operation
Figure 9.48 Control band.
noise gets attenuated. Therefore, K should be a function of frequency w. However, note that if K is made small
at high frequencies then the bandwidth is limited which makes the speed of response lower. Thus the whole
gamut of different control system techniques is basically for analyzing this gain K(w) and to shape this
frequency-dependent gain such that noise is attenuated but the performance specifications are met.
9.9 State Space Method
T raditionally, the classical control theory was employed for the design of the controllers of single-input
single-output (SISO) systems. The classical control notions are generally in the frequency domain and
the pole-zero domain. Relying on the transform methods, it is primarily applicable for linear time-invariant
(LTI) systems. However, an exact description of the internal system dynamics is not needed for the classical
design. Only the input–output behaviors of the system are of importance. A real system has disturbances
and measurement noise that may not be possible to describe exactly by a mathematical model. Classical
theory is natural for designing control systems that are robust to such disorders, yielding good closed-loop
performance even though the controllers are not optimal with respect to a specific performance index. By
robustness, it is meant that the system retains its performance and stability even in the face of disturbances
and parameter variations in the system.
However, the classical control theory is difficult to apply in multiple-input and multiple-output (MIMO)
systems like the case of a high performance induction motor vector-controlled drive system. Due to the
interaction of the control loops in a multivariable system, each SISO transfer function can have acceptable
properties in terms of step response and robustness, but the coordinated control motion of the system can
fail to be acceptable. Therefore, MIMO design using classical techniques requires painstaking effort using
Chapter 09.indd 542 3/14/2009 2:54:21 PM

the approach of closing one loop at a time. This is a trial and error procedure that may require multiple itera-
tions and it does not guarantee good results or even closed-loop stability.
On the other hand, by using state space control techniques, many of the limitations of the classical con-
trols for multi-variable feedback control systems can be overcome. State space control design is fundamen-
tally a time-domain technique. A state space model of the system to be controlled is required. The state
space model can represent both MIMO as well as SISO systems. State space control techniques were first
firmly established for linear systems and then later extended to non-linear systems. Topics like optimal and
robust control are also time domain methods that essentially depend on the state space techniques.
Some remarks can be made on the features of the state space approach: (a) As long as the system is con-
trollable, the feedback gains guarantee the stability of the closed-loop system. (b) This technique is easy to
apply even for multiple-input plants. (c) The approach relies on the solution of the matrix design equations
which are unsuited for hand calculations. As a consequence, computer-aided design is an essential feature.
(d) All the controller gains of all the loops are determined at the same time by using the matrix design equa-
tions. Thus all the loops of a MIMO system are closed simultaneously which is in complete contrast to the
one loop at a time procedure of classical controls design.
If one were to consider for example optimal controller like the linear quadratic regulator (LQR). Linear
quadratic design using state feedback guarantees closed-loop stability. However, all the state components are
seldom available for feedback purposes in a practical design problem. As a consequence, the non-available
state components are estimated from the available states and the system model, using a reduced order
observer. But unfortunately, the state estimates will depend on the accuracy with which the system parame-
ters, that is, the elements of the A matrix are known. To overcome these problems, the output feedback
method is also popular. In the output feedback approach, only the measurable or the available state compo-
nents of the system are used for feedback control. Further, unlike the full-state feedback, the output feed-
back control can be used to design compensators with any desired dynamical structure, thereby regaining
much of the intuition of classical controls design. These concepts are discussed in the succeeding two
chapters.
At the outset, this chapter may appear to be appar- To understand the discrete domain system repre-
ently disconnected with power electronics systems. sentations, the z-transform, digital filters and sam-
However, as stated earlier in the introduction, this pling aspects are discussed. The mapping between the
chapter is a cocktail of concepts and principles that s-plane and the z-plane will give a visual comparative
are needed to better understand and design the con- insight between the continuous and the discrete
trollers for the power electronics systems. For the domain representations. Performance specifications
design of controllers, first the physical power elec- and performance measures are introduced in the later
tronics system is modeled using the methods dis- part of the chapter to provide the framework for spec-
cussed in the previous chapter. The continuous ifying the control-related measures. Finally, the next
domain model of the system is required to be shifted section gives few tutorial exercises that should be
to the discrete domain in order to design the digital studied by simulating in a programming environment
controllers. For this, the continuous to discrete like SciLAB or MATLAB. These exercises will
domain conversion tools are required. Various meth- strengthen the discrete domain and control basics that
ods have been discussed, but the zero-order hold or are needed for the controller design which will be dis-
ZOH is by far the most popular. cussed in the next chapter.
Chapter 09.indd 543 3/14/2009 2:54:21 PM

The following tutorial exercises can be studied in Vary the sampling frequency from 100 Hz to
either SciLAB/SciCos or MATLAB/Simulink 5 kHz and plot the frequency spectrum. What
simulation environment. is effect of sampling frequency on the frequency
spectrum?
1. Construct a close-loop system with plant
5. The correct mapping between the s-plane and
( s + 1)( s + 2) the z-plane is given by z = e sT. Graphically give
( s + 1.5)( s + 2.5)( s + 3.5) the s-plane to z−plane mapping along with the
stability boundary for the following methods:
and a gain block K as the controller. Give a step
a. Euler’s forward rule
input as the reference. Probe the error wave-
b. Euler’s backward rule
form and the control input to the plant.
c. Tustin’s rule
a. Vary K from 0.1 to 1000 and observe the
d. Zero-order hold
waveforms.
b. Keeping K at 0.1, introduce an integra- 6. A digital filter has two poles at z = 0 and two
tor between the controller and the plant. zeros, one at z = i and the other at z = −i. To
Observe the waveforms. this system apply a sine wave input of fre-
c. Introduce a limiter between the control- quency 25 Hz which is sampled at 100 Hz.
ler output and the plant. Repeat Steps Build the above system in the simulation envi-
(a) and (b). ronment and observe the input and output on
d. Introduce some high-frequency noise the scope. Also observe the frequency response
before the plant by including a random of the above filter. Discuss and comment on
noise generator in the simulation sche- the observations. What happens to the output
matic. Repeat Steps (a) and (b) above. when there is a slight deviation in the frequency
e. Modify the plant by removing the zeroes. of the sine wave about 25 Hz? Why?
Now conduct Steps (a) and (b).
7. A digital filter is designed such that its output
2. A system has a transfer function given by is the average of the current input sample
together with previous (L − 1) input samples.
H(z) = 1 + 5z-1 − 3z−2 + 2.5z-3 + 5z-8
Write a function for the above L-point averag-
Determine and plot the output sequence y(n) ing FIR filter. The function has the following
when the input is d(n). features:
a. Input the value “L” for L-point averaging.
3. A system is described by the following input–
b. Output the pole-zero map of the L-point
output relation
average.
y(n) = [x(n) + x(n − 1) + x(n − 2)]/3
c. Output the magnitude and phase fre-
Determine the system transfer function H(z) quency response.
for the system. Plot the poles and zeroes of
8. Determine the range of controller gain K for
H(z) in the z-plane. What is the output if
stability of the unity feedback control system
the input is x(n) = 4 + cos[0.25p(n − 1)]
whose open-loop transfer function is
− 3cos(2pn/3)?
K
4. Sample a sinusoidal waveform of 50 Hz using a G (s ) =
1 kHz sampler and plot the sampled sequence. s ( s + 1)( s + 2)
Chapter 09.indd 544 3/14/2009 2:54:21 PM

9. Obtain the velocity error coefficients of the fol- 10. Consider the unity feedback control system
lowing systems through simulation: whose open-loop transfer function is
10 100
a. G ( s ) = G(s ) =
( s + 1)(5s 2 + 2 s + 10) s (0.1s + 1)
3s + 10 Determine the steady-state error when the

b. G ( s ) = input is r (t ) = 1 + t + at 2 where a ≥ 0.
5s 2 + 2 s + 10

1. Majority of the controllers exist in the 11. In a physical sense, z−1 is a element in
domain whereas the plant itself is in the digital processor and z− m means that there
the or the domain. are elements in the digital processor.
2. A system exists always in the 12. Digital filters are broadly classified as
domain. filter and filters.
3. The system dynamics and waveshape best 13. FIR filter has only -forward gains.
viewed in the domain.
14. All the poles of the FIR filter are located at the
4. Information like bandwidth, harmonic con- of the z-plane.
tents, spectrum, energy spectrum and power
15. FIR filter is an -loop filter.
spectral density are better understood when
viewed in the domain. 16. In a FIR filter of order m, for a unit impulse
input, the output response will exist upto
5. The domain provides information
the sample time.
about damping, natural frequencies and domi-
nant time constants for analysis and design. 17. In an IIR filter of order m, for a unit impulse
input, the output response will exist upto
6. To represent sampling of a continuous signal
.
x(t), one needs to multiply x(t) with a train of
. 18. IIR filter is a -loop filter. It contains
both the gains and gains.
7. There exists one-to-one correspondence
between the signal representation in the 19. The origin of the s-plane maps to the point
and the . in the z-plane.
8. The co-efficients are in both n- and 20. If the poles lie within the circle in the
z-domains. z-plane then the system is stable.
9. Passing a signal through a z−1 block introduces 21. The region outside the unit circle in the z-plane
a unit time . is the region.
10. If one were to pass a signal through a z− m 22. A continuous time signal can be converted into
block, the input signal will be time shifted by a discrete time signal by .
delays.
Chapter 09.indd 545 3/14/2009 2:54:21 PM

23. The sampling frequency must be greater than 25. In a system, a step input will disturb
the highest frequency component of the output response which will just oscillate.
the analog signal.
26. In a system, a step input will disturb
24. In an system, a step input will dis- the output response which will exponentially
turb the output response which will grow in decay to zero.
amplitude.
1. Why is there a shift from the analog domain 12. What is aliasing? Discuss.
control to digital domain control?
13. How is anti-aliasing usually achieved?
2. How are computation, repeatability and re-
14. What are the different methods of converting
usability aspects better in the digital domain
the continuous domain models to the discrete
vis-à-vis the analog domain?
domain models?
3. For controller design, it is essential that all the
15. Write short notes on the following: (a) pole-
system blocks are equivalently either in the
zero map, (b) numerical integration methods,
analog domain or the digital domain. Based on
(c) hold equivalence and (d) discretizing state
this what are the two possible approaches and
equation.
which is the preferred method from the perfor-
mance point of view? 16. Why is hold equivalence generally the most
preferred method of converting the continuous
4. What are the common domains in which a
domain model to the discrete equivalent?
system is usually represented?
17. What are the three performance measures for a
5. What is z-transform of a discrete signal?
system to be controlled?
6. “z−1 block introduces a unit time delay to the
18. Write short notes on the following: (a) stability,
input stream”. Discuss.
(b) steady-state accuracy and (c) speed of
7. Discuss the FIR and IIR filters using the unit response.
delay blocks.
19. What are positional error constant, velocity
8. Discuss the canonical or standard form of rep- error constant and acceleration error constant?
resentation of the IIR filter.
20. How is stability measured and specified?
9. Discuss the mapping of the various critical
21. How is steady-state accuracy measured and
points of the s-plane onto the z-plane.
specified?
10. Every strip in the left half s-plane of height 2p/T
22. How is speed of response measured and
beyond +p/T and −p/T maps onto the same
specified?
unit circle in the z-plane. Why? Explain.
23. What is bandwidth of a system?
11. What is the effect of sampling on the signal
viewed in the time domain, the frequency 24. What is rise time and settling time?
domain and the pole-zero domain?
Chapter 09.indd 546 3/14/2009 2:54:21 PM

25. Write short notes on the following: (a) peak 28. What is control band?
overshoot, (b) delay time and (c) error band.
29. What is limit cycle oscillation? Why does it
26. “For an ideal noiseless system with no limit on the occur?
system’s control input value, the controller gain
30. What are the features of the state space method
that gives zero error during transients and steady
of controller design?
state is infinite”. Explain.
27. What is the effect on the controller gain due to
presence of noise in the system?
| PROBLEMS
1. Obtain the discrete transfer function of the fol- 5. The mapping between the s-plane and the
z-plane variables is given as z = e . However,
sT
lowing digital filters and check their stability:
a. y(k) = 0.7y(k − 1) − 0.4y(k − 2) + the mappings that result from the numerical
0.4x(k) integration methods are different from the
b. y(k) = 1.8y(k − 1) − y(k − 2) + x(k) actual mapping relation given above. How does
c. y(k) = y(k − 1) + 0.3y(k − 2) + 0.1x(k) the approximate mapping relation affect the
mapping of the critical points of the s-plane
2. Derive the z-transform for the following sig-
onto the z-plane for the following integration
nals: (a) unit step, (b) unit ramp, (c) exponen-
methods: (a) Euler’s forward rule, (b) Euler’s
tial decay with time constant t, (d) sinusoidal
backward rule and (c) Tustin’s trapezoidal rule.
function of frequency w, (e) co-sinusoidal
function of frequency w. 6. Map the pole and zero locations of the follow-
ing discrete domain transfer functions to the
3. Determine the z-transforms for the following
s-plane using (a) exact mapping, (b) Euler’s for-
continuous domain transfer functions:
ward rule, (c) Euler’s backward rule and
10 (d) Tustin’s trapezoidal rule:
a. G ( s ) =
s ( s + 1)2 z ( z + 1)( z + 2)
a. H ( z ) =
5 ⋅ s ⋅ e st ( z − 0.3)( z 2 − 0.5z + 0.5)
b. G ( s ) =
s ( s + 1)( s + 2)
( z − 2)2
b. H ( z ) =
100 ( z + 0.3)( z + 0.5)( z − 0.7 )
c. G ( s ) =
s + 10 s + 25
2
z ( z − 1)
c. H (z ) =
10( s − 1) z 2 − 3z + 1
d. G ( s ) =
s ( s + 2)3 7. A continuous domain system is given as
10 1
e. G ( s ) = H (s ) =
s 2 − 5s s ( s + 4)
2
4. Show that the z-plane variable “z”, the s-plane This continuous system is sampled by means of
variable “s” and the sampling time T are related a sampler followed by a zero-order hold device.
by z = e sT . This discrete system is in a closed-loop unity
Chapter 09.indd 547 3/14/2009 2:54:21 PM

feedback configuration. The controller gain is This system is connected in a closed-loop unity
K. Then feedback configuration. Find the positional,
a. What is the discrete transfer function of velocity and acceleration error constants for the
the open-loop system? closed-loop transfer function.
b. What is the discrete transfer function of
9. For the system of Problem 8, determine the
the closed-loop system?
steady-state error resulting from the reference
c. Is the system stable at K = 1?
input of a ramp waveform which is represented
8. A type 1 system has a transfer function given as r(t) = 5t.
as
10. For the system of Problem 8, determine the
5 steady-state error resulting from the input that
H (s ) =
s ( s + 4) is given as r (t ) = 5 + 6t + 2t 2.
| ANSWERS
Fill in the Blanks
1. digital; continuous; analog 10. m time 18. close; feed-forward; feedback
2. time 11. memory; m memory 19. z=1
3. time 12. finite impulse response; 20. unit
4. frequency infinite impulse response 21. unstable
5. pole-zero 13. feed 22. sampling
6. unit impulses 14. origin 23. twice
7. n-domain; z-domain 15. open 24. unstable
8. identical 16. m times 25. marginally stable
9. delay 17. infinity 26. stable
Chapter 09.indd 548 3/14/2009 2:54:22 PM

Digital Controller Design
Learning Objectives
CHAPTER
10
design digital controllers using transfer function methods like Bode diagram and root
locus plots.
design digital controllers using state space methods.
design estimators using state space methods.
T his chapter uses the concepts discussed in the previous two chapters to design controllers for systems
and more specifically power electronic systems. As majority of the controllers today are in the discrete
domain, this chapter primarily addresses the design of digital controllers. Insights into the digital domain
design concepts are explained with the help of examples of power electronic systems.
Most literature on control systems devotes considerable space to the stability issues and the methods for
finding out the stability and relative stability of the systems. Methods such as Routh–Hurwitz criterion, Jury
test, etc. are discussed at length in literature. However, with a mathematical model of the system available,
there are a host of programing environments such as SciLAB and MATLAB that will compute the exact
location of the poles and zeros of a system of any given order. Therefore, this text does not focus much on
the classical methods of stability determination. Instead more space is devoted towards design and examples
that will enhance the insights into the system.
The controller design addresses the three performance criteria: (a) stability, (b) steady-state accuracy,
(c) speed of response, as discussed in the previous chapter. These three criteria form the specifications for
which the controller should be designed. In Section 9.8 of Chapter 9, the general structure of a closed-loop
system is discussed. The main function of the controller is to make its input zero by appropriately adjusting
its output. This implies a controller with infinite gain. However, as discussed in the above-mentioned section,
there are two major issues with infinite gain as controller: (a) The noise problem – every component generates
noise and this will also get amplified by the high-gain controller making the output noisy and (b) finite
signal limits – the output of the controller and the system control input have finite upper and lower limits.
This will cause the infinite gain controller output to clamp to the permissible extreme limits leading to limit
cycle saturation operation, which is also called limit cycle oscillation. Both these problems will cause the
system to malfunction and the three performance criteria mentioned above may not be met. This chapter
will discuss the ways and means of meeting the above-mentioned performance criteria in the presence of
noise or external disturbance and with finite control input values.
Chapter 10.indd 549 3/18/2009 1:13:01 PM

10.1 Controller Design Techniques
T he control of power electronic systems can be visualized by the block schematic shown in Figure 10.1.
The controlled plant is a power electronic system wherein either the output voltage or the input current
or some specific plant parameter needs to be controlled. The parameter to be controlled is sensed and fed
back to be compared with the desired reference after being appropriately processed by a signal processing
circuit. The error between the reference and the fed back signal is given to the controller which will appro-
priately generate a control voltage. For most switched power systems the plant is controlled by duty cycle
variation. Therefore a voltage-to-time converter is used to map the controller output to the controller input,
which is invariably the duty cycle in switched power systems.
For the system to meet the performance requirements of stability, steady-state accuracy and speed of
response, numerous design methods are available for the controller design. The design methods can be
broadly classified into the following:
1. Transfer Function Based Controller Design Methods: Here the system is modeled primarily as a linear
time-invariant (LTI) system with single-input and single–output (SISO). The transfer function along
with block diagram algebra is eminently applicable in these methods.
2. State Space Controller Design Methods: Here the system is modeled as a general multiple-input and
multiple-output (MIMO) plant. The state equation representation of the system is used in these methods.
The traditional and classical techniques are based on the transfer function models of the system. These
methods are:
1. Nyquist diagram method.
2. Inverse polar plot method.
3. Nichols chart method.
4. Bode diagram method.
5. Root locus method.
The modern techniques are based on the state equation models of the system. These methods are:
1. full-state feedback;
2. full-state feedback with estimator;
3. output feedback;
4. optimal controller – linear quadratic;
Reference Voltage Power Vo

+ Controller to electronics
− e Vc u
time system
Signal
processing
Figure 10.1 Block diagram of the control scheme of a typical power electronic system.
Chapter 10.indd 550 3/18/2009 1:13:02 PM

Digital Controller Design 551
5. optimal controller with optimal estimator;

6. robust controller;
7. self-tuning and adaptive controllers.
Among the classical methods, though any of the above methods can be used for low-order systems, for
high-order systems, the root locus method would in general be a more appropriate approach. The Bode
diagram approach can also be used; however, that approach cannot be used for non-minimum phase
systems as will be explained later. Calculation of the root loci and the roots of the characteristic equations
was a tedious and cumbersome manual process before the advent of the computational tools. Now with
software such as SciLAB, MATLAB and other similar computational tools available, plotting the root
loci is no longer an issue for complex systems. Therefore, the root locus method itself would be sufficient
to design any LTI SISO system of any order by the classical method. However, as Bode diagram is also
a popular tool in use by the designers; both these methods will be discussed in the following
sub-sections.
The state space methods will also be discussed in this section. The full-state and output feedback methods
are discussed in this chapter; however, the optimal and robust controller design methods will be discussed in
the next chapter. The self-tuning and adaptive controllers are beyond the scope of this book and are therefore
not discussed.
10.2 Bode Diagram Method
T his method is a popular method that is used by the practicing engineers as it is relatively easy to check the
anticipated results by just obtaining the open-loop frequency response of the system. Bode’s contribution
to the field of control is in the form of two important statements called Bode’s theorems. These two theorems
form the basis for the Bode diagram method.
Bode’s Theorem I
Statement: The slopes of the asymptotic amplitude–log frequency curve imply a certain corresponding phase shift,
and the slope at crossover, where the amplitude–log frequency curve crosses the 0 dB line, is weighted more towards
determining system stability than a slope further removed from this frequency.
A slope of 0 dB/decade results in a phase shift of 0, a slope of +20 dB/decade results in a phase shift of
+90o, a slope of +40 dB/decade results in a phase shift of +180o, a slope of –20 dB/decade results in a phase
shift of –90o, a slope of –40 dB/decade results in a phase shift of –180o and so on. Every slope of +20 dB/
decade adds a phase shift of 90o and every slope of −20 dB/decade adds a phase shift of −90o. Further, the
slope at the gain crossover frequency is a measure of the relative importance towards determining the system
stability. Specifically, at the gain crossover frequency, the phase shift at this frequency is measured to deter-
mine the phase margin with respect to 180o.
If a system has a slope of –20 dB/decade at the gain crossover and the other slope sections are far away
from the gain crossover frequency, then the phase shift at the gain crossover frequency is approximately 90°
and the corresponding phase margin is 90o. Such a value of phase margin implies a stable system. However,
if the slope at the gain crossover frequency is –40 dB/decade and the other slope sections are far away from
the gain crossover frequency, then the phase shift is approximately 180o and the corresponding phase margin
is 0o. This implies that such a system will be on the verge of instability. If the slopes are steeper at the gain
crossover frequency, the phase margins would be negative and indicate unstable systems. Therefore, by Bode
diagram approach, it is important to maintain the slope of the amplitude–log frequency curve in the vicinity of the
gain crossover frequency at a slope of –20 dB/decade.
Chapter 10.indd 551 3/18/2009 1:13:02 PM

Bode’s Theorem II
Statement: The amplitude and phase characteristics of linear, minimum phase shift systems are uniquely related.
This means that when the slope of the amplitude–log frequency curve over a particular frequency interval
is specified, the corresponding phase shift characteristics over that frequency interval are automatically
defined. Conversely, if the phase shift over a particular frequency interval is specified, the corresponding
amplitude–log frequency characteristics over that frequency interval are automatically defined.
Bode’s second theorem may appear trivial at the outset. However, its implications are quite significant.
It should, however, be noted that the second theorem implies that Bode’s method should be used only for
minimum phase systems because they have minimum phase shift possible for the number of energy-storage
elements in the system. This restricts the poles and zeros of the systems to lie on the left half of the s-plane
or within the unit circle in the z-plane. For example, a pole on the left half of the s-plane contributes a slope
of –20 dB/decade and a phase shift of –90o. A zero on the left half of the s-plane contributes a slope of +20
dB/decade and a phase shift of +90°. However, a zero on the right half of the s-plane contributes a slope of
+20 dB/decade and a phase shift of –90o. Therefore, in such a case even though the system is stable, specifying
the phase shift will not define the amplitude–log frequency characteristics and vice-versa. Therefore, Bode
diagram method should not be applied for non-minimum phase systems.
Bode Diagrams
For a system that has the transfer function model given as G (s), as the s-plane variable “s” traverses the imag-
inary axis, it takes the value of the real frequency w and the plots corresponding to the amplitude and phase
angle of G(jw) are obtained. The amplitude and the phase angle that are plotted with w as the common axis
are called the Bode diagrams.
The amplitude or magnitude of the transfer function is expressed in decibels (i.e., dB) and the phase
angle in degrees. The magnitude M is given as
M = 20 ⋅ log10 G ( jω) (10.1)

and the phase angle is
⎛ Imag part G ( jω) ⎞
φ( jω) = ∠G ( jω) = tan −1 ⎜ ⎟ (10.2)
⎝ Real part G ( jω) ⎠
Constant
For a transfer function G(s) = K, a constant
M = 20 ⋅ log10 K (10.3)
⎧⎪0o K ≥ 1
φ( jω) = ⎨ o (10.4)
⎩⎪180 K < 1
Figure 10.2(a) shows the Bode diagrams for the constant transfer function.
Integrator
The transfer function of an integrator is given as G(s) = 1/s. Therefore
1
G ( jω) =
jω
Chapter 10.indd 552 3/18/2009 1:13:02 PM

The magnitude is given as

1
M = 20 ⋅ log10 = −20 ⋅ log10 ω (10.5)
ω
The phase angle is given
⎛ 1/ ω ⎞
φ( jω) = tan −1 ⎜ − ⎟ = −90° (10.6)
⎝ 0 ⎠
Figure 10.2(b) shows the Bode diagrams for the integrator. It can be seen from Eq. (10.5) and Figure 10.2(b)
that the magnitude plot has a slope of –20 dB/decade and the corresponding phase shift angle is –90o.
Pure Differentiator
Pure differentiator does not exist in nature. However, the transfer function is given as G(s) = s and G(jw) = jw.
The magnitude is given as
M = 20 ⋅ log10 ω (10.7)
and the phase angle is
⎛ω⎞
φ( jω) = tan −1 ⎜ ⎟ = 90o (10.8)
⎝0⎠
Figure 10.2(c) illustrates the Bode diagrams for the differentiator.
First-Order Low-Pass Filter

The transfer function is given as G ( s ) = a / ( s + a ) and G ( jω) = a / ( jω + a ). The magnitude is
⎛ a ⎞
M = 20 ⋅ log10 ⎜ ⎟ (10.9)
⎝ jω + a ⎠
and the phase is
⎛ a a − jω ⎞ ⎛ a2 aω ⎞ ⎛ω⎞
φ( jω) = ∠G ( jω) = ∠ ⎜ ⋅ ⎟ = ∠⎜ 2 −j 2 2 ⎟
= − tan −1 ⎜ ⎟ (10.10)
⎝ a + j ω a − j ω ⎠ ⎝ a + ω 2
a + ω ⎠ ⎝a⎠
Figure 10.2(d) illustrates the Bode diagrams for the first-order low-pass filter. The dotted line indicates the
exact magnitude characteristics and the dark line indicates the asymptotic approximation. They would differ
at most by 3 dB at ω = a. The asymptotic approximations are generally preferred by the designers to quickly
obtain the frequency character of the system. For example, in the case of the low-pass filter discussed above,
for frequencies much less than “a”, the magnitude is 0 dB and for frequencies much greater than “a ”, the
magnitude is a function of w. Thus,
⎧20 ⋅ log10 (1) = 0 ωa

M =⎨
⎩−20 ⋅ log10 (ω / a ) ωa
The phase shift is as given by Eq. (10.10). At the break frequency where the slope changes, the phase shift is
–45o. The first-order low-pass filter can be realized with a resistor and a capacitor. There is only one energy-
storing element in the network and it can be seen that the phase shift is the minimum swing between 0o and
–90o. This is, therefore, a minimum phase network.
Chapter 10.indd 553 3/18/2009 1:13:02 PM

∞
dB
M −20 dB/dec
dB
M
0 w
1
20 log10K
0 w f
0
f w
−90°
0 w
(a) (b)
M w=a
+ 20 dB/dec 0
w
0 w Exact −20 dB/dec
1
f
−∞
90° 0
−45° w
f −90°
0 w
(c) (d)
Figure 10.2 Bode diagrams for (a) constant; (b) pure integrator; (c) pure differentiator;
(d) first-order low-pass filter.
Lead–Lag Network
The transfer function is given as G ( s ) = ( s + a ) / ( s + b ) and G ( jω) = ( jω + a ) / ( jω + b ). The magnitude and
phase angle are given as
jω + a
M = 20 ⋅ log10 (10.11)
jω + b
⎛ a + jω b − jω ⎞
φ( jω) = ∠G ( jω) = ∠ ⎜ ⋅ ⎟
⎝ b + jω b − jω ⎠
⎛ ω2 + ab ω (a − b ) ⎞ ⎛ ω (a − b ) ⎞
= ∠⎜ 2 −j 2 2 ⎟
= − tan −1 ⎜ 2 ⎟ (10.12)
⎝b +ω b +ω ⎠ ⎝ ω + ab ⎠
2
Chapter 10.indd 554 3/18/2009 1:13:02 PM

To obtain the asymptotic approximation, consider a > b. Then from w = 0 to w = b, the magnitude is a
constant with amplitude M = 20 ⋅ log10 ( a / b ) with a magnitude slope of 0 dB/decade. From w = b to w = a,
the magnitude slope changes by –20 dB/decade with respect to the previous slope due to introduction of the
pole or lag effect at w = b. The phase angle at w = b is –45o and drifts towards –90o. From w = a onwards,
the magnitude slope changes by +20 dB/decade with respect to the previous slope due to the introduction
of zero or lead effect at this point. The phase angle is back again at –45o at w = a due to positive angle addi-
tion to the phase because of the zero. The magnitude slope now becomes 0 dB/decade.
The Bode diagrams for the lead–lag network are illustrated in Figure 10.3. It should be noted that
plotting the Bode diagrams is no longer done manually. With the availability of a host of computer programs
like SciLAB and MATLAB, the exact bode diagrams are plotted and used for design purposes.
The Bode diagram for any general minimum phase system is shown in Figure 10.4. There are two
important parameters that are used in the analysis and synthesis: (a) gain margin and (b) phase margin.
In any feedback control system, the feedback is negative. If the controlled system introduces a phase shift of
180o, then along with the negative feedback, the overall phase shift around the loop becomes 360o and
therefore positive feedback. If the system gain is greater than 1, the positive feedback will make the output
grow till the output saturates or goes into limit cycle oscillation. Therefore, the critical points on the open-loop
Bode diagrams are as follows:
1. At the gain crossover frequency, the magnitude M = 0 dB or a gain of 1. At this frequency, the phase
shift should be within ±180o for the closed-loop system to be stable. The margin that the phase angle
has with respect to 180° (or instability) is called the phase margin. The phase margin is measured from
the ±180o phase angle line to the phase angle curve as shown in Figure 10.4. It is considered positive if
the phase margin arrow head is towards the 0o phase line.
2. At the frequency where the phase angle is ±180°, the gain should be less than 1 (i.e., attenuation) or
–ve dB such that the system output will not grow. The margin that the gain or magnitude has with
respect to 0 dB or unity gain (or instability) is called the gain margin. The gain margin is measured from
the magnitude curve to the 0 dB line. It is considered positive if the arrow head points upwards.
20 log10 a
b −20 dB/dec
0 w=a w
w=b
Pole or
lag effect Zero or
lead effect
0
w
−45°
−90°
Figure 10.3 Bode diagram for lead–lag network.
Chapter 10.indd 555 3/18/2009 1:13:02 PM

dB
M
Gain crossover
0 w
Gain margin
0° w
−180°
Phase margin
Figure 10.4 Bode diagram indicating gain margin and phase margin.
Both gain and phase margins are measures of relative stability. The gain towards w = 0 frequency region
gives a measure of the steady-state accuracy. The steady-state portion of the dynamics is near the zero
frequency region and the transient portion of the dynamics is the high-frequency region. For the steady-state
error to be zero, the gain near the steady-state region should be close to infinity so that the output parameter
by the gain will result in almost zero error.
Controller Design with Bode Diagram

The controller design using the Bode diagrams is based on the following three constraints:
1. Gain margin should be positive (e.g., 6 dB gain margin).
2. Phase margin should be positive (e.g., 45o phase margin).
3. The gain or magnitude near the zero-frequency regions should be high to achieve low steady-state
error.
The steps for designing the controller are as follows:
Step 1: Plot the amplitude–log frequency and the phase–log frequency Bode diagrams for the system
with a specific open-loop transfer function.
Chapter 10.indd 556 3/18/2009 1:13:02 PM

Step 2: Find out the gain margin and the phase margin for the open-loop Bode plots.
Step 3: Find out the magnitude near the zero-frequency regions.
Step 4: If the gain margin, phase margin and the low-frequency gains do not satisfy the specifications,
then the open-loop Bode plot curves should be shaped accordingly. The shaping is done by including
the controller or compensator which will alter the shape of the plant’s open-loop Bode diagram. As the
controller and the plant are in series, the controller Bode plot will add onto the plant Bode plot to
result in the modified Bode plot. This shaping is done based on the following notions:
1. Including a gain in the controller will shift the Bode diagram up or down without affecting the
phase angle plot.
2. Including a pole or lag in the controller will add a –20 dB/decade slope to the magnitude curve
from the frequency at which the pole or lag is inserted. The phase angle curve from the frequency
of pole or lag insertion will shift by at most an additional –90o as given by the lag phase angle.
3. Including a zero or lead in the controller will add a +20 dB/decade slope to the magnitude curve
from the frequency at which the zero or lead is inserted. The phase angle curve from the frequency
of lead insertion will shift by at most an additional +90o as given by the lead phase angle.
4. The Bode diagram shaping should be done such that at the gain crossover, the slope of the magni-
tude curve should be –20 dB/decade.
The above steps can be iterated with the help of SciLAB or MATLAB program. Though the Bode diagram
has been discussed using the s-plane variable, it is applicable without loss of generality to the discrete domain
using the z-plane variable. In the s-plane, the s-plane variable traverses the w-axis to obtain the Bode magni-
tude and phase diagrams. In the z-plane, the w-axis maps to the circumference of the unit circle. As the
z-plane variable traverses the circumference of the unit circle, the Bode magnitude and phase diagrams are
obtained for the frequency range of 0–ws/2, where ws is the sampling frequency.
EXAMPLE 10.1 Consider the separately excited DC motor that is discussed and modeled in Chapter 8
(Figure 10.5). Let it be required that the speed of the DC motor be controlled.
ld
fie
a nt
nst
if Co
i a(t )
Ra La
J
V(t ) Vb(t ) B (friction)
Figure 10.5 Armature-controlled DC servomotor.
Chapter 10.indd 557 3/18/2009 1:13:02 PM

Large-Signal Model of Plant

In state space form, the plant model is given by
⎡dia / dt ⎤ ⎡− R / La − K / La ⎤ ⎡ia ⎤ ⎡1 / La 0 ⎤⎡ v ⎤
⎢dω / dt ⎥ = ⎢ K / J +
−B / J ⎥⎦ ⎢⎣ω⎥⎦ ⎢⎣ 0 −1 / J ⎥⎦ ⎢⎣TL ⎥⎦
⎣ ⎦ ⎣
ω (s ) K / La J
G(s ) = = 2
V ( s ) s + [(B / J ) + ( Ra / La )]s + [( Ra B / La J ) + ( K 2 / La J )]
Plant Parameters
Armature resistance, Ra = 2 Ω
Armature inductance, La = 0.5 H
Torque/back emf constant, K = 0.1
Friction coefficient, B = 0.2 N ms
Load inertia reflected onto the motor shaft, J = 0.02 kg m2/s2
Sampling time, Ts = 20 ms
Transfer Function in the s-Domain
The transfer function of the angular frequency with respect to the armature voltage
is given as
w(s) = 10
----- = ----
V(s) = (s + 9.828) (s + 4.172)
Transfer Function in the z-Domain

Conversion from continuous to discrete domain is performed by zero-order hold
(ZOH) equivalence.
w(z) = 0.0018233 (z + 0.9109)
----- = --------------------
V(z) = (z − 0.92) (z − 0.8215)
State Space Model in the z-Domain

Conversion from continuous to discrete domain is performed by ZOH
equivalence.
⎡0.92294 − 0.00348 ⎤
A=⎢
⎣0.08698 0.81856 ⎥⎦
⎡0.03844 0.00182 ⎤
B=⎢
⎣0.00182 − 0.90629 ⎥⎦
C = [0 1]
Chapter 10.indd 558 3/18/2009 1:13:02 PM

D = [0 0]
x = [ia; w] u = [v; TL]
Bode Plot of Open-Loop Plant
The open-loop Bode plot of the plant is shown in Figure 10.6. It can be observed that
1. The magnitude plot indicates that the gain is never above unity. This implies
that the low-frequency or DC gain is very low and therefore the steady-state
error is large.
2. Gain margin = 43.3497 dB at –180o phase crossover frequency of 37.131 rad/s.
3. As there is no gain crossover frequency, there is no concept of gain margin here.
To compensate the system, the open-loop Bode plot can be modified by includ-
ing a compensator or controller such that
1. The DC gain is infinite. This implies that an integrator be used. The integrator
scaling factor can be adjusted such that the magnitude curve cuts the 0 dB line
so that the phase margin is at least 45o and the gain margin is at least 10 dB.
2. Include a zero in the controller such that the magnitude plot flattens out
beyond the zero and the phase shift remains more or less unchanged beyond
the zero frequency.
−10
−20
−30
M (dB)
−40
−50
−60
−70
10−1 100 101 102
w (rad/s)
−50
Phase (degree)
−100
−150
−200
180°
−250
10−1 100 101 102
w (rad/s)
Figure 10.6 Open-loop Bode plot of the plant.
Chapter 10.indd 559 3/18/2009 1:13:03 PM

The Bode plot shown in Figure 10.7 is that of the controller or compensator. The
magnitude plot shows the presence of an integrator. The inclusion of the zero in
the compensator leads to the controller magnitude slope becoming 0 dB/decade.
This controller is called a PI controller that is discussed in the next section.
The controller corresponding to the Bode plot of Figure 10.7 is
6( z − 0.9)
Gc ( z ) =
z −1
The Bode plot of the compensated system is shown in Figure 10.8 wherein it is
compared with the Bode plot of the uncompensated plant. Observe that the gain
margin is about 24 dB and the phase margin is 50.8o. The step response of the
closed-loop system is shown in Figure 10.9. The controller and the plant are con-
nected in negative feedback and the equivalent transfer function from the output
to the reference input is considered for plotting the closed-loop step response. The
step response plot shows that the steady-state error reaches zero. There is a peak
overshoot of about 20%. This can further be improved by tuning the waveshape of
the compensated system.
50
45
40
−20 dB/dec
M (dB)
35
30
25
Introduction of zero
20
15 −1
10 100 101 102
w (rad/s)
−20
Phase (degree)
−40
−60
−80
−100
10−1 100 101 102
w (rad/s)
Figure 10.7 Bode plot of the controller.
Chapter 10.indd 560 3/18/2009 1:13:03 PM

40
Compensated system
20
0 24 dB gain
Uncompensated system
M (dB)
margin
−20
−40
−60
−80
10−1 100 101 102
w (rad/s)
0
Uncompensated system
−50
Phase (degree)
Compensated system
−100
−150 50.8° phase margin

−180
−200
−250
10−1 100 101 102
w (rad/s)
Figure 10.8 Bode plot of the compensated system.
1.4
1.2
0.8
Amplitude
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70 80 90 100
Time (s)
Figure 10.9 Step response of the closed-loop system.
Chapter 10.indd 561 3/18/2009 1:13:03 PM

10.3 PID Controller
P ID is an acronym for Proportional (P) – Integral (I) – Derivative (D). The controller with infinite gain
would have been the best controller in an ideal sense. If it were not for the noise problem and the finite
control input values, the infinite gain controller would have provided zero error under all conditions (either
transient or steady state) irrespective of the system that is being controlled. However, in the presence of
noise and finite control input limits, the controller cannot be an infinite gain controller as discussed in the
previous chapter. Alternately, one can relax the constraints on the controller by stating that instead of
providing zero error under all conditions, the controller should at least provide zero error under steady-state
conditions. Herein lies the origins of the PID controller. This section will explain the evolution of the PID
controller that is probably the most popular controller in the industry today and used in about 85–90% of
all control systems.
The noise magnitude is in general a function of the frequency. It is most dominant at higher frequencies.
Figure 10.10(a) shows the noise amplitude as a function of the radian frequency. It can be observed that the
noise amplitude monotonically increases with frequency. The frequency axis (i.e., the x-axis) can be divided
into two zones: (a) the low-frequency (LF) zone where the noise is not very significant and (b) the high-
frequency (HF) zone where the noise is very dominant. It should, however, be noted that it is difficult to
have a very strict line of demarcation between the LF and HF zones. In reality, the transition from the LF to
the HF zone and vice-versa is rather hazy. The LF region is associated with the steady-state part of the time
domain response and the HF region is associated with the transient part of the time domain response.
Referring to Figure 10.10(a), the controller can have high or infinite gain in the LF zone and high or
infinite attenuation in the HF zone. Such a controller assures zero error in the LF region. However, it is
difficult or practically impossible to build a controller that swings from +∞ to −∞ as shown (dashed line)
G G
∞
−20 dB/dec
Noise amplitude
DC w DC w
LF HF
LF HF
−∞
(a) (b)
Figure 10.10 (a) Noise amplitude as a function of frequency; (b) integrator gain as a
function of frequency.
Chapter 10.indd 562 3/18/2009 1:13:03 PM

in Figure 10.10(a). Alternately, one can think of a controller gain that has an infinite gain at DC and the
gain gradually decreases with frequency such that in the HF region, the attenuation becomes high as shown
in Figure 10.10(b). The simplest block that has a gain profile like that shown in Figure 10.10(b) is that of an
integrator. The integrator having a transfer function of 1/s has an infinite gain at DC and the gain decreases
with frequency at the rate of –20 dB/decade.
More insight into the working of the controller can be gathered by visualizing both the time domain
response and the frequency domain gain plot. The LF portion of the frequency domain gain plot is the
steady-state portion of the time domain response and the HF portion of the frequency domain gain plot is
the transient portion of the time domain response. However, in order to visualize both the plots in a common
frame, the time domain response is plotted with the x-axis or the time flipped as shown in Figure 10.11(a).
This way the LF zone and the HF zones of the frequency domain and time domain response will have
reasonable correspondence.
Consider a system that is controlled by a controller having a gain that varies with frequency in a manner
as depicted in Figure 10.11(a). The evolution of the output with respect to time is plotted in the time response
graph as shown. The time increases from right to left starting from zero. The gain also evolves from right to
left on the frequency plot. The “DC” state indicates the steady-state or equilibrium condition of the system.
G G
int Ki /s
−20 dB/dec −20 dB/dec
DC w DC w
LF HF LF HF
DC DC
t t
LF HF LF HF
(a) (b)
Figure 10.11 (a) Gain versus frequency and time response; (b) scaling the
integrator gain.
Chapter 10.indd 563 3/18/2009 1:13:03 PM

If a step change is given to the system in terms of either reference change or disturbance, the time frame in
the neighborhood of the step change relates to the HF region of the gain plot. During this period when the
step change has occurred, the gain is very low and therefore, the error between the output and the reference
is very large. As time progresses towards the steady-state or DC conditions, the gain also gradually increases
thereby decreasing the error between the output and the reference. At the steady state, the gain is infinite
and the error between the output and the reference is zero. This is indicated in Figure 10.11(a). Thus it can
be observed that using a simple integrator, one of the most important performance criteria is met, that is,
zero steady-state error. Such a controller where only an integrator is used is called I-control. Figure 10.11(b)
illustrates the effect of the scaling parameter Ki on the integrator gain and the time response. A degree of
design freedom can be incorporated into the I-controller by introducing a scalar gain scaling factor Ki. The
controller transfer function Gc(s) is now given as
K
Gc ( s ) = i (10.13)
s
The gain plot is in dB and is given as
20 ⋅ log10 Gc (ω) = 20 ⋅ log10 ( K i ) − 20 ⋅ log10 (ω) (10.14)
From Eq. (10.14), it can be observed that at Ki = 1, the gain profile is same as that of the pure integrator,
that is, Gc ( s ) = 1 / s . For values of Ki > 1, the gain profile will be higher than the pure integrator gain profile
by 20 ⋅ log10 ( K i ) and for values of Ki < 1, the gain profile will be less than the pure integrator gain profile by
20 ⋅ log10 ( K i ) . This is illustrated in Figure 10.11(b). There are two important points that need to be
observed:
1. The value of Ki does not affect the DC gain. The DC gain remains at infinity and hence zero steady-state
error is assured for any value of Ki.
2. The value of Ki affects the controller bandwidth. The point of intersection of the gain profile with the 0 dB
or the unity gain axis is decided by Ki. This affects the unity gain–bandwidth product of the controller.
Therefore, Ki gives a degree of design freedom in the choice of the controller bandwidth or the control
speed. If the value of Ki is greater than 1, the speed of the response is faster and if the value of Ki is lesser
than 1, the speed of the response is slower as illustrated in Figure 10.11(b).
Figure 10.12(a) shows the next stage in the evolution of the PID controller. The I-controller takes care of the
steady-state performance. However, the transient performance is poor. To improve the transient perfor-
mance, the gain during transients, or in other words, the gain in the HF region, should be increased. In the
HF region, if a zero is introduced, it will change the slope of the gain profile by adding +20 dB/decade.
Thus, the slope becomes zero and the gain profile flattens out as shown. This is achieved by incorporating
a scalar gain Kp as shown in the block diagram of Figure 10.12(a). The controller input is passed not
only through the I-part, but also through a proportional gain part. The outputs of the two blocks are
summed and used as the total controller output. Thus, the overall controller transfer function becomes
K
Gc ( s ) = i + K p = K p ⋅ [{ s + ( K i / K p )} / s ] (10.15)
s
From Eq. (10.15), it can be seen that a zero is included into the controller by the introduction of the scalar
proportional gain Kp in a manner as shown in the block diagram of Figure 10.12(a). This changes the slope
of the gain profile from –20 dB/decade to 0 dB/decade from the point on the frequency axis as dictated by
the ratio Ki/Kp (i.e., the location of the zero).
It can be observed from Figure 10.12(a) that the gain profile shows an increase in the HF gain with
frequency as compared to the pure I-controller. This results in a better transient response as indicated in the
time response plot.
Chapter 10.indd 564 3/18/2009 1:13:03 PM

G G
Ki /s + Kp Ki /s + Kp + Kds
−20 dB/dec −20 dB/dec
Zero included
DC DC
LF HF Improved LF HF
HF gain
Zero included
DC i DC i
pi pid
t t
LF HF LF HF
(a) (b)
Figure 10.12 Gain profile and time response for: (a) PI controller; (b) PID controller.
One more degree of freedom can be incorporated by adding a derivative term as shown in the controller
block diagram of Figure 10.12(b). Here, the controller transfer function is
Gc ( s ) = ( K i / s ) + K p + K d ⋅ s = K d ⋅ [{ s 2 + ( K p / K d )s + ( K i / K d )} / s ] (10.16)
From Eq. (10.16), it can be observed that one more zero is added by the derivative term. This zero further
adds another +20 dB/decade making the gain profile to increase with a rate of +20 dB/decade as shown in
Figure 10.12(b). This further increases the gain improvement with frequency and pulls up the transient
response further to make the system fast. However, one must be cautious in introducing the derivative part
as this will result in increase in gain in the HF zone which is supposed to have significant noise. Therefore,
the scalar Kd is used to scale down the derivative effect such that the system noise does not get amplified. It
should also be noted that one cannot in practice build a pure derivative circuit. This will always be associated
with an energy-storing element indicating a pole also. Thus, Eq. (10.16) is a first-order transfer function
only in theory, but in actual practice it is a second-order transfer function. One order is due to the integrator
and the other order is contributed by the non-ideal differentiator circuit or algorithm. However, the pole
contributed by the non-ideal differentiator is not a dominant pole and is far away from the origin in the
left half of the s-plane. Figure 10.13 shows the consolidation of the Bode diagrams of both the plant and the
PID controller. Figure 10.13(a) is the magnitude Bode diagram of an arbitrary plant with break frequencies
at a and b. Figure 10.13(b) is the PID controller’s magnitude Bode diagram as discussed in this section.
Figure 10.13(c) gives the magnitude Bode diagram of the composite system comprising the plant and the
Chapter 10.indd 565 3/18/2009 1:13:03 PM

Mplant
−20 dB/dec
−40 dB/dec M −20 dB/dec

(a)
0 a b w
= w
MPID c a b d
+
(b) (PID + Plant)

c d
0 w
(c)
Figure 10.13 Magnitude plot of: (a) Plant; (b) PID controller; (c) PID controller and plant.
PID controller. The block schematic of the PID controller in a typical closed-loop system is shown in
Figure 10.14.
The blocks shown in the PID controller schematic in Figure 10.14 are generic blocks that can be applied
for either the continuous systems or the discrete systems. For discrete systems the integrator and differentiator
blocks are implemented using the unit delays as explained in Chapter 9.
Figure 10.15 shows the traditional time response of the output of the system to a step change in the
reference. Here the time evolves from left to right as is normally the convention. The time response shows
the regions where the different parts of the PID controller are active. The I-part or the integral part of
the PID controller is active during the steady-state operation wherein the error is controlled to zero.
PID controller
Ki ∫
Reference
+
− Error (e)
Feedback +
Kp ++ Plant
Controller
output (Vc)
Kd d/dt
Figure 10.14 PID controller block schematic.
Chapter 10.indd 566 3/18/2009 1:13:03 PM

D P I
DC
t
HF LF
Figure 10.15 Time response showing the dominant operating regions of the PID controller.
During the transient, the D-part or the derivative part comes into action first as it is more dominant in the
HF zone. The D-part will die out quickly and the P-part or the proportional part comes into action to act
as a bridge between the D-part and the I-part. It should be noted that if the plant already has an integrator
in-built into it or that it has an integrator pole either s = 0 or z = 1, then Ki should be zero. As the integrator
of the plant will provide the necessary infinite gain at steady-state conditions, the external integrator in the
controller is not necessary. Therefore, a propotional derivative (PD) controller is sufficient.
PID Controller Design

The PID controller parameters Ki, Kp and Kd can be designed
1. by the Bode diagram controller design method if the transfer function of the plant is known;
2. by experimental method if the model of the plant is not known.
The Bode diagram controller design method has been discussed in the previous sub-section. This sub-section
will discuss the experimental method.
The experimental method for determining the parameters of the PID controller is used for plants where
the mathematical model of the plant is not known. The experimental setup is as shown in Figure 10.16.
Reference Ki ∫
r∗
+ +
+ − e
rd
Kp Σ Plant
Disturbance Controller
output (Vc)
Kd d/dt
Feedback
Figure 10.16 Experimental setup for tuning the parameters of the PID controller.
Chapter 10.indd 567 3/18/2009 1:13:03 PM

The reference for the system is now modified to include a disturbance signal rd. The disturbance signal is a square
pulse waveform that has an amplitude that is 10% of the actual reference r *. Thus the modified reference that is
applied to the system is r * + rd. This modified reference consists of the actual reference r * superimposed with the
square pulse disturbance. The feedback signal along with the modified reference is shown in Figure 10.17.
The steps for experimentally tuning the parameters of the PID controller are as follows:
Step 1: Set Kp = 0 and Kd = 0.
Step 2: Set the frequency of the superimposed square pulse to 1 s. This should be adjusted such that
the period is much greater than the time constant of the system.
Step 3: Observe the feedback signal with respect to the modified reference as shown in Figure 10.17
on an oscilloscope.
Step 4: Start with a very small value of Ki and keep increasing Ki till it overshoots and/or damped oscillations
result in the feedback signal. Select the value of Ki to be half the value of Ki that just results in overshoots
and/or damped oscillations in the feedback signal.
Step 5: Set Ki to a value as obtained in Step (4). Set Kd = 0 and repeat the exercise of Step (4) by
gradually increasing Kp from a very small value.
Step 6: Set Ki to a value as obtained in Step (4) and Kp to the value as obtained in Step (5). Gradually
increase Kd till satisfactory time response in the transient region is obtained.
Integrator Windup in PID Controller

In some cases, the control input to the plant will have a limiter either built into the plant or included external
to it in order to limit the control input signal swing to be within the limits accepted by the interface circuitry.
For example, the PID controller may be designed with power supplies of ±15 V and the control input to the
plant may be constrained to accept signals only within ±5 V. In such cases, to protect the input circuitry of
the plant, the controller output will be passed through a limiter as shown in Figure 10.18. In such situations,
due to the memory effect of the integrator in the controller, the integrator winds up and causes a delay in the
response of the controller to the changes in the error. This problem is called integrator windup.
The integrator windup is illustrated in Figure 10.19. Consider an error signal e as shown in Figure 10.19(a).
The output of the integrator Vc linearly builds up as shown. The control input u follows the controller
output till the limiter’s upper limit of u+ is reached. When Vc reaches u+, the control input u is clamped to
r∗ + rd
r∗ + rd
r∗
Feedback
signal
t
Figure 10.17 Response of the plant to the square pulse superimposed reference.
Chapter 10.indd 568 3/18/2009 1:13:04 PM

Reference e Vc u
+ Ki ∫
− Plant
Limiter
I -control
Figure 10.18 Schematic showing the integral portion of the controller (I-control) with
output limiter.
(a)
0 t
Vc+
Vc
u+
Windup delay
u
(b)
0 t
u−
Vc−
Vc+
u+
u ,Vc
(c)
0 t
u−
Vc−
Figure 10.19 (a) Error waveform; (b) integrator output and control input showing windup
delay; (c) integrator output and control input with an anti-windup solution.
u+ due to limiter action. However, Vc continues to rise as the integrator limit is Vc+. Due to feedback
action, if the error e changes as shown in Figure 10.19(a), the integrator de-accumulates and Vc starts
decreasing as shown in Figure 10.19(b). However u starts to change and follows the controller output
only after Vc drops below u+. The time taken from the instant the error signal changes to the instant
when the control input u changes to take corrective action, is the delay in the controller response. This is due
Chapter 10.indd 569 3/18/2009 1:13:04 PM

to the fact that the integrator output had wound up to a higher value and takes some time to unwind and
come within the control band.
This problem is solved by incorporating anti-windup circuitry or anti-windup algorithm in the controller
implementation such that the integrator does not integrate beyond the limiter’s limit. In other words, the
integrator is inhibited from accumulating if the control input has reached its limits for a given error signal
polarity. Figure 10.19(c) shows the integrator output and the control input with anti-windup solution
incorporated into the integrator. It can be observed that the control input responds immediately to changes
in the error signal as the integrator output is prevented from winding up to a higher value than the limits of
the limiter.
In the case of PID controllers built with op-amps, back-to-back Zener diodes are connected to clamp
the voltage across the capacitor element that acts as the accumulator. This will ensure that the capacitor voltage
does not windup. In the case of the discrete implementation, the input to the integrator is controlled
through a Boolean logic as follows:
1. For a positive error, if the output of the PID controller is greater than or equal to the control input
upper limit, then the integrator input is set to zero.
2. For a negative error, if the output of the PID controller is lesser than or equal to the control input lower
limit, then also the integrator input is set to zero.
3. The integrator is allowed to integrate only if the above two conditions are not violated.
Practical PID Controllers

The PID controller transfer function as given by Eq. (10.16) is
Gc ( s ) = ( K i / s ) + K p + K d ⋅ s = K d [{ s 2 + ( K p / K d )s + ( K i / K d )} / s ]
However, this is not a practical transfer function as a pure differentiator cannot be implemented. The cont-
roller transfer function will also have another pole in addition to the integrator pole to make it a controller
that can be implemented. Thus the PID controller transfer function becomes
Ki s
Gc ( s ) = + Kp + Kd
s s + sd
⎛ K p + K d ⎞ ⎛ s + [( K i + K p ⋅ sd ) / ( K p + K d )]s + [( K i ⋅ sd ) / ( K p + K d )] ⎞
2
=⎜ ⎜
⎟ ⎜ ⎟ (10.17)
⎝ sd ⎠ ⎝ s [( s / sd ) + 1] ⎟
⎠
where sd is the differentiator pole frequency. This will change the high-frequency portion of the PID controller
Bode diagram as shown in Figure 10.20.
For a discrete implementation, the PID controller transfer function in the z-domain is given as
Ki K ⎛ z +1⎞
Gc ( z ) = + Kp + d ⎜ ⎟ (10.18)
z −1 T ⎝ z ⎠
where the integrator is based on Euler’s method and the differentiator is based on the first-order difference
equation as discussed in Chapter 9.
The analog implementation of the PID controller circuit using op-amps is shown in Figure 10.21. The
op-amp-1 is a difference amplifier that gives the error between the reference and the feedback signal. The
error is fed to three sets of op-amps in parallel. Op-amp-2 is the integral controller. Two back-to-back
Chapter 10.indd 570 3/18/2009 1:13:04 PM

M M
Corresponds to sd
0 w 0 w
(a) (b)
Figure 10.20 Magnitude plot of (a) an ideal PID; (b) a practical PID with differentiator pole.
Anti-windup
clamp
I output
Ri Ci Summer
− R
Reference R2 R1
+ R R
2
−
R
R2 −
e
+ RP1 − RP2
Feedback 1
+ Vc
5
R1 +
3
Cd Rd P-output
+ 4 D-output
Figure 10.21 PID circuit schematic.
Zeners are connected across the capacitor Ci to provide a simple clamp to prevent windup of the voltage
across Ci. Op-amp-3 is a simple inverting amplifier that provides the proportional gain to the error signal.
Op-amp-4 is the differentiator. The outputs of the three stages are summed up in op-amp-5 to provide the
overall PID controller output. The transfer function of the circuit is
Vc ( s ) 1 / RiC i Rp2
Gc ( s ) = = + + RdC d s (10.19)
E (s ) s Rp1
One should note that the differentiator is not ideal as indicated in Eq. (10.19). The op-amp acts as a
low-pass filter and provides the high-frequency differentiator pole to make the system a second-order
system. On comparing Eq. (10.19) with Eqs. (10.16) and (10.17), the relationships between Ki, Kp and Kd
Chapter 10.indd 571 3/18/2009 1:13:04 PM

with the circuit parameters can be established. After Ki, Kp and Kd are designed as per the discussion in a
previous section, the circuit parameters can be determined by comparing Eqs. (10.16) and (10.19).
Figure 10.22(a) shows another PID circuit that uses only one op-amp. The circuit introduces a 180°
phase reversal. Therefore, the error signal is inverted by interchanging the reference signal and the feedback
signal in the difference amplifier op-amp circuit. This will give the inverter error signal output that will
cancel out the phase reversal introduced by the PID op-amp circuit. The transfer function is given as
Vc ( s ) sR1C1 + 1 ⎛ sR1C1 + 1 ⎞ C 3
Gc ( s ) = = +⎜ ⎟
E (s ) sC1R2 ⎝ sR3C 3 + 1 ⎠ C1
⎛ R C + R3C 3 + R2C 3 ⎞ ⎛ 1 ⎞
s2 + ⎜ 1 1 ⎟s +⎜ ⎟
⎛ R1R3C 3 + R1R2C 3 ⎞ ⎝ R1C1R3C 3 + R1R2C1C 3 ⎠ ⎝ R1C1R3C 3 + R1R2C1C 3 ⎠
=⎜ ⎟ (10.20)
⎝ R2 ⎠ s ( sR3C 3 + 1)
The transfer function given in Eq. (10.20) is of the form given in Eq. (10.17). In many applications, a PI
controller alone is sufficient. Figure 10.22(b) gives the circuit schematic of PI controller. The transfer function
is given as
V ( s ) sR C + 1 1 / R2C1 R1
Gc ( s ) = c = 1 1 = + (10.21)
E (s ) sC1R2 s R2
PID
C3 R3 R1 C1 R1 C1
R2
−e
−e Vc
− −
R2
Vc +
+
(a) (b)
R1 C1
Feedback
R2
−
+ Vc
Reference R2
R1
C1
(c)
Figure 10.22 (a) Another PID circuit schematic; (b) single op-amp PI circuit schematic; (c) single
op-amp comparator and PI circuit schematic.
Chapter 10.indd 572 3/18/2009 1:13:04 PM

Anti-windup
Reference rule
+
e
Ki z −1
+
−
+
0 +
Feedback
Vc
z−1
Kp +
Kd/T +
+
z−1 −
Figure 10.23 Discrete implementation of PID controller.
This is of the form

Vc ( s ) K i
Gc ( s ) = = + Kp
E (s ) s
where
1 R
Ki = and K p = 1
R2C1 R2
The PI controller of Figure 10.22(c) is interesting in that it uses only one op-amp even eliminating the need
for the difference op-amp needed to obtain the error. This acts like a difference amplifier and has the transfer
function as given by Eq. (10.21). The disadvantage of this circuit is that two sets of matched R1, R2 and C
are needed to obtain accurate performance.
The discrete implementation is shown in Figure 10.23. The anti-windup solution for the integrator is
implemented by comparing the error signal and the PID output signal. The integrator input is either connected
to the error signal or to the zero value in accordance to the logic described in the section “Integrator Windup in
PID Controller”. Other numerical integration and differentiator algorithms can also be used in place of the
simple algorithms used in Figure 10.23 to obtain better controller output accuracy.
10.4 Root Locus Method
T he root locus method is a graphical technique for determining the closed-loop poles of a system as a
function of the gain. The relationship that exists between the poles of the closed-loop transfer function
and the gain, poles and zeros of the open-loop transfer function is the basic principle of this method. The
root locus method gives a complete and accurate transient and steady-state picture of the closed-loop system
as the closed-loop poles can be directly obtained from the root loci. Plotting the root locus on the s-plane or
the z-plane by manual means is very cumbersome and tedious. Therefore, this method, though complete
and detailed, was not very popular with the designers till the advent of computer programs that removed the
drudgery and tedium of plotting the root loci. Among the classical techniques for SISO systems, this method
by far offers the most complete, detailed and accurate transient and steady-state solution for the system.
Chapter 10.indd 573 3/18/2009 1:13:04 PM

e u y
+ K Gc Gp
Reference, r −
Plant
Controller
Figure 10.24 General structure of closed-loop system.
Figure 10.24 shows the general structure of a closed-loop system wherein the plant Gp is an LTI SISO
system. The block H is the signal processing circuitry of the feedback signal. The controller is divided into
two parts: (a) the gain part K and (b) the part representing the pole–zero structure of the controller, Gc.
Without loss of generality, the closed-loop system can represent either a discrete domain system or a
continuous domain system. In a continuous domain system, the plant transfer function is Gp(s), the feed-
back block transfer function is H(s) and the controller pole–zero structure part is expressed as Gc(s). In the
case of the discrete domain system, the continuous domain plant is converted into the discrete equivalent as
discussed in Chapter 9 and is expressed as Gp(z), the feedback block is expressed as H(z) and the controller
pole–zero structural part is expressed as Gc(z). The discussion to follow is equally applicable to both con-
tinuous domain and discrete domain systems. However, as majority of the controllers in the products today
are in the discrete domain, majority of the design examples will be in the discrete domain.
Root Locus Plot

Consider the system shown in Figure 10.24. The closed-loop transfer function is given as
Y (s ) K ⋅ Gc ⋅ Gp
Gclose ( z ) = = (10.22)
R ( s ) 1 + K ⋅ Gc ⋅ Gp ⋅ H
From the closed-loop transfer function, the closed-loop poles are given by the roots of the denominator of
Eq. (10.22), that is, the roots of the characteristic equation. Thus the following equation,
1 + K ⋅ Gc ⋅ Gp ⋅ H = 0
is the characteristic equation. All the components of the characteristic equation are obtainable from the
open-loop model of the system. Here Gp is the plant model, H the model of the feedback signal conditioner,
Gc the pole–zero structure of the controller that is defined by the designer, K the controller gain. Therefore,
to find the poles of the closed-loop transfer function,
1 + K ⋅ Gc ⋅ Gp ⋅ H = 0
K ⋅ Gc ⋅ Gp ⋅ H = −1 = 1∠(2n + 1)π (10.23)
where n = 0, ±1, ±2, …. From Eq. (10.23) the existence of a closed-loop pole at a point on the s-plane or z-plane
is obtained if the following two constraints are satisfied:
1. Magnitude constraint, KGcGp H = 1.
2. Angle constraints, ∠KGcGp H = ( 2n + 1)π.
Chapter 10.indd 574 3/18/2009 1:13:04 PM

In the two constraints stated above, at different values of K, different points on the s- or z-plane are obtained
that would satisfy the constraints and become eligible to be closed-loop poles. Thus as K is varied from
0 to ∞, all possible closed-loop poles are located on the s- or z-plane for any given K. The locations of the
closed-loop poles on the s- or z-plane form lines as K is varied and are called the loci of the roots of the charac-
teristic equation or the root loci.
There will be as many root loci as the order of the system. This is due to the fact that there will be as
many closed-loop poles as the order of the system and therefore as many root loci. Each segment or branch
of the root locus describes the variation of the particular closed-loop pole as the gain is varied.
The open-loop poles define the start of the root locus when K = 0 and the open-loop zeros define the
termination of the root locus when K = ∞. This can be easily concluded from the magnitude condition of
Eq. (10.23), that is, KGcGp H = 1 . At the open-loop poles, GcGp H is infinite. To satisfy the magnitude
constraint, K = 0. At the open-loop zeros, GcGp H is zero. To satisfy the magnitude constraint, K = ∞ .
When the order of the denominator is greater than the numerator, that is, when there are more open-loop
poles than zeros, the number of root loci amounting to the difference between the number of open-loop
poles and zeros will terminate at infinity.
Controller Design
With the concepts of the root locus discussed above, a controller may be designed for a given system using
the root locus method to meet required performance specifications. In designing a controller for a system
based on the root locus method, the following guidelines will help in reducing the design iterations:
1. The root locus method should be performed with the aid of a computer program like SciLAB or
MATLAB in order to be effective.
2. The controller structure should be decided first. This is normally done intuitively. Therefore, the control-
ler design for a given system and performance requirement need not be unique.
3. If the controlled plant does not contain an open-loop pole at s = 0 or z = 1, then the steady-state error
will not be zero. An integrator pole should be built into the controller pole–zero structure in such
cases. If the controlled plant contains an open-loop pole at s = 0 or z = 1, then avoid the integrator
pole in the controller pole–zero structure. If it is not possible to place an integrator pole in the control-
ler due to unsatisfactory root loci, then choose the closed-loop poles at as large a value of K as possible
to reduce the steady-state error.
4. A given system may have unstable poles or right half plane zeros. An example of right half plane zero
is the boost converter. An example of a real system having an unstable pole is a magnetic levitation
system where the operating point is inherently unstable but has to be maintained by control action. In
all such applications the closed-loop poles should be brought inside the unit circle in z-plane or the
left of the s-plane by proper selection of the controller pole–zero structure.
5. Poles of the controller cannot be placed outside the unit circle or on the right half of the s-plane.
6. Since the closed performance is determined by the closed-loop poles, to increase the speed of response,
the root loci should be shaped to move closer to the origin in the z-plane or away from the negative
real axis in the s-plane.
7. As far as possible do not select closed-loop pole locations that are close to the boundary of stability. The
gain and other system parameters will vary and may cause the root loci to drift during operation, and as
a result the closed-loop pole location at the selected gain K may drift into the unstable region. To design
robust controllers, select the closed-loop poles that are far from the boundary of stability (i.e., the unit
circle in the case of discrete controller or the imaginary axis in the case of continuous controller).
Chapter 10.indd 575 3/18/2009 1:13:05 PM

8. While determining K, it must be remembered that the number of zeros introduced cannot exceed the
number of poles introduced. Also, complex poles or zeros will always occur in pairs. This implies that
energy is stored in both the potential and the kinetic storage components, resulting in exchange of energy
between them. It never reaches equilibrium if there is no dissipation, that is, if the poles are on the
imaginary axis. If there is a dissipative element in the system, then the pole is away from the imaginary
axis (s-plane) or within the unit circle (z-plane).
9. The pole–zero structure of the controller has to be defined to appropriately shape the root loci.
This is the most uncertain part in the whole design process. This will improve with experience.
However, the following intuitive insight can be used for selecting the pole–zero structure of the con-
troller to shape the root loci. As a general guideline, zeros attract the root loci and poles repel the
loci. Hence to shape a portion of the loci closer to a desired region in the s- or z-plane, zeros may be
used. In a similar manner, controller poles should also be placed. It must be remembered that
addition of a zero means that you must add a pole also to make it practically implementable.
Care should be taken to add just the minimum number of poles and zeros so that the overall order
of the system is not unduly increased which will deteriorate the system dynamics.
10. The origin in the z-plane corresponds to negative infinity region in the s-plane. This means that if all
closed-loop poles are chosen to be close to these regions, the system response will be faster. This also
implies the closed-loop system will be more robust to system parameter variations because the drift in
the root loci in the face of variations in the system parameters will not affect the relative stability sig-
nificantly as the closed-loop poles are far removed from the instability boundary.
Design Steps
Referring to Figure 10.24 and the guidelines discussed above, the controller design procedure using the root
locus method is given in the following steps which are algorithmic in nature. A simple program can be written
and used to automate the steps. This will help in quick iterations and fine tuning the controller.
Step 1: Model the plant in the continuous time domain. This is based on the discussion in Chapter 8.
This will result in the plant transfer function Gp(s).
Step 2: Transform the plant model to the discrete domain using the continuous-to-discrete domain
transformation methods discussed in Chapter 9. The ZOH method is the continuous-to-discrete trans-
formation method that fits most applications that use sample-and-hold (S/H) circuits to perform
analog-to-digital data conversions. At the end of this step, the discrete domain plant transfer function
Gp(z) is obtained. If the controller is to be designed in the continuous domain, this step is skipped. Let
the plant transfer function be
np
Gp = (10.24)
dp
where np is the numerator polynomial of the plant transfer function that denote the zeros and dp is the
denominator polynomial of the plant transfer function that denote the poles.
Step 3: Obtain the transfer function of the feedback sensing and processing circuitry and transform it
from the continuous domain to the discrete domain using the ZOH. Let the feedback block transfer
function be
nh
H= (10.25)
dh
Chapter 10.indd 576 3/18/2009 1:13:05 PM

where nh is the numerator polynomial denoting the zeros and dh the denominator polynomial denoting
the poles.
Step 4: Define the pole–zero structure for Gc(z) or Gc(s) depending on whether the controller design is
in discrete domain or continuous domain, respectively. Let the controller transfer function without the
gain be
n
Gc = c (10.26)
dc
where nc is the numerator polynomial of the controller transfer function that denotes the zeros and dc
is the denominator polynomial of the controller transfer function that denotes the poles.
Step 5: Plot the root locus of the loop transfer function GcGpH with the controller gain K as the param-
eter that is varied from 0 to ∞. Root locus plotting functions are available in computer programs like
SciLAB and MATLAB. The loop transfer function without K is
nc ⋅ np ⋅ nh
Gl = (10.27)
dc ⋅ dp ⋅ dh
Step 6: Locate the closed-loop poles on the root locus by applying the guidelines discussed above. Note
down the value of the controller gain K at the selected closed-loop pole location.
Step 7: With value of gain K, obtain the closed-loop transfer function. The closed-loop transfer function is
K ⋅ Gc ⋅ Gp K ⋅ nc ⋅ np ⋅ d h
Gclose = = (10.28)
1 + K ⋅ Gc ⋅ Gp ⋅ H d c ⋅ d p ⋅ d h + K ⋅ nc ⋅ np ⋅ nh
Step 8: Obtain the step response of the closed-loop transfer function given by Eq. (10.28). The step
response plotting functions are available in computer programs like SciLAB and MATLAB.
Step 9: Observe the step response to check if the performance specifications are met.
Step 10: If performance specifications are not met, Steps (4)–(9) should be iterated till the performance
is satisfactory. In performing the iteration, first iterate by changing the locations of the poles and zeros
of the controller without increasing the order of the system. Only if it is not possible with this pole–zero
structure, should one change the order of the controller to obtain satisfactory response.
In addition to the guidelines discussed above, following are some tips and hints for defining the pole–
zero structure of the controller for achieving suitable root loci:
1. All open-loop poles and zeros are within unit circle:
• If there is no pole at z = 1, place 1 there for zero steady-state error.
• Place a zero such that PI controller structure is obtained.
2. All poles are within the unit circle (z-plane) and some zeros are outside it or all poles are in the left half
of the s-plane and some zeros are on the right half:
• As the closed-loop poles move towards zeros for higher gains, such a system would necessarily be a
low-gain system.
• Such a system is a non-minimum phase system and the Bode plot approach cannot be used.
3. Poles of GpH are outside the unit circle or in the unstable zone:
• Here for low gains the closed-loop poles would be close to the open-loop poles and would therefore
be unstable. Place controller zeros (along with poles) inside the unit circle and pull the loci into the
stable zone.
Chapter 10.indd 577 3/18/2009 1:13:05 PM

• Since the closed-loop poles are needed to be close to the open-loop zeros, such a system would
necessarily be a high-gain system.
4. GpH has an integrator (pole at z = 1):
• To take the steady-state error to 0, an additional integrator is not needed. Hence the controller for
this will be like a PD controller.
1
0.5π/T
0.6π/T 0.4π/T
0.8 0.1
0.7π/T 0.3π/T
0.2
0.6 0.3
0.8π/T 0.4 0.2π/T
0.5
0.4 0.6
0.9π/T 0.7 0.1π/T
0.8
0.2
Imaginary axis
0.9
π/T
0
π/T
−0.2
0.9p /T
0.1π/T
−0.4
0.8π/T 0.2π/T
−0.6
0.7π/T 0.3π/T
−0.8
0.6π/T 0.4π/T
0.5π/T
−1
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real axis
Figure 10.25 Root locus plot of the open-loop uncompensated system.
EXAMPLE 10.2 The example of the speed control of the DC motor that is discussed in Section
10.2 is considered here wherein a controller based on the root locus approach is
designed. The transfer function of the uncompensated DC motor is given as
w (z) 0.0018233 (z + 0.9109)

----- = ---------------------------
V(z) (z − 0.92) (z − 0.8215)
Here the output is the angular speed of the DC motor shaft and the input is the
armature voltage. The armature voltage is the control input that is varied to obtain
the desired angular speed. The root locus plot of the open-loop uncompensated
system is shown in Figure 10.25.
Chapter 10.indd 578 3/18/2009 1:13:05 PM

It consists of two poles and a zero within the unit circle. The root loci go out of the unit circle as shown.
One can start with the controller that was designed with the Bode approach. The controller designed with
the Bode approach is
6( z − 0.9)
Gc ( z ) =
z −1
Using the controller structure without the gain, one obtains
z − 0.9
Gc ( z ) =
z −1
This controller structure is combined with the DC motor plant to obtain the loop transfer function which
is given as
z − 0.9 ⎛ 0.0018233 ⋅ ( z + 0.91099) ⎞
Gloop ( z ) = KGcGp H = K
z − 1 ⎜⎝ ( z − 0.92) ⋅ ( z − 0.8215) ⎟⎠
The root locus of the above loop transfer function with K as the gain parameter is shown in Figure 10.26.
For a gain of K = 3.3, the step response of the closed-loop system is shown in Figure 10.27. Let the controller
structure be changed to a PID controller from a PI controller structure. Introduction of the derivative portion
will significantly improve the transient response of the system. Introduce a zero at z = –0.95 that performs the
1
0.5π/T
0.6π/T 0.4π/T
0.8
0.7π/T 0.3π/T
0.6 0.2π/T
0.8π/T
0.4
0.9π/T 0.1π/T
0.2
Imaginary axis
π/T
0
π/T
0.9
−0.2
0.9π/T 0.8
0.7 0.1π/T
−0.4 0.6
0.5
0.8π/T 0.4 0.2π/T
−0.6 0.3
0.2 0.3π/T
0.7π/T
−0.8 0.1
0.6π/T 0.4π/T
0.5π/T
−1
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real axis
Figure 10.26 Root locus of the plant with PI controller.
Chapter 10.indd 579 3/18/2009 1:13:05 PM

1.4
1.2
0.8
Amplitude
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70 80 90 100
Time (s)
Figure 10.27 Step response of the closed-loop system for a gain of K = 3.3.
derivative action. One cannot introduce only a zero as it will become a non-causal system. Therefore, a pole has
to be introduced. Introduce a pole at z = 0. The PID controller structure is
⎛ z − 0.9 ⎞ ⎛ z − 0.95 ⎞
Gc ( z ) = ⎜ ⎟ ⎜ ⎟
⎝ z −1 ⎠ ⎝ z ⎠
This controller structure is combined with the DC motor plant to obtain the loop transfer function which
is given as
⎛ z − 0.9 ⎞ ⎛ z − 0.95 ⎞ ⎛ 0.0018233 ( z + 0.9109) ⎞

Gloop ( z ) = KGcGp H = K ⎜ ⎟⎜ ⎟⎜ ⎟
⎝ z −1 ⎠ ⎝ z ⎠ ⎝ ( z − 0.92)( z − 0.8215) ⎠
The root locus of the above loop transfer function with gain K as the parameter is shown in Figure 10.28.
The derivative zero is judiciously placed so that the root loci are brought well within the unit circle. The
gain can now be chosen higher to improve the transient response. For a gain K = 227, the step response of
the closed-loop system is shown in Figure 10.29. From the step response as shown in Figure 10.29, it can
be observed that the initial transient is due to the derivative part. The steady-state portion of the response
is due to the integral part that ensures zero steady-state error. The proportional part acts as the bridge
between the derivative part and the integral part. The three operational zones of the PID controller can
be easily distinguished from Figure 10.29.
Chapter 10.indd 580 3/18/2009 1:13:05 PM

1
0.5π/T
0.6π/T 0.4π/T
0.8 0.1
0.7π/T 0.3π/T
0.2
0.6 0.3
0.8π/T 0.4 0.2π/T
0.5
0.4 0.6
0.9π/T 0.7 0.1π/T
0.8
0.2
Imaginary axis
0.9
π/T
0
π/T
−0.2
0.9π/T
0.1π/T
−0.4
0.8π/T 0.2π/T
−0.6
0.7π/T 0.3π/T
−0.8
0.6π/T 0.4π/T
0.5π/T
−1
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real axis
Figure 10.28 Root locus of the plant with PID controller.
1.4
1.2
0.8
Amplitude
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70 80 90 100
Time (s)
Figure 10.29 Step response of the closed-loop system with PID controller structure for a gain K = 227.
Chapter 10.indd 581 3/20/2009 3:07:43 PM

10.5 State Space Method
T he state space model is a more general representation of a system compared to the transfer function
model. The transfer function assumes zero initial conditions and is applicable only for SISO systems.
The state space approach suffers from neither of these limitations. It is applicable for MIMO systems and
can even be used for non-linear and time-varying systems. In the discussions to follow, the discrete domain
state space controller design methods will be used unless otherwise stated, even though the concepts and
principles are equally valid for continuous domain controller design.
The state space equations for a general discrete domain system as discussed in Chapter 9 are in the form
xk +1 = Axk + Buk (10.29a)
yk = Cxk + Duk (10.29b)
Equation (10.29a) gives the next state, given the present state and the input. Equation (10.29b) gives the
output, given the present state and the inputs. Together the two equations are called the state equations for
the system where xk is the state vector at the discrete time instant kT with sampling time T, uk the input
vector at the discrete time instant kT; A the characteristic matrix for the system; B the input matrix; C the
output matrix; D the feed-through matrix.
Any system can be represented in the state space form as indicated above. Chapter 8 gives a detailed
discussion on modeling and Chapter 9 discusses the conversion of continuous domain models to discrete
domain models. Computer programs like SciLAB and MATLAB have functions that will do these
continuous-to-discrete domain transformations thereby eliminating the tedious manual conversion pro-
cesses. These could be used during the design process.
For any physical system, there are two important conditions that must be met: (a) controllability and
(b) observability. If these two conditions are not satisfied then one should re-look at the model and include
certain non-idealities that may have been overlooked.
Controllability
A system represented by xk +1 = Axk + Buk is controllable if there exists a sequence of inputs
{u0 , u1 , u2 , …, un−1} such that it transfers the system from any initial state to any final state xn with n being
finite.
From the state equation of the system,
x1 = Ax0 + Bu0
x 2 = Ax1 + Bu1
⇒ x 2 = A 2 x0 + ABu0 + Bu0
By extending to state at nth time interval, one obtains
⎡ un −1 ⎤
⎢u ⎥
⎢ n−2 ⎥
xn = A n x0 + ⎡⎣B AB A 2 B … A n −1B ⎤⎦ ⎢un −3 ⎥
⎢ ⎥
⎢ ⎥
⎢⎣ u0 ⎥⎦
Chapter 10.indd 582 3/18/2009 1:13:05 PM

⎡ un −1 ⎤
⎢u ⎥
⎢ n−2 ⎥
( xn − A n x0 ) = ⎡⎣B AB A 2 B … A n −1B ⎤⎦ ⎢un −3 ⎥
⎢ ⎥
⎢ ⎥
⎢⎣ u0 ⎥⎦

⎡ un −1 ⎤
⎢u ⎥
⎢ n−2 ⎥ −1
⎢un −3 ⎥ = ⎡B AB A 2 B … A n −1B ⎤⎦ ( xn − A n x0 )
⎢ ⎥ ⎣
⎢ ⎥
⎢⎣ u0 ⎥⎦
It can be seen from the above expression that to transfer the system from any state x0 to another state xn, it
is possible to find a sequence of inputs to achieve the task only if the matrix
⎡B AB A 2 B … A n−1B ⎤
⎣ ⎦
is invertible. This matrix is called the controllability matrix. This implies that any system can be controlled
only if the controllability matrix is invertible.
For any real system modeled correctly, the matrix is always invertible and the system is therefore control-
lable. If the inverse of the controllability matrix does not exist, it implies that the model of the system is
improper.
Observability
A system represented by xk +1 = Axk + Buk as the state equation and yk = Cxk as the output equation
wherein there is no direct feed-through to the output from the input is said to be observable provided that
the initial state x0 can be calculated from n measurements y0, y1, …, yn–1 with n being finite.
This is actually the dual of controllability. Let the system be transferred to any state xn by application of
a certain sequence of inputs. Then the output at nth time interval is
yn = Cxn
yn −1 = Cxn −1

y0 = Cx0
The above equations can be represented as
⎡ y0 ⎤ ⎡ C ⎤
⎢ y ⎥ ⎢ CA ⎥
⎢ 1⎥ ⎢ ⎥
⎢ y 2 ⎥ = ⎢CA 2 ⎥ x0
⎢ ⎥ ⎢ ⎥
⎢⎥ ⎢ ⎥
⎢⎣ yn ⎥⎦ ⎢CA n ⎥
⎣ ⎦
Chapter 10.indd 583 3/18/2009 1:13:05 PM

The initial state x0 can be obtained as

−1
⎡ C ⎤ ⎡ y0 ⎤
⎢ CA ⎥ ⎢y ⎥
⎢ ⎥ ⎢ 1⎥
⎢
x0 = CA 2⎥
⎢ y2 ⎥
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢⎥
⎢ n⎥
⎣CA ⎦ ⎣⎢ yn ⎥⎦
The matrix
−1
⎡ C ⎤
⎢ CA ⎥
⎢ ⎥
⎢CA 2 ⎥
⎢ ⎥
⎢ ⎥
⎢ n⎥
⎣CA ⎦
is called the observability matrix. Only if the inverse of the observability matrix exists, can any initial state
be computed from n measurements. If the inverse of the observability matrix does not exist, then the system
is not observable which again implies incorrect system model.
10.6 Full-State Feedback
I n the full-state feedback control, all the state variables of the system are available and therefore fed back to
the controller. As the states contain all the information about the plant, the controller does not need
additional poles and zeros to include more degrees of freedom. Therefore, the controller needs to be just
simple gains for each input. This implies that in the case of the full-state feedback design the order of the
system remains unchanged as the controller is only a gain matrix.
In the case of the classical root locus method discussed in the previous section, the closed-loop poles
were placed on the root loci portion that is within the unit circle. However, in the case of the full-state feed-
back control, due to the availability of full-state information, there is increased flexibility in the controller
design in that the closed-loop poles of the system can be placed anywhere within the unit circle. As a conse-
quence this is a very powerful design tool.
Figure 10.30 depicts the block diagram of a system represented in the state space form with full-state
feedback. The plant is represented in the state space form as given by Eqs. (10.29a) and (10.29b). The
block schematic representation of these equations is as shown within the marked square in Figure 10.30.
The block interconnections are shown as double-lined arrows to indicate that the variables are vectors. The
present state vector xk is fed back and compared with the reference vector r. The difference between the
reference vector and the fed back state vector results in the error vector ek. This error vector is given as the
input to the controller K which is a gain matrix. The output of the controller becomes the control input to
the plant.
The state space controller is generally classified by the variable that is fed back to the controller and the
reference variable. The classification is called the control law. The control law for the above system is
uk = K (r − xk ) (10.30a)
Chapter 10.indd 584 3/18/2009 1:13:05 PM

Plant
r xk+1 xk
+
−
K B ∑ z−1 C ∑
yk
uk
Figure 10.30 Block diagram describing the full-state feedback.
Further, there are two classes of the controller based on the nature of the reference waveform. They are
1. regulator;
2. tracking.
Regulator
In this case the output needs to be maintained at a fixed set point. The reference in this case is a constant R.
From Eq. (10.30a), the control law for the regulator is
uk = K ( R − xk ) (10.30b)
Substituting the control law in the system x k +1 = Axk + Buk , one obtains
x k +1 = ( A − BK )xk + BKR (10.30c)
The system dynamics is decided by (A – BK ) where K is the controller. The controller should perform for
any value of the constant reference R. Therefore, without loss of generality, the controller designed for R = 0
is valid for controller designed for any R. Thus for the regulator design, the control law is simplified by
considering R = 0. Thus the control law for regulator design is given as
uk = − Kx k (10.30d)
The block schematic of the regulator is shown in Figure 10.31. Figure 10.31(a) shows the regulator structure
and Figure 10.31(b) shows the control structure used for controller K design considering R = 0.
Tracker
This is a more versatile control structure as compared to the regulator control structure. In this the reference
is not fixed but changes with time. This corresponds to large-signal changes in the output. Equation (10.30a)
r=R y r=0 y
e
+ K Plant −K Plant
− u x x
(a) (b)
Figure 10.31 (a) Regulator control structure; (b) regulator structure for controller design with R = 0.
Chapter 10.indd 585 3/18/2009 1:13:06 PM

is itself the control law where the input and state vectors represent large-signal variations. The controller has
to be designed using the large-signal model of the system. Figure 10.32 illustrates the tracker control struc-
ture. The large-signal state variable is fed back and compared with the reference r. As r will have the vector
dimensions of the controlled outputs y, the block matrix Nx is used to make r compatible with the vector
dimensions of the state x. To improve the dynamics, a feed-forward control can also be implemented as
shown by directly feeding a part of r to the control input u through the block matrix Nu.
The tracker problem can be considered as the regulator problem plus a feed-forward portion. The con-
troller is designed to take care of the variations of the state variables in the neighborhood of the operating
point using the small-signal model of the system. The control input to the plant is the sum of the constant
reference plus the correction action to reject the perturbations in the neighborhood of the operating point.
Thus,
u =U + u = R + u

where R is the constant reference indicating nominal operating point and u is the small-signal variations of
the control input about the nominal control input U = R. The feedback state variable xk represents the
Eq. (10.30d),
small-signal variations about the operating point and the small-signal reference is zero. From
the control law for the regulator problem is
uk = − Kxk

where xk = xk − X with xk being the actual large-signal state value and X the state value at the nominal
point. Figure 10.32 illustrates the tracker control structure by splitting the tracker problem of
operating
Figure 10.32 into the regulator problem and the feed-forward portion.
Figure 10.33(a) shows the block schematic of the regulator plus feed-forward control. The controller K
corrects for the state perturbation around the operating point. In the steady state, xk will become zero. X is
the steady-state value or the nominal operating point corresponding to the input U = R. The steady-state
value X is obtained from the steady-state model of the plant with U = R as the input. Figure 10.33(b) shows
a simplified regulator plus feed-forward control structure.
To design the controller K, the small-signal model of the plant is sufficient as the controller handles only
the small-signal dynamics. The control block diagram used for designing the controller K is shown in
Figure 10.34 wherein the plant is only the small-signal model.
Feed-forward
control
Nu
+ u
Nx + K + Plant y
r − u
~ x
Figure 10.32 Block diagram of the tracker control structure.
Chapter 10.indd 586 3/18/2009 1:13:06 PM

U=R
+ u
+ K + Plant y
−
∼r = 0
u
∼
x
+
x
∼ −
(Small signal variation X
about operating (State at nominal
point) operating point)
(a)
U=R
+ u
−K + Plant y
u
∼
x
+
∼x −
X
(b)
Figure 10.33 (a) Block diagram of regulator plus feed-forward control structure;
(b) reduced regulator plus feed-forward control structure.
Plant
−K (small signal y
u ~
~ model) x
~
x
~
Figure 10.34 Block diagram for controller design for tracker problem.
10.7 Regulator Design by Pole Placement

T his section discusses the controller design for the regulator control problem. The control law is given in
Eq. (10.31) as
uk = − Kxk (10.31)
The plant model is given as
xk +1 = Axk + Buk (10.32a)
yk = Cxk + Duk (10.32b)
Chapter 10.indd 587 3/18/2009 1:13:06 PM

Equation (10.32a) gives the dynamics of the system. The input to this system is as given in Eq. (10.31).
Incorporating the control law into the state equation [Eq. (10.32a)], one obtains
xk +1 = Axk − BKxk = ( A − BK )xk (10.33)
The closed-loop dynamics of the system is determined by A – BK. From Eq. (10.33), the characteristic equa-
tion is obtained as the determinant of zI − ( A − BK ). Thus,
zI − ( A − BK ) = 0 (10.34)
This gives the characteristic equation of the full-state feedback system in the following equivalent polyno-
mial form as
ξp ( z ) = z n + βn −1z n −1 + + β1z + β0 = 0 (10.35)
The coefficients bi will be functions of K and the parameters of the system. The roots of Eq. (10.35) give the
closed-loop pole location of the full-state feedback system.
From the model of the plant, the order of the system is known and therefore the number of closed-loop
poles to be selected is also known. Let the closed-loop poles for the full-state feedback system be selected as
l1, l2, …, ln within the unit circle. From the selected or desired closed-loop pole location values, the
desired characteristic equation is
( z − λ1 )( z − λ2 )( z − λ3 )... = 0 (10.36)
which can be represented in the equivalent polynomial form as
ξd ( z ) = z n + αn −1z n −1 + + α1z + α0 = 0 (10.37)
where ai are functions of the desired closed-loop pole locations.
Equation (10.35) gives the characteristic equation that relates with the system parameters and the con-
troller gain K. Equation (10.37) gives the characteristic equation that relates with the chosen or desired
closed-loop pole locations for the full-state feedback system. Equating these two characteristic equations and
comparing the coefficients, a set of n simultaneous linear equations is obtained with K as the unknown (K is
the gain matrix containing n gains). The solution to this set of simultaneous equations gives the values of the
controller gain matrix K which will result in the selected or chosen closed-loop pole locations, l1, l2, …, ln.
Computer programs can be used to obtain the solution for the set of n simultaneous equations. One of the
algorithms is based on the Ackerman’s formula which is given as
−1
K = [ 0 0 ... 0 1] ⎡⎣B AB A 2 B ... A n −1B ⎤⎦ ⋅ ξd ( A ) (10.38)
where ξd ( A ) = A n + αn −1 A n −1 + + α1 A + α0 is obtained from the desired characteristic equation of ξd ( z )

by replacing the variable z by A. The design is algorithmic in nature and summarized in the following
steps:
Step 1: Model the plant in the continuous domain.
Step 2: Convert the model from the continuous-to-discrete domain as discussed in Chapter 9 and
obtain the A, B, C and D matrices.
Step 3: Select all the desired closed-loop poles locations within the unit circle of the z-plane. The closer the
poles are to the origin of the unit circle, faster is the speed of response. In the limit, if all the closed-loop poles
are placed at the origin, then the system becomes an nth-order delay. This is the fastest response that one can
obtain from the system. Such a controller where all the li = 0 is called a dead beat controller. It should,
however, be noted that closer the closed-loop poles to the origin, larger will be the resulting controller gains.
This implies high control energy requirement which the plant should be capable of providing.
Chapter 10.indd 588 3/18/2009 1:13:06 PM

Step 4: Obtain the controller gains K by providing A, B and l as inputs to the Ackerman’s formula or
an equivalent algorithm for solving the n simultaneous equations.
Step 5: Obtain the step response of the closed-loop system. Use a computer program to plot the step
response.
Step 6: Iterate Steps (3)–(5) till the step response satisfactorily meets all the performance requirements.
EXAMPLE 10.3 Consider the example of the DC motor speed control, the model of which is
described in Section 10.2. The controller for the discrete domain state space model
given in the above-mentioned example is designed here. For the parameters of the
DC motor given in the earlier section, the state space model is given as
⎡0.92294 − 0.00348⎤
A =⎢
⎣ 0.08698 0.81856 ⎥⎦
⎡ 0.03844 0.00182 ⎤
B =⎢
⎣0.00182 − 0.90629 ⎥⎦
C = [ 0 1]
D = [ 0 0]
x = [ia; ω] u = [ v; TL ]
where x is the state vector and u is the input vector.
The plant has two inputs: (a) v – the armature voltage that actually controls
the speed, (b) TL – the load torque. The load torque acts as a disturbance input to
the system. The control input is only the armature voltage. Hence the control
input vector is [v] considering the disturbance input TL = 0 for purposes of con-
troller synthesis without loss of generality. As it is a second-order system, let the
desired closed-loop poles be chosen as z = 0 and z = 0. This means that both the
closed-loop poles are placed at the origin of the unit circle. This will result in dead
beat control and will also give the fastest response. As the order of the system is
two, in two times the sample time, the system will reach steady state. The gains are
computed by means of a computer program which solves the Ackerman’s formula.
If the function computing the Ackerman’s formula is called “acker”, then
K = acker(A, B, P)
where P is the vector of desired closed-loop pole locations and in this case
P = [0, 0]. The state and the input responses of the closed-loop system are given in
Figure 10.35. The value of the controller gain vector is
K = [35.22729, 212.4682]
From Figure 10.35, it can be observed that the steady state is reached in two sample
times. As the reference input in the regulator design is assumed to be zero, the states
will reach the zero reference value in two sample times. Figure 10.36 shows the state
and input response for an arbitrary location of the closed-loop poles within the unit
circle. For z = 0.1 and z = 0.3 as the desired closed-loop pole locations,
K = [29.29581, 118.1327]
Chapter 10.indd 589 3/18/2009 1:13:06 PM

State error and input for desired closed-loop pole location at z = 0 and z = 0
1 250
w
0 200
ia V (armature voltage)
−1 150
−2 100
−3 50
State
Input
−4 0
−5 −50
−6 −100
−7 −150
K = [35.22729 212.4682]
−8 −200
−9 −200
0 5 10 15 20 25 0 5 10 15 20 25
No. of samples, n No. of samples, n
Figure 10.35 State error and input for dead beat controller.
State error and input for desired closed-loop pole location at z = 0.1 and z = 0
1 100
w
V (armature voltage)
0
ia 50
−1
0
State
Input
−2
−50
−3
K = [29.29581 118.1327] −100

−4
−5 −150
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Figure 10.36 State error and input for arbitrary closed-loop pole placement.
Chapter 10.indd 590 3/18/2009 1:13:06 PM

10.8 Estimator Design
I n full-state feedback, it has been considered that all the state variables of a system are measurable. In many
practical situations all the state variables may not be easily measurable or may not be accessible due to
physical limitations. In such cases, the states of the system will have to be estimated based on the informa-
tion available in the measurable states. Such a unit is called an estimator or an observer. The estimator or
observer that observes the state values is in actuality a replacement for the physical sensors that would gener-
ally be used in the case of measurable states.
A model of the plant, which should be proper, is run alongside the plant. Control inputs are given to
this model as well as the plant. The controller uses the estimated state variable values to compute the control
action. The output of the plant model is compared with the actual plant output and the error is passed
through a gain vector G called the estimator gain. The output of the estimator is used to apply corrective
action to the estimator states.
Figure 10.37 depicts the full-state feedback system with an estimator. Only few of the states may be
measurable. The measured states are brought out as the system output yk as indicated. The estimator gives
an estimate of the measured output ŷk . The difference between the measured output and the estimated
output is passed through an estimator gain G and given as a local corrective negative feedback to the estima-
tor as shown. The estimator uses the plants model A, B, C and D and the actual input to the plant u to
estimate the states x̂k . The estimated x̂k is fed back to the controller K to perform the control action.
Here G is a simple gain vector. The variables with the hat “^” represent the estimated values. From
Figure 10.37, it can be observed that the state equation for the estimator is
xˆk +1 = Axˆk + Buk + G ( yk − yˆk ) (10.39)
On re-arranging, one obtains
xˆk +1 = ( A − GC )xˆk + Buk + Gyk (10.40)
Equation (10.40) is the estimate equation. It is called the prediction estimator. x̂k represents the estimator
state and uk and yk form the inputs to the estimator. The estimator dynamics is determined by A – GC.
yk
Actual plant
+
−
D
∧
yk
∧ ∧
r + X k+1 Xk
+ K B ∑ z −1 C ∑
− +
uk
+
A
Estimator
Figure 10.37 The schematic of the full-state feedback system with an estimator.
Chapter 10.indd 591 3/18/2009 1:13:06 PM

Estimator Error Dynamics

The estimator error gives a measure of the closeness of the actual state and the estimated state. Defining the
estimator error as the difference between the actual state variable and the estimated state variable, one
obtains
ek = xk − xˆk (10.41)
ek +1 = xk +1 − xˆk +1
= Axk + Buk − ( A − GC )xˆk − Buk − GCxk
= ( A − GC )ek (10.42)
Equation (10.42) gives the error dynamics of the system which is same as the estimator dynamics. It is
desired that the error should not increase with time. Hence if the estimator is stable, the error will also be
stable and decay to zero. The equation is of a similar form as A – BK which gives the closed-loop dynamics of
the full-state feedback system. The estimator may be viewed as a local closed-loop system with dynamics
determined by A – GC. The method described to design the controller K for the full-state feedback can be
used to find the value of G.
The controller K is designed using the Ackerman’s formula or a similar algorithm. In the controller
design, the closed-loop dynamics is determined by A – BK where A is the n × n matrix for an n-state system,
B the n × m matrix for m control inputs, K the m × n matrix. In the case of the estimator, the estimator
dynamics is decided by A – GC where A is the n × n matrix, C the m × n matrix (in a practical system there
should be one control input for every controlled output; thus for m control inputs, the maximum number
of controlled outputs is m), G the n × m matrix.
To dimensionally match with the components of A – BK,
( A − GC )T = A T − C TG T (10.43)
By transposing A – GC, the dynamics of the estimator is not affected, but it is dimensionally trans-
formed exactly to the form of A – BK. Now the same Steps (3)–(6) given in Section 10.7 can be applied
to design the estimator gain GT by using AT, CT and the desired estimator closed-loop pole locations.
The design steps are:
Step 1: Select all the desired closed-loop poles locations λe of the estimator within the unit circle of the
z-plane. As a general guideline, the estimator should be faster than the plant. So for obtaining G, the
closed-loop poles for the estimator will be chosen closer to the origin compared to the closed-loop poles
chosen for the controller K design.
Step 2: Obtain the estimator gains G by providing AT, CT and λe as inputs to the Ackerman’s formula
as given in Eq. (10.38) or an equivalent algorithm for obtaining GT. In Ackerman’s formula A is replaced
by AT, B is replaced by CT and K is replaced by GT.
Step 3: Obtain the error step response of the estimator system. Use an appropriate computer program
to plot the step response.
Step 4: Iterate Steps (1)–(3) till the step response satisfactorily meets all the performance
requirements.
Chapter 10.indd 592 3/18/2009 1:13:07 PM

EXAMPLE 10.4 For the DC motor plant, the prediction estimator gain G is designed by first plac-
ing the estimator poles. The fastest estimation is done when all closed-loop esti-
mator poles are at the origin. In general, the estimator poles should be placed
closer to the origin than the controller closed-loop poles. Figure 10.38 shows the
error between the actual and the estimated states for estimator poles placed at the
origin and at 0.2. Here again the Ackerman’s formula is used. The function is
called with the following parameters to the Ackerman’s function:
GT = acker(AT, CT, Pe)
where Pe is the vector of desired closed-loop estimator pole locations.
1 1
0
0
−1
−2 −1
State error
State error
−3 −2
−4
−3
−5
GT = [9.7895 1.7415] GT = [6.0051 1.3415]
−6 −4
−7
−5
−8 Estimator closed-loop poles = [0,0] Estimator closed-loop poles = [0.2, 0.2]
−9 −6
0 5 10 15 20 25 1 2 3 4 5 6 7 8 9 10 11
Figure 10.38 State errors for two different estimator closed-loop pole locations.
Separation Principle
In the full-state feedback system, there are two sets of dynamics: (a) closed-loop plant dynamics as decided
by A – BK and (b) the estimator dynamics as decided by A – GC. The question that needs to be answered is
whether there is any sequence in which the estimator and controller gains should be designed? For example,
should the estimator gain be designed first and then the controller gain or the other way round? Or can both
be designed separately and independent of each other for a given system?
For a system that is represented as
xk +1 = Axk + Buk
applying the regulator control law uk = − Kxk , the closed-loop system with full-state feedback is given as
xk +1 = Axk − BKxk = ( A − BK )xk

Instead of actual state feedback, if the estimated states are fed back then
xk +1 = Axk − BK xˆk (10.44)
Chapter 10.indd 593 3/18/2009 1:13:07 PM

Due to the use of the estimated states rather than the actual states for feedback, the state error is ek = xk − xˆk
and therefore, x̂k = xk − ek . Substituting this in Eq. (10.44), the system state equation is given as
xk −1 = Axk − BKxk + BKek (10.45)
where the error becomes an implicit input to the system. The error dynamics is given by Eq. (10.42).
Considering xk and ek as the composite state of the system, the overall system dynamics with the estimator is
given by the following state equation which is a combination of Eqs. (10.42) and (10.44).
⎡ xk +1 ⎤ ⎡ A − BK BK ⎤ ⎡ xk ⎤
⎢e ⎥ = ⎢ 0 A − GC ⎥⎦ ⎢⎣ ek ⎥⎦
(10.46)
⎣ k +1 ⎦ ⎣
From the state equation [Eq. (10.46)], the characteristic equation may be written as
zI − A + BK zI − A + GC = 0 (10.47)
If λ1 , λ2 , …, λn are the closed-loop pole locations of the controller and λe1 , λe 2 , …, λen are the estimator
pole locations, then from Eq. (10.47), the characteristic equation can be represented as
{( z − λ1 )( z − λ2 ).....( z − λn )}{( z − λe1 )( z − λe 2 ).....( z − λen )} = 0 (10.48)

Closed-loop poles Estiimator poles
Equation (10.48) is the characteristic equation of the complete full-state feedback system with estimator and
it can be split into the following two equations:
{( z − λ1 )( z − λ2 ).....( z − λn )} = 0
Closed-loop poles
{( z − λe1 )( z − λe 2 ).....( z − λen )} = 0

Estimator poles
It can be observed that the choice of the closed-loop poles is not dependent on the choice of the estimator
poles and vice-versa. This implies that the controller design K and the estimator gain design G can be done
separately and independent of each other. This is called the “Separation Principle”.
Current Estimator
The state equation for the prediction estimator is given in Eq. (10.40) and is reproduced here:
xˆk +1 = ( A − GC )xˆk + Buk + Gyk
In the prediction estimator, the values of the output yk (i.e., the measured states) correspond to the previous
time interval to estimate the next state estimate. In other words, the measured states of the previous time
interval are used to predict the next state values. Hence, it is called “prediction estimator”. In the case of the
current estimator, the latest or the measured states of the current time interval are used to correct the
estimate in the current time interval. Therefore, the name “current estimator”. The current estimator is a
two-step process. The two steps are:
1. time update;
2. measurement update.
During the time update process, the next state of the state vector is estimated, that is, the (k + 1)th state is
estimated using the kth values of the state and measured outputs. This process of estimate update from k to
k + 1 is called the time update. Next, the fresh values of the outputs in the current time interval [i.e., (k + 1)]
Chapter 10.indd 594 3/18/2009 1:13:07 PM

is used to correct the predicted estimate to obtain the current estimate. This process is called the measure-
ment update. Thus the two processes of the current estimation are:
1. Time Update:
xk +1 = Axˆk + Buk (10.49)
where xk +1 is the time-updated estimate.
2. Measurement Update:
xˆk +1 = xk +1 + G ( yk +1 − Cxk +1 ) (10.50)
where the intermediate time-updated state value xk +1 is corrected with fresh or current measurement yk + 1
to obtain the current estimate x̂k +1 .
Substituting Eq. (10.49) into Eq. (10.50), one obtains
xˆk +1 = ( A − GCA )xˆk + Gyk +1 + ( B − GCB )uk (10.51)
This is the consolidated current estimator’s estimate equation. The error dynamics for the current estimator
is obtained in a similar manner as discussed for the simple prediction estimator. The error is given as
ek = xk − xˆk
ek +1 = xk +1 − xˆk +1
ek +1 = Axk + Buk − ( A − GCA )xˆk − Gyk +1 − ( B − GCB )uk

ek +1 = ( A − GCA )ek (10.52)
Equation (10.52) gives the error dynamics of the system. The design of the current estimator gain is exactly
similar to the prediction estimator gain. By transposing A – GCA, the dynamics of the estimator is not
affected, but it is dimensionally transformed exactly to the form A – BK:
( A − GCA )T = A T − ( A TC T )G T
The same steps of the predication estimator design can be applied to design the estimator gain GT by using
AT, (ATCT) and the desired estimator closed-loop pole locations. The design steps are as follows:
Step 1: Select all the desired closed-loop poles locations λe of the estimator within the unit circle of the
z-plane. The estimator should be faster than the plant. So for obtaining G, the closed-loop poles for the
estimator will be chosen closer to the origin compared to the closed-loop poles chosen for the controller
K design.
Step 2: Obtain the estimator gains G by providing the AT, CT and λe as inputs to the Ackerman’s for-
mula as given in Eq. (10.38) or an equivalent algorithm for obtaining GT. In the Ackerman’s formula
A is replaced by AT, B is replaced by ATCT and K is replaced by GT.
It should be noted that the current estimator is the basis for many optimal estimators. The principles of the
current estimator such as time update and measurement update are used in optimal estimators like the
Kalman filter as is discussed in the next chapter.
Chapter 10.indd 595 3/18/2009 1:13:07 PM

EXAMPLE 10.5 For the DC motor plant, the current estimator gain G is designed by first placing
the estimator poles. Figure 10.39 shows the error between the actual and the esti-
mated states for estimator poles placed at 0.2. Here again the Ackerman’s formula
is used. The function is called with the following parameters to the Ackerman’s
function:
GT = acker(AT, ATCT, Pe)
where Pe is the vector of desired closed-loop estimator pole locations.
−1
State error
−2
GT = [6.5101 0.94707]
−3
−4
Estimator closed-loop poles = [0.2, 0.2]
−5
1 2 3 4 5 6 7 8 9 10 11
No. of samples, n
Figure 10.39 State errors for current estimator.
Reduced Order Estimator

If some of the state variables are available for measurement, it makes sense to estimate only the remaining
state variables instead of estimating all the state variables. The state variables that are measured are fed back
as it is and the non-measurable state variables are estimated through an estimator and the estimated states
along with the measured states are fed back. This reduces the errors because some of the state values are the
actual values existing in the plant. Such an estimator that estimates only part of the state vector is called a
“reduced order estimator”.
Let the state vector be partitioned into two parts: (a) a part xm where the state variables are measurable
and (b) a part xe where the state variables need to be estimated. Thus,
⎡ xe ⎤
x=⎢ ⎥
⎣ xm ⎦
The system xk +1 = Axk + Buk can now be represented as
⎡ xe ⎤ ⎡ xe ⎤
⎢ x ⎥ = A ⎢ x ⎥ + Buk (10.53)
⎣ m ⎦k +1 ⎣ m ⎦k
Chapter 10.indd 596 3/18/2009 1:13:07 PM

where
⎡ Aee Aem ⎤ ⎡ Be ⎤ ⎡ xe ⎤
A=⎢ ⎥ , B = ⎢ ⎥, x = ⎢ ⎥
⎣ Ame Amm ⎦ ⎣Bm ⎦ ⎣ xm ⎦
Now,
⎡ xe ⎤ ⎡ Aee Aem ⎤ ⎡ x e ⎤ ⎡ Be ⎤
⎢x ⎥ = ⎢ A + u
Amm ⎥⎦ ⎢⎣ x m ⎥⎦k ⎢⎣Bm ⎥⎦ k
(10.54)
⎣ m ⎦k +1 ⎣ me
From Eq. (10.54), the state equation for the state variables to be estimated is given as
x e k +1 = Aee x e k + Aem x m k + Be uk

Bnunk
Aem x m k + Be uk can equivalently be represented as Bn unk because xmk is measurable. Thus
⎡ xmk ⎤
Bn unk = [ Aem
Be ] ⎢ ⎥ (10.55)
⎣ uk ⎦
From Eq. (10.54), the state equation for the measurable state variables is given as
x m k +1 = Amm x m k + Ame x e k + Bm uk
x m k +1 − Amm x m k − Bm uk = Ame x e k
x m k +1 − Amm x m k − Bm uk = Ame x e k

ynk C n x ek
The system given by Eq. (10.54) can now be represented only in terms of the state variables that need to be
estimated. A full-order state estimator can be built for x e with un as the equivalent input and yn as the corre-
sponding estimated output. This estimated output will be compared against the actual value of yn which is
completely measurable. As yn includes a k + 1 term, which lies in the future, the rest of the estimator inputs
are delayed by one time instant to make this the current time instant term.
The system given by Eq. (10.54) is now represented as
x e k +1 = Aee x e k + Bn unk
ynk = C n x e k (10.56)
where
⎡ xmk ⎤
unk = ⎢ ⎥; Bn = [ Aem Be ] ; C n = [ Ame ]
⎣ uk ⎦
The full-order prediction or current estimator for the system given by Eq. (10.56) is as explained in the pre-
vious sections. The design steps are as follows:
Step 1: Select all the desired closed-loop poles locations λe of the reduced order estimator within the
unit circle of the z-plane. The estimator should be faster than the plant. So for obtaining G, the closed-
loop poles for the estimator will be chosen closer to the origin compared to the closed-loop poles chosen
for the controller K design.
Chapter 10.indd 597 3/18/2009 1:13:07 PM

Step 2: Obtain the estimator gains G by providing the Aee T , C T and

n λe as inputs to the Ackerman’s
formula as given in Eq. (10.38) or an equivalent algorithm for obtaining GT. In the Ackerman’s formula
T , B is replaced by C T and K is replaced by GT.
A is replaced by Aee n
EXAMPLE 10.6 For the DC motor plant, let the armature current be estimated and the shaft
angular speed be measured. There is only one state to be estimated, the armature
current. The reduced order estimator gain G is designed by placing the estimator
pole at z = 0.3. Figure 10.40 shows the error between the actual and the estimated
state. Here again the Ackerman’s formula is used. The function is called with the
following parameters to the Ackerman’s function:
GT = acker(Aee
T , AT A T , P )
ee me e
where Pe is the vector of desired closed-loop estimator pole location.

1
0.9
0.8
0.7
0.6
State error
0.5
0.4
0.3 GT = [7.1617]
0.2
0.1
Estimator closed-loop pole = [0.3]

0
0 2 3 4 5 6 7 8 9 10 11
No. of samples, n
Figure 10.40 State errors for reduced order estimator.
Chapter 10.indd 598 3/18/2009 1:13:08 PM

10.9 Tracker: Controller Design
T he controller designed for the regulator problem drives the state values to 0 or a constant reference R.
The task of the controller now consists of tracking a varying reference.
Figure 10.41 shows the block diagram of the tracking system. Here r is the reference input which may
vary. The length of the reference vector r will be same as the number of controlled outputs. The length of the
input vector also will be the same as the number of controlled outputs as each controlled output needs at
least one control input. The matrix Nx makes the r vector compatible with the state vector. This part of the
input handles the disturbances of the system in the neighborhood of the operating point. The Nu matrix is
responsible for setting the operating point so that the controller has a 0 output when the system is at steady
state. Thus, the controller is inactive when the system has reached steady state since the required control
input to maintain the output at r is provided by the Nu matrix gain. To find the values of Nx and Nu, one
may proceed as follows. At steady state,
y = r ; y = CX; X = Nx*r
because the error e is 0. Therefore
CNx = I (10.57)
The control law is given as
uk = KN x r − Kxk + N u r = ( KN x + N u )r − Kxk (10.58)
Substituting the control law given above into the system equation xk +1 = Axk + Buk , one obtains
xk +1 = ( A − BK )xk + B ( KN x + N u )r
The feedback from the state elements is not affected by the addition of these two matrices and hence the
system dynamics remain the same and are dictated by A – BK.
Since in steady state, e = 0, uk = Nu × r
x ∞+1 = x ∞ = Ax ∞ + Bu∞
( A − I )x ∞ + Bu∞ = Θ
Substituting x ∞ = N x r and u∞ = N u r , one obtains
Nu D
r + xk+1 xk yk
Nx + K + B ∑ z −1 C ∑
Nxr − e uk
∧
xk
Estimator
Figure 10.41 Block schematic of the tracking control system.
Chapter 10.indd 599 3/18/2009 1:13:08 PM

( A − I )N x + BN u = Θ (10.59)
−1
⎡ N x ⎤ ⎡ A − I B ⎤ ⎡Θ ⎤ (10.60)
⎢N ⎥ = ⎢ C 0 ⎥⎦ ⎢⎣ I ⎥⎦
⎣ u⎦ ⎣
The controller K design has the same steps as discussed previously and the system dynamics are dictated by
A – BK. The full-state feedback is used and the feedback state vector can consist of only measured states or
all estimated states through a full-order estimator or part measured states and part estimated states through
a reduced order estimator. The Nx and Nu matrices for tracking input are decided by Eq. (10.60).
EXAMPLE 10.7 Consider the example of the speed control of the DC motor discussed in the
previous examples. The controller gain K is calculated based on A – BK dynamics
using the Ackerman’s formula as discussed in the regulator controller design
example. Only the scaling matrices Nx and Nu need to be determined. Applying
Eq. (10.60) and with the help of a computer program, the Nx and Nu matrices for
the DC motor example are
⎡2⎤
Nx = ⎢ ⎥
⎣1⎦
N u = [ 4.1]
Referring to the control block diagram shown in Figure 10.41, there is only one
controlled output or performance output (i.e., the angular speed, w). Therefore
the reference vector consists of only one variable. The control input vector is also
only one variable, that is, the armature voltage. Thus,
⎡ 2 ⋅ ω ref ⎤
x ref = N x ⋅ r = N x ⋅ ω ref = ⎢ ⎥
⎣ ω ref ⎦
This x ref is compared with the fed back state vector x. The error is fed as input to
the controller K. The control input u is given as
uk = N u ⋅ r + ek K = N u ⋅ ωref + ( N x ⋅ ωref − xk )K
The fed back state vector xk can be directly fed back or obtained from an estimator.
10.10 Controlling Voltage
T he previous sections discussed the methods for designing the controllers based on the general control
structure as depicted in Figure 10.1. If the plant is an inverter generating an AC voltage, then sensing
the AC voltage and feeding back the voltage is generally done with the help of potential transformers. If the
plant is a DC–DC converter, then voltage sensing is done using opto-couplers.
Chapter 10.indd 600 3/18/2009 1:13:08 PM

The general control structure for the converters is shown in Figure 10.41 where the plant considered is
a typical flyback converter. The output voltage Vo is the controlled output and the duty cycle for the power
switch is the control input. The output voltage is sensed by a sensing and isolation circuit and fed back to
the controller. The fed back voltage is compared with the reference Voref. The error between the reference and
the fed back signals is given as input to the PI or PID controller. The controller output is a voltage signal
which has to be converted to time information. The voltage-to-time conversion is done by comparing the
controller output with a triangular waveform at the switching frequency. The output of the comparator will
give a pulse-width modulated (PWM) signal wherein the controller voltage information is converted to
pulse-width modulation or duty cycle variation. The pulse signals are passed through the power switch drive
circuit (as described in Chapter 2) to control the power semiconductor switch state.
The output voltage sensing may be done in a simple manner by attenuating the output voltage with a resistive
divider. However, in that case the output and the primary-side grounds are common and the circuit looses the
galvanic isolation property. To retain the output and the primary-side ground isolation, the opto-isolation circuit
as depicted in Figure 10.42 is a popular-sensing circuit used in DC–DC converter circuits. The output voltage is
used to drive a current through the diode of the opto-isolator. The current through the opto-isolator diode is
Vo − V z
iopto = (10.61)
Ropto
The current iopto through the opto-coupler diode optically drives the opto-transistor Q. If the output voltage
Vo increases, then iopto increases leading to greater drive for the opto-transistor Q and the voltage across the
Vi
Triangle waveform Vo
Voref + d
+ PI − Drive
−
Vref
Ropto
Q1
Vo
Opto-isolation
Feedback R1
Vz
Figure 10.42 Control block diagram for DC–DC converters.
Chapter 10.indd 601 3/18/2009 1:13:08 PM

transistor decreases. This implies that the fed back voltage across R1 increases. Thus, Vo feedback across R1 is
directly proportional to Vo. The voltage reference Vz is used for nominal drive bias and also for temperature
compensation of the opto-coupler diode.
The comparator, controller, triangle waveform generation and comparison with the controller output and
pulse-width modulation generation are generally available as a single IC. Many such PWM ICs from various
manufacturers are available in the market for the designer to make an appropriate selection based on the
application. The PWM ICs are mixed signal ICs and the control is performed in the continuous or analog
domain. Thus the plant model and the controller design should be done in the s-domain. Alternately, the
controller and the modulator can be programmed within a microcontroller wherein the controller is in the
z-domain. In such a case the plant model should be converted into the z-domain equivalent by providing
the sampling time information and then the controller should be modeled in the z-domain. As microcon-
trollers are more popular and more ubiquitous in usage, the z-domain controller design is more relevant in
practical applications.
In most applications the DC–DC converters supply multiple loads with isolated grounds and different
voltage requirements. In such cases, the transformer consists of multiple secondary windings that supply
power to the various loads. From the control viewpoint, as there is only one control input (i.e., the duty
cycle of the power switch), only one output can be controlled. Then the question arises as to how the other
outputs should be controlled?
For all such applications involving multiple outputs, the voltage corresponding to the highest power
output is fed back in the manner described above for controlling it. The other outputs are controlled in one
of the following methods:
1. Linear regulator method
2. Non-isolated converter method
3. Coupled inductor method
4. Magnetic amplifier method
Figure 10.43 illustrates the use of a linear regulator to control the non-controlled outputs. The combina-
tion of the switched-mode and linear regulator is a very popular method to obtain good quality regulated
outputs. The switched-mode converter helps in reducing the input–output differential voltage drop across
the linear regulator by providing the necessary voltage scaling. The linear regulator provides the line and
load regulation that is required by the output.
Alternately, the linear regulator can be replaced by a non-isolated converter that can be one of the
three primary DC–DC converter topologies (Figure 10.44). The power switch of the secondary-side
non-isolated converter is controlled by a local feedback loop system that is powered by the output voltage
itself.
Another interesting method of controlling the uncontrolled outputs is by coupling the inductors of each
of the isolated outputs as indicated in Figure 10.45. Referring to Figure 10.45, let Vo1 be controlled
in the regular manner by feeding back the output voltage through the opto-isolator circuit and let Vo2 be
the uncontrolled output. For the sake of this discussion, consider the secondary diodes as ideal and that the
leakage inductances between N1 and N2 windings is zero and the coupling between L1 and L2 as tight.
The following constraint is applied to the above circuit:
Vs1 N 1 N L1
nc = = = (10.62)
Vs2 N 2 N L2
Chapter 10.indd 602 3/18/2009 1:13:08 PM

Vi
Linear
regulator
Vo1
Reference
+
−
Opto-isolator
Figure 10.43 Control of multiple outputs with linear regulators.
Vi
Vo1
Reference
+
−
Opto-isolator
Figure 10.44 Controlling multiple outputs using non-isolated converters.
Chapter 10.indd 603 3/18/2009 1:13:08 PM

Vi
VA
L1
N1
Vo1
Vs1
VB
L2
N2
Vo2
Vs2
Figure 10.45 Control of multiple outputs through coupled inductors.
where NL1 and NL2 are the number of turns of the inductors L1 and L2, respectively. Node voltage VA is
equal to Vs1 during period DTs and is equal to zero during (1 – D)Ts. Likewise, the node voltage VB is equal
to Vs2 during period DTs and is equal to zero during (1 – D)Ts. From Eq. (10.62), it can be deduced that
V A = nc ⋅VB
VL1 = nc ⋅VL2 (10.63)
The uncontrolled output voltage Vo2 is given as
Vo 2 = VB − VL2 = ncV A − ncVL1 = nc (V A − VL1 ) = ncVo1 (10.64)
Equation (10.64) implies that as nc is a constant, if Vo1 is controlled then Vo2 is automatically controlled to
nc times Vo1. Any variation in Vo2 reflects as variation in VL2 and hence variation in VL1 and Vo1. The varia-
tion in Vo1 is controlled by the control loop which in turn brings back Vo2 to the nominal value. It should,
however, be noted that the relationships given in Eq. (10.63) is for an idealized condition where the second-
ary diodes are ideal and that the coupling between the two secondary windings and the inductors is perfect.
In practice this is not possible. The diode drops vary with the load currents and are different in each winding
and for different load values. Further, the coupling between the inductors and the transformer secondary
windings is not perfect. Therefore the relationship Vo 2 = ncVo1 is not exactly correct. This will cause a cer-
tain amount of de-regulation in both the outputs. As a consequence each individual power supply will have
to be tuned for coupling to obtain the best regulation. This, therefore, is not a very good method for mass
productions.
Another interesting method is the magnetic amplifier method. Actually the name magnetic amplifier is
a misnomer as no amplification is involved. It uses a saturable inductor to control the output voltage. The
saturable inductor Ls is included in the uncontrolled output circuit as shown in Figure 10.46. A controlled
current ifR is injected at node B to reset the inductor Ls as indicated.
Chapter 10.indd 604 3/18/2009 1:13:08 PM

ifR
Vi DR
A C
1:n
Ls B D1 L
Vo1
D2
Vo2
Reference
+
−
Opto-isolator
Figure 10.46 Control of multiple outputs using magnetic amplifier.
Operation
The operation of the magnetic amplifier can be understood by referring to the waveforms shown in
Figure 10.47. Consider the steady-state operation wherein the forward-converter switch is OFF and the
output inductor L is freewheeling through the diode D2. The voltage across the secondary is such that the
dot end is negative making D1 reverse-biased and OFF. Now when the forward-converter primary-side
switch is turned ON, the dot ends of the transformer windings are positive with respect to the non-dot
ends. The inductor Ls offers high impedance and does not allow the voltage to pass immediately to the
node C. All the secondary winding voltage drops across Ls. The output inductor L continues to freewheel
through D2. A current flows through the inductor Ls through the path starting from A–Ls–B–D1–node
C–the non-dot end of the secondary winding. The current flows from node C to the non-dot end of the
secondary through the diode D2. As long as this current is less than the freewheeling current of the output
inductor L, the diode D2 is forward-biased and conducts. Due to the current flowing in the inductor, the
flux within the core of Ls increases at the rate of nVi / N Ls where NLs is the number of turns of the satura-
ble inductor winding. The saturable inductor saturates once the flux in the core reaches fsat as indicated in
Figure 10.47(b). Once the flux in the saturable-core saturates, the inductor cannot support any voltage
and offers zero impedance. The secondary voltage is now passed on the node C which will turn OFF D2
and the output inductor L magnetically starts charging up. The node voltage C waveform is shown in
Figure 10.47(c).
At the onset of the (1 – D)Ts period, the forward-converter primary-side switch is turned OFF. The non-
dot ends are positive with respect to the dot ends. The output inductor starts freewheeling through D2 and
D1 is OFF. The saturable inductor core is now reset by allowing a current ifR to flow through it in the path
Vo–B–A–non-dot end–negative of output. The reset current flowing through the inductor Ls sets up an mmf
Chapter 10.indd 605 3/18/2009 1:13:08 PM

DTs (1 − D)Ts DTs (1 − D)Ts DTs

VLS
(a) (VAB)
0
t
nVi /NLS nVi /NLS nVi /NLS

fsat
(b) fLS
0
t
−fLS1
−fLS2
−fsat
Vo
VC
(c)
0
t
Figure 10.47 Waveforms explaining the operation of magnetic amplifier.
(NLsifR) that resets the core. The value of the current ifR decides the extent to which the saturable inductor
core gets reset. If ifR is small then the core can get reset to a flux value of –fLs1 as indicated in Figure 10.47(b).
If ifR is high then the core can get reset to a flux value of –fLs2 as indicated in Figure 10.47(b). The depth to
which the Ls core gets reset determines the time taken for the core to saturate in the immediate next DTs
period. As the rate of rise of the flux in the core during the DTs period is fixed and equal to nVi / N Ls , the
time taken for the core to saturate is wholly dependent on the depth to which the core is reset in the previ-
ous cycle. As can be observed from the node C waveform of Figure 10.47(c), if the core reset is not deeply
in the negative side as shown for –fLs1, the inductor reaches the saturation during the DTs period in a
shorter time as compared to the –fLs2 reset value. This implies that if the core is reset to a greater depth in
the negative direction, then voltage pulse width available at node C to charge the output inductor is smaller
and vice-versa. Thus larger ifR core reset current will try to reduce the output voltage and a smaller ifR core reset
current will try to increase the output voltage.
Figure 10.48 gives the practical implementation of the magnetic amplifier. The core reset current ifR is
obtained by the transistor circuit. The output voltage is compared with a reference and the error is used to
drive the PNP transistor Q. The resistor Re along with Q controls the reset current ifR. If the output Vo
increases, the error at the output of the op-amp reduces. This will increase the drive for the PNP transistor
Q which drops less emitter–collector voltage across it. The drop across Re will increase, which will increase
the reset current ifR. An increased value of ifR will reset the core deeper in the negative direction thereby
trying to reduce the output voltage. A reduction in the output voltage will in a similar manner lead to
reduced ifR which will try to increase the output voltage.
Chapter 10.indd 606 3/18/2009 1:13:09 PM

Vi
Ls
Vo
iφR
Re
−
f
+
DR Vref
Figure 10.48 Magnetic amplifier circuit implementation.
10.11 Controlling Current
I n most applications, generally the output voltage is the controlled output. However, there are some appli-
cations where the input current or the switch current or the output current is controlled. Typical current
control applications are
1. current-controlled DC–DC converter or current programmed converter;
2. boost converter based unity power factor converter;
3. single-phase and three-phase front-end converter for reactive power compensation.
These topologies will be discussed to get a flavor for controlling the current instead of the voltage.
Current-Controlled Converter
In this method of control, the DC–DC converter is operated with constant frequency but current-
limited pulse width. The turn-ON times of the power switch are clocked periodically and the turn-OFF
times are determined by the time at which the current through the switch reaches a reference current
limit. This method can be applied to any of the single switch, dual switch or the derived forward con-
verters. This involves inner fast current control loop as shown in Figure 10.49. The advantages of current
control are
1. The power semiconductor switch is turned OFF when the switch current reaches a reference current
limit. This implies that there is inherent over-current protection for the entire converter.
2. Several similar converters may be connected in parallel to share the load. This is possible as the current
through the power switch of each converter is limited by the reference current limit, each converter will
deliver only the proportion of power that it is programmed for by its reference current limit.
3. The peak inductor current is determined by the reference current limit that limits the maximum
current through the power switch. Thus, the maximum energy-storing capability of the inductor
Chapter 10.indd 607 3/18/2009 1:13:09 PM

is determined and as overloads cannot occur in this type of control, the inductor saturation is
eliminated.
4. As the power switch currents are limited, the problem of flux walking that is discussed in the case of
push–pull, half-bridge and full-bridge topologies does not exist. If there is a mismatch in the ON-
state drops of the two sets of switches, then the transformer flux starts drifting towards a saturation
limit. As the flux nears the knee of the flux curve with respect to the magnetization or mmf, the cur-
rents in the power switches increase which will turn OFF the switch. This will automatically change
the duty cycle and will maintain the volt–second balance across the transformer even in the face of
mismatched set of power switches. This implies that one need not use the flux-walking capacitor in
this mode of control.
Figure 10.49 gives the control schematic of the current programmed control wherein an inner fast current
control loop is incorporated in addition to the outer voltage control loop. The DC–DC converter is a typical
forward converter. The switch current is sensed using a unidirectional current transformer (CT). The output
of the CT is amplified and compared with the output of the voltage controller Vc which sets the reference
current limit. At the start of every period, a clock generates a pulse that sets the S–R flip–flop which turns
the forward-converter power switch ON. The switch has a linearly increasing pattern due to the reflected
current from the output inductor. When the switch current reaches the reference current limit Vc, the
comparator will generate a negative pulse that will reset the flip–flop and turn OFF the forward-converter
power switch.
Slope Compensation
In current programmed control, there is an instability that is introduced due to the inner current control
loop for duty cycle operation greater than 0.5. Consider the current flowing in the output inductor at the
Vi
vo
Clock
S
Voref
+ PI + R Q Drive
− Vc −
Vo
Feedback
A CT
Figure 10.49 Block schematic of current programmed control.
Chapter 10.indd 608 3/18/2009 1:13:09 PM

steady-state condition. The current through the inductor has a linear rise and fall with the slopes dictated by
the voltage across the inductor during DTs and (1 – D)Ts periods. Figure 10.50(a) illustrates the instability
problem with the help of the inductor current in steady state.
Consider a forward converter with transformer turns ratio of 1:1 such that the switch current relates
directly to the inductor current for the sake of this discussion. However, this discussion is valid for any turns
ratio. In Figure 10.50(a), the reference current limit waveform is a constant. During the DTs period the
inductor current and the switch current are same wherein the current rises with a slope of m1. When the
switch current reaches Vc, the switch is turned OFF and the inductor current falls with a slope of –m2. Thus
in the steady state,
ΔiL ΔiL m2 D
m1 = ; m2 = − ; = (10.65)
DTs (1 − D )Ts m1 1 − D
Consider a perturbation δiL0 in the inductor current at the beginning of the period. The perturbed
inductor current follows the dotted lines shown in Figure 10.50(a). At the end of one period, the
resulting perturbation is δiL1 . This becomes the starting perturbation for the next cycle and so on.
At the end of the kth period, the resulting perturbation is δiLk . From Figure 10.50(a), it can be
deduced that
m2 D
δiL1 = δiL0 = δiL0 (10.66)
m1 1− D
At the end of the kth period, the perturbation δiLk is given as
k
⎛ D ⎞
δiLk = ⎜ ⎟ δiL 0 (10.67)
⎝ 1− D ⎠
DTs (1 − D) Ts DTs (1 − D) Ts
iL iL
Vc
Vc (Ref current limit) (Ref current limit)
−mc
−m2
ΔiL
m1 diL0 m1
diL0
diL1 −m2 diL1
(a) (b)
Figure 10.50 (a) Disturbance propagation with constant reference current limit;
(b) disturbance propagation with inverted ramp reference current limit.
Chapter 10.indd 609 3/18/2009 1:13:09 PM

The term [ D / (1− D )]k grows for D ≥ 0.5 causing the instability problem. Therefore, for a constant refer-
ence current limit, the maximum duty cycle should be limited to 0.5.
The instability problem can be solved by introducing an inverted ramp for the control voltage Vc that
determines the reference current limit as indicated in Figure 10.50(b). The current limit voltage Vc is an
inverted ramp voltage with a slope of –mc. From Figure 10.50(b), it can be deduced that
k
⎛ m − mc ⎞
δiLk = ⎜⎜ 2 ⎟⎟ δiL0 (10.68)
⎝ m1 + mc ⎠
From Eq. (10.68), it can be observed that the initial perturbation δiL0 can be made to decay by an appro-
priate choice of –mc even for duty cycles beyond D = 0.5. In particular, if mc is chosen equal to m2, the per-
turbation decays within one period. Thus, incorporating a stabilizing ramp in the control voltage enables the
inner current control loop to be stable. This incorporation of the stabilizing ramp in the control voltage is
called slope compensation. The modified control block diagram with the slope compensation incorporated is
Unity Power Factor Converter

Another application that is very common is the unity power factor converter. In Chapter 3, the rectifier–
capacitor filter circuit is discussed. The rectifier–capacitor filter circuit is the most popular method of obtain-
ing a DC voltage from the AC mains. However, the input current is of pulsed nature and as a consequence
Vi
−mc vo
Clock
S
Voref +
+ PI + + R Q Drive
− Vc −
Vo
feedback
A CT
Figure 10.51 Block schematic of current programmed control with slope compensation.
Chapter 10.indd 610 3/18/2009 1:13:09 PM

the power is very poor. To improve the power factor, a boost converter can be used as shown in Figure 10.52(a).
The input stage of a boost converter has an inductor that has a filtering effect on the line current. The power
semiconductor switch of the boost converter is controlled in a manner such that the input current is in-
phase with the rectified source voltage to achieve unity power factor operation. Figure 10.52(b) shows the
rectified source voltage that appears at the input to the boost converter. The requirement for the input cur-
rent of the boost converter is also shown. The input current of the boost converter should be in-phase with
the input voltage as shown in Figure 10.52(b), to achieve unity power factor.
The control block diagram of the boost converter for unity power factor operation is shown in
Figure 10.53(a). The objective of the controller is to control the input current of the boost converter. The
input current ii of the boost converter is fed back and compared with the reference iiref that has a rectified
sinusoidal waveshape such that it is in-phase with the boost input voltage Vi. The error is passed to a PI con-
troller whose output is compared with a triangular signal to generate the PWM signal. The generated PWM
is passed through the power switch drive circuit to turn ON and OFF the boost converter switch. The PI
controller ensures that the error at its input is zero, which implies that ii tracks iiref .
Reference Current Generation

The generation of a proper tracking reference iiref is important in this topology. iiref should have a waveshape
that follows the waveshape of the rectified source voltage. However, the amplitude of iiref is determined by
the output power requirement. Consider the input source voltage
v s = Vm sin ω t
ii
L
Vs = Vm sin wt Vi Vo
(a)
Vi, ii Vi Requirement
ii
0 t
(b)
Figure 10.52 (a) Boost converter used as power factor improvement circuit; (b) required input
current for unity power factor.
Chapter 10.indd 611 3/18/2009 1:13:09 PM

ii
L
is
Vs = Vm sin wt Vi Vo
ii Feedback
iiref −
+ PI − Drive
+
(a)
Po = VoIo
Step Virms Iirms

Vs = Vmsin wt Rect rms
down
sin wt
1
√2 u1u2 iiref
2
(b)
Figure 10.53 (a) Block diagram of the control scheme for power factor
improvement; (b) reference current generation.
This vs is normalized and rectified to obtain

vs
vn = = sin ωt (10.69)
Vm
Equation (10.69), which is the normalized and rectified source voltage waveform, defines the reference cur-
rent waveshape. The amplitude is determined by the output power. If Po is the output power, then
Po = Vo I o (10.70)
Assuming 100% efficiency of the boost converter,
Virms I irms = Po = Vo I o (10.71)
Chapter 10.indd 612 3/18/2009 1:13:09 PM

From Eq. (10.71),

Po V I
I irms = = o o (10.72)
Virms Virms
Equation (10.72) determines the amplitude of the reference current waveshape. From Eqs. (10.70) and
(10.72), the reference current is given as
Vo I o
iiref = 2 I irms sin ωt = 2 sin ωt (10.73)
Virms
The block diagram for the reference current generation is shown in Figure 10.53(b).
Front-End Converter
The unity power factor converter discussed in the previous section is a unidirectional power flow converter.
To have a bi-directional power flow wherein the source also sinks power, the front-end converters are used.
The topology of a single-phase front-end converter is shown in Figure 10.54(a). The AC source and the two-
arm bridge are interfaced by an inductor L as shown. The two-arm bridge is operated as an inverter which
generates a voltage Vinv such that the current through the inductor or source is in-phase with the source volt-
age Vs. The internal body diodes act as rectifiers to allow the power flow from the source to the capacitive
DC bus, Vdc.
The control objective here is to maintain the source current is to be in-phase with the source voltage Vs,
independent of the nature of the load. The control block diagram is shown in Figure 10.54(b). The voltage
to be controlled is the inverter output, Vinv. The duty cycle to the four switches are generated that will con-
trol the amplitude and the phase of the sinusoidal PWM waveform of the inverter output, such that the
source current is in-phase with the source voltage. The voltage Vinv or VAB is fed back and compared with a
reference Vinv* as shown in Figure 10.54(b). The error from the comparator output is passed as input to the
controller that tries to make the error zero. The controller output is compared with a triangular waveform to
generate the PWM signal. The PWM signal is passed to the drive circuit in order to turn ON and OFF the
power switches of the inverter.
Reference Voltage Generation

The inverter output voltage controls the phase and amplitude of the current waveform. Therefore, a proper
tracking voltage reference has to be generated that will make the current in-phase with the source voltage
independent of the nature of the load. The source current is given as
1
is = ∫ (v s − v inv ) dt (10.74)
L
Let v s = Vm sin ω t be the source voltage. If the source has to only deliver unity power factor current, then
the current waveshape should be in-phase with the source voltage. If the peak current is Im, then
is = I m sin ω t (10.75)
Substituting for vs and is in Eq. (10.74), the required inverter output voltage to achieve unity power factor is
* = ⎛ v − L s ⎞ = V sin ω t − ω L ⋅ I cos ω t
di
v inv ⎜⎝ s ⎟ (10.76)
dt ⎠
m m
If the inverter output vinv generates a voltage as given in Eq. (10.76), the source current is in-phase with the
source voltage. The current amplitude is determined from the output power drawn from the source. If the
DC power drawn from the DC bus is Po, then for a 100% efficient inverter.
Chapter 10.indd 613 3/18/2009 1:13:10 PM

S1 S3
Vs is L Vinv
Vdc RL
S2 S4
(a)
Idc
g1 g3
S1 S3
A
Vs is L Vinv
Vdc RL
B
g2 g4
S2 S4
Vinv feedback (VAB)

g1 g2 g3 g4
∗
Vinv −
+ PWM
PI Drive circuits
generation
(b)
Figure 10.54 (a) Single-phase front-end converter topology for power factor improvement; (b) control
schematic for the control of the single-phase converter.
Chapter 10.indd 614 3/18/2009 1:13:10 PM

Vm I m
Po = Vdc I dc = Vsrms I srms = (10.77)
2
From Eq. (10.77),
2Vdc I dc
Im = (10.78)
Vm
Substituting Eq. (10.78) in Eq. (10.76), one obtains
2V I
v inv * = Vm sin ωt − ωL dc dc cos ωt (10.79)
Vm
All the variables and parameters in Eq. (10.79) are known or measurable and hence the reference voltage can
be generated to achieve unity power factor operation for the single-phase front-end converter.
Three-Phase Extension
The concepts of the single-phase front-end converter can be extended to the three-phase front-end con-
verter. The three-phase front-end converter topology is shown in Figure 10.55. Here the control objective is
to achieve unity power factor operation for the three-phase source. The three-phase currents should be in-
phase with the three-phase voltages of the source. Space-vector control method is well-suited for applying to
three-phase systems. From the discussion in Section 8.9, the space vector for the three-phase source voltage
is given as
v s = van + v bn ⋅ e i( 2π /3) + v cn ⋅ e i( 4π /3) (10.80)
Similarly, the space vector for the source phase currents is
i s = ia + ib ⋅ e i( 2π /3) + ic ⋅ e i( 4π /3) (10.81)
The space vector generated at the output of the inverter at nodes a, b and c is v inv . The three-phase equiva-
lent circuit is shown in Figure 10.56.
Vdc
Idc
S1 S3 S5
VaN a
La
VbN b C RL
Lb
VcN c
Lc
N
S2 S4 S6
Figure 10.55 Three-phase front-end converter.
Chapter 10.indd 615 3/18/2009 1:13:10 PM

is
Labc
Vs Vinv
Figure 10.56 Equivalent circuit of the three-phase front-end converter.
The required reference voltage space vector is given as

⎛ di ⎞
v inv * = ⎜ v s − L s ⎟ (10.82)
⎝ dt ⎠
The space vectors for the source voltage and the source current are shown in Figure 10.57 for a general lagging
power factor where the current space vector lags the voltage space vector. If the voltage space vector axis is
considered as the reference axis, then the reference axis also rotates synchronously with the voltage space
vector. With respect to this axis, the current space vector can be resolved into two orthogonal components, isd
and isq. The isd component is along the rotating reference axis and the isq component is perpendicular to the
reference axis. For the power factor to be unity, the spatial angle between the voltage and current space vector
should be zero. In other words, the current space vector should be along the reference axis. This implies that
isq should be zero. Thus, isq reference should be set to zero and isd reference should be set according to the
power drawn from the supply, and an inverter voltage space vector v inv should be so generated that achieves
this objective. All the voltages and currents can be expressed with respect to the reference axis that is aligned
along the source voltage space vector and rotating synchronously along with it. This implies that all the cur-
rent and voltage values with respect to the rotating reference axis will be DC quantities (Figure 10.58).
Reference axis
b
Vs
is
q
isd is
f
g
a
Figure 10.57 Source voltage and current space vectors.
Chapter 10.indd 616 3/18/2009 1:13:10 PM

Vinv Reference
w axis
Vs d
q w isd
Vinvq Vinvd is
g
0 Stationary
axis
isq
Figure 10.58 Representation of the source voltage, source current and inverter space vectors with
respect to the rotating reference axis.
With respect to the rotating reference axis, the inverter voltage space vector v inv can be decomposed into
the direct and quadrature components as v invd and v invq , respectively. The value v invq should be generated
such that isq is made zero and v invd should be generated such that isd supplies the power to the load.
All the voltage and current quantities are measured in the stationary axis. Transferring them to the refer-
ence axis that is at an angle γ is a two-step process. First the three-phase quantities are converted to two-
phase quantities in the same stationary axis. This is called the three-phase to two-phase conversion as
discussed in Section 8.9 and given in Eq. (8.89). The second step is to convert the two-phase quantities in
the stationary axis to the rotating reference axis by using Eq. (8.92). The variables in the two-phase station-
ary axis are represented with a and b subscripts. The variables in the rotating reference frame are represented
with d and q subscripts. If χ is either a voltage or a current variable, then for transferring variables from
three-phase stationary axis to the rotating reference axis the following two-step process is used:
1. Three-phase to two-phase transformation in stationary axis:
⎡ 1 1 ⎤
⎡ χa ⎤
1 − −
⎡ χα ⎤ ⎢ 2 2 ⎥⎢ ⎥
⎢χ ⎥ = ⎢ ⎥ χb (10.83a)
⎣ β ⎦ ⎢0 3 3⎥⎢ ⎥
− ⎢χ ⎥
⎢⎣ 2 2 ⎥⎦ ⎣ c ⎦
2. Two-phase stationary axis to two-phase rotating reference axis transformation:
⎡ χd ⎤ ⎡ cos γ sin γ ⎤ ⎡ χα ⎤
⎢χ ⎥ = ⎢ ⎢ ⎥ (10.83b)
⎣ q ⎦ ⎣− sin γ cos γ ⎥⎦ ⎣ χβ ⎦
For transferring variables from the two-phase rotating reference axis to the three-phase stationary axis the
following two-step process is used:
1. Two-phase rotating reference axis to two-phase stationary axis transformation:
⎡ χα ⎤ ⎡cos γ − sin γ ⎤ ⎡ χd ⎤
⎢χ ⎥ = ⎢ ⎢ ⎥ (10.84a)
⎣ β ⎦ ⎣ sin γ cos γ ⎥⎦ ⎣ χq ⎦
Chapter 10.indd 617 3/18/2009 1:13:10 PM

2. Two-phase to three-phase transformation in the stationary axis:
⎡ χa ⎤ ⎡ 2 / 3 0 ⎤
⎢ χ ⎥ = ⎢−1 / 3 1 / 3 ⎥ ⎡ χα ⎤ (10.84b)
⎢ b⎥ ⎢ ⎥ ⎢χ ⎥
⎢⎣ χc ⎥⎦ ⎢−1 / 3 −1 / 3 ⎥ ⎣ β ⎦
⎣ ⎦
Idc
a
RL Vdc b
c
Drive circuits
Vinv1
∗
isd + PI Vinvα
−
dq ab
isd to to PWM
ab abc GEN Vbn
Vinv2 Vinvβ
∗ =0
isq + PI
−
Vcn
g g Van
isq
Vα Van
Vsd 3f
ab
to Vbn
Vsq to
dq 2f Vcn
Vβ
iα ia
isd 3f
ab
to ib
isq to
dq 2f ic
iβ
g
Obtained from
g ∫ w phase voltage
waveforms
Figure 10.59 Control block diagram of the three-phase front-end converter.
Chapter 10.indd 618 3/18/2009 1:13:10 PM

The angle γ for the axis transformation is obtained from the source voltage waveforms as
vα
γ = cos −1 (10.84c)
vα + v β 2
2
At the steady state, γ = ω t where ω is the source voltage frequency.

The reference value for the isq* is zero. The reference value for isd is obtained from the power delivered to
the load. Assuming 100% efficiency of the inverter, the power delivered to the load is
⎛V I ⎞
Po = Vdc I dc = 3 ⎜ m m ⎟ (10.85)
⎝ 2 ⎠
where Vm is the peak phase voltage and Im is peak phase current value of the source. In the rotating reference
frame, the amplitude of the space vector is 1.5 times the peak amplitude of the phase quantity in the station-
ary axis due to the transformation. Therefore, in the rotating reference frame,
⎛V I ⎞
Po = 3 ⎜ m m ⎟ = Vsd I sd + Vsq I sq (10.86)
⎝ 2 ⎠
Note that Vsq = 0 along the rotating reference axis, Vsd = Vs = 3Vm/2 and Isd = 3Im/2. Therefore, to ensure
that the power calculated in the stationary axis is the same as the power calculated in the rotating reference
axis, a factor of 2/3 is included to make the power balance. That is
⎛V I ⎞ 2
Po = 3 ⎜ m m ⎟ = Vsd I sd + Vsq I sq = Vs I sd (10.87)
⎝ 2 ⎠ 3
From Eqs. (10.85) and (10.87), the reference value for isd is
Vdc I dc
isd * = (10.88)
( 2 / 3)Vs
The control block diagram for the three-phase front-end converter control for unity power factor is shown
in Figure 10.59.
10.12 Control of Induction Motor
T he speed control of induction motor has evolved over the years from the simple stator voltage control
wherein the stator voltage is varied to achieve a small speed range. However, the stator voltage control
is very inefficient as the flux decreases with decrease in voltage. The generated torque is the product of
the flux and the stator current. If the flux decreases with reduction in stator voltage, the generated torque
reduces. To produce the same generated torque, the stator current should increase which will also increase the
losses. Next in the evolution of the induction motor control is the flux control. Depending on the type of
flux control, whether based on the steady-state flux control or dynamic flux control, the scalar (V/f ) control
and the vector control strategies have evolved.
Scalar Control
The scalar control of induction motor uses the steady-state model of the induction motor. The steady-state
model of the induction motor is given in Figure 10.60. The equivalent circuit is similar to that of a trans-
former. The stator side comprises the stator resistance Rs and the stator leakage Lss. The rotor side comprises
the rotor leakage Lsr and the equivalent load is given as Rr/s where s is the slip of the induction motor. The
inductance M represents the mutual coupling between the stator and the rotor.
Chapter 10.indd 619 3/18/2009 1:13:10 PM

is ir
Rs Lσs Im Lσr
Eb Rr
Vs M
S
Figure 10.60 Steady-state model of the induction motor.
The voltage across the mutual inductance M is the back emf which is represented as Eb. If Im is the magnetiz-
ing current responsible for the air-gap flux, then E b = ωMI m. The flux linkage in the air gap is given as
Eb E
φ = MI m = = b (10.89)
ω 2π f
From Eq. (10.89), it can be observed that the flux in the core can be maintained constant if the ratio of the
back emf (Eb) to the frequency (f ) is kept constant. If the flux in the core is maintained as constant, then as the
stator voltage is varied the frequency is also varied proportionally. This will give a wide speed range of operation
without compromising the torque-generation capability of the induction motor. Figure 10.61(a) gives the rela-
tionship between the back emf and the frequency. Upto the base frequency fb, the back emf and the frequency
are linearly related with a constant slope. Beyond base frequency, the stator supply voltage reaches the rated
voltage of the induction motor and cannot be increased any further. As only the frequency increases, the flux
reduces according to Eq. (10.89). This region beyond base frequency is called the field weakening region.
However, it is not easy to measure the back emf for the induction motor as the equivalent mutual induc-
tance terminal is virtual and not accessible. Therefore, the terminal voltage Vs is considered instead of the
back emf. If Vs/f is maintained constant, then the flux within the air gap is reasonably constant. The stator
terminal voltage versus the frequency plot is shown in Figure 10.61(b). The terminal voltage Vs differs from
the back emf Eb by an amount equivalent to the drop across the stator resistance Rs and the drop across the
leakage reactance wLss. The leakage reactance is generally very small compared to the mutual reactance and
Eb Vs Constant
f region
Field
weakening
Eb Vs region
=f ∼f
f f
Vb
0 fb f 0 fb f
(a) (b)
Figure 10.61 (a) Back emf versus frequency; (b) stator terminal voltage versus frequency.
Chapter 10.indd 620 3/18/2009 1:13:11 PM

hence may be neglected. The stator resistance drop is a small fraction of the terminal voltage at high values
of Vs, that is, at higher frequency and speed. However, at low frequency, the stator terminal voltage is
comparable to the stator resistance drop. As a consequence, the terminal voltage is much different from the
actual back emf Eb at low frequencies. To compensate for the stator resistance drop, a small boost to the
stator terminal voltage is provided at low frequencies as indicated in Figure 10.61(b). Beyond the base fre-
quency fb, the terminal voltage is held constant and the field weakens on increasing the frequency.
Open-Loop V/f Control

The control structure of the open-loop V/f control of the induction motor is shown in Figure 10.62. The
power circuit consists of the mains voltage that is rectified and filtered with a capacitor as shown. The capacitor
voltage acts as a DC bus for the inverter. The three-phase inverter supplies the induction motor. The switches
of the inverter are controlled by a pulse-width modulator followed by relevant switch drive circuits. The
control command is obtained by setting the synchronous frequency ws for the induction motor. The required
stator terminal voltage Vs is obtained from a look-up table that includes the characteristic as given in
Figure 10.61(b).
Integration of ws gives the angle g of the voltage space vector Vs at any given instant of time. The two
phase quantities in the stationary axis are obtained as
Vsα = Vs cos γ
(10.90)
Vsβ = Vs sin γ
The two-phase to three-phase conversion is performed as discussed in the previous section and given in
Eq. (10.84b). The three-phase stator voltage signal is passed through a PWM generator to drive the
switches of the inverter.
Closed-Loop V/f Control

The block diagram for the closed-loop V/f control is shown in Figure 10.63. The shaft speed of the induc-
tion motor is sensed by a tacho and fed back to the controller. The fed back shaft speed wm is compared with
reference shaft speed wm*. The error is passed through a PI controller and added with the shaft speed to give
the synchronous speed ws. The rest of the block diagram is same as that of the open-loop V/f control. If the
actual induction motor speed increases due to an external disturbance, the speed error goes negative and
decreases the slip speed wsl. This will reduce the synchronous speed ws which in turn will reduce Vs. This will
have a reducing effect on the actual induction motor shaft speed trying to bring back the induction motor
Mains Rect Inverter IM
ws Vs Vsα Va
2f
Vb
to PWM
Vc
∫ 3f
g Vsβ
Figure 10.62 Open-loop V/f control.
Chapter 10.indd 621 3/18/2009 1:13:11 PM

Mains Rect Inverter IM
Tacho
Vs Vsα Va
Vs cos γ
2f Vb
to PWM
w∗ wsl ws 3f Vc
m
+ PI + ∫ Vs sin γ
− + g Vsβ
wm wm
Figure 10.63 Closed-loop V/f control.
speed to the set reference speed. A similar control action will take place even when the induction motor
shaft speed reduces due to an external disturbance.
Vector Control
In high-performance applications, it is necessary that the torque of the induction motor exhibits a fast
response. The dynamics of the electromagnetic torque is shown in Figure10.64(a). Figure 10.64(a) is derived
from the generated torque model of Eq. (8.150) that is discussed in Chapter 8. The torque model is based
on the rotor flux linkage components and the stator current components in the arbitrary reference frame.
The dynamics of the rotor flux linkage components yrd and yrq shown in Figure 10.64 are evident from the
rotor flux state model of the induction motor as given in the state space model of Eqs. (8.128)–(8.130).
If the stator current space vector i s or its components isd or isq are changed, then both the flux in the motor
and the torque developed by the motor are affected. Thus, in conventional control strategies, there is no way
to independently control the flux in the machine and the torque produced by the machine. The torque
isd isd
Rrnr isq Rrnr isq
+ +
tr x tr x
+ +
yrd yrd = yr
wsl wsl
+ +
2pnr /3 2pnr /3
− Td − Td
wsl yrq wsl yrq = 0
− −
tr x tr x
+ +
isq isd isq isd
Rrnr Rrnr
Figure 10.64 (a) Induction motor torque dynamics; (b) torque dynamics under field orientation.
Chapter 10.indd 622 3/18/2009 1:13:11 PM

dynamics will be limited by the dynamics of the flux, which has a large time constant (∼300 ms). To
improve the torque dynamics, the load torque-producing and flux-producing components of the stator cur-
rent space vector (i.e., isq and isd) must be decoupled. Thus on decoupling, the torque will have dynamics as
that of the stator currents (∼10 ms) which is generally 20 to 50 times faster than the flux dynamics. The
decoupling of the torque-producing and the flux-producing components of the stator current to achieve
high dynamic performance in torque is performed by a method called the vector control or field-oriented
control.
With reference to Figure 10.64, it is evident that any change in the stator current components isd and/or
isq will affect the rotor flux linkage components yrd and yrq, with a time constant tr. From Figure 10.64,
it is clear that if yrq is somehow made equal to zero, Td will depend only on isq and yrd [see shaded portion
of Figure 10.64(b)]; and yrd will depend only on isd (without depending on isq). The drive torque Td now
responds with a time constant τr for any deviations in the direct component isd of the stator current space
phasor and will have an almost instantaneous response for any deviations in the quadrature component isq.
Figure10.65(a) shows the space-vector representation of the stator currents and the rotor flux linkage
components with respect to an arbitrary reference frame that is not aligned with the rotor flux linkage
space vector. The quadrature component of the rotor flux linkage space vector yrq is made zero by aligning
the direct axis of the arbitrary reference frame along the rotor flux linkage space vector. When yrq is made
zero, then from Eq. (8.150) the drive torque Td is given by
2
Td = p ⋅ υ r ⋅ψ rd ⋅ isq (10.91)
3
It is evident that if isd is kept constant, then yrd is constant and as a consequence the torque will now have
the dynamics of the stator current isq only. Clearly the isd component of the stator current can be used to set
the magnitude of the rotor flux linkage space vector yr (with yrd = yr; yrq = 0) and the isq component of the
stator current can be used to control the magnitude of the drive torque. Thus, by aligning the arbitrary refer-
ence axis along the rotor flux linkage space vector, one obtains decoupling between the torque-producing
and the flux-producing components of the stator current space vector and as a consequence the drive torque
is
ax
d-
ry
i tra
is isq rb is
− A
is) −
isd ax isq
flu
x xis is
or d-a r-ax
yrq (rot y
rar g
y
is bit on
rd
ax Ar d al
y
y r- isd e
gn
yr ali
r r
Stator axis Stator axis
(a) (b)
Figure 10.65 (a) Space vectors represented with respect to the arbitrary reference frame, where
the direct axis is not aligned along the rotor flux linkage space vector yr; (b) the direct
axis of the arbitrary reference frame aligned along rotor flux linkage space vector yr.
Chapter 10.indd 623 3/18/2009 1:13:11 PM

dynamic performance is improved. Therefore, this technique of control of the induction motor is referred to
as the rotor field-oriented control or simply field-oriented control. In the above scheme, the stator current
space vector is controlled by controlling the individual component vectors isd and isq. Therefore, this method
is also referred to as the vector control.
It is interesting to note from Eq. (10.91) that the dynamic performance of a field-oriented controlled
induction motor resembles closely to that of the separately excited DC motor which has a similar electro-
magnetic torque expression. Therefore, it is evident that the field-oriented controlled induction motor or a
vector-controlled induction motor can achieve dynamic performances that were in the past achievable only
by separately excited DC motors.
It may be noted that one may orient the arbitrary reference axis either along the rotor flux axis
(rotor field-oriented control) or the stator flux axis (stator field-oriented control) or the air-gap flux
axis (air-gap field-oriented control). All three strategies of field orientation lead to high dynamic per-
formance with respect to torque, but there exists considerable difference in implementing the various
field-orientation strategies.
Control of Induction Motor by Field Orientation

The speed control of an induction motor would consist of (a) an inner fast-acting torque control loop to
improve the dynamic performance as well as to limit the large transient currents through the power con-
verter and (b) an outer slower acting speed control loop. In the case of the DC motor, the armature current
is directly proportional to the electromagnetic torque. For a separately excited DC motor, torque control
loop reduces to armature current control with the field excitation being constant. However, in the case of
the induction motor, there is no easy way to reduce the torque control to a current control loop unless a
component of the stator current that is proportional to the torque is available.
It would be interesting to compare the control structures of the DC motor drive system and that of the
field-oriented controlled induction motor drive system. This would enhance the understanding of field-
oriented control for induction motors. Figure 10.66 gives the speed-controlled DC motor drive system.
Compare this with the speed-controlled induction motor drive system using the vector control strategy as
given in Figure 10.67.
In the DC motor drive system as shown in Figure 10.66, the speed reference wmref is compared with the
measured speed wm from the tacho. The speed error is passed through a suitable speed controller. The output
of the speed controller becomes the torque or the armature current reference iaref . This armature current ref-
erence is compared with the measured armature current ia. The armature current error is passed through a
current controller. Likewise the field current reference ifref is compared with the measured field current if ,
and the field current error is passed through a field controller. The field current error is used to modulate the
field converter to set the flux level in the DC motor. Similarly, the armature current error is used to modu-
late the armature converter to control the torque in the DC motor.
The vector-controlled induction motor drive is shown in Figure 10.67. It can be observed that there are a
lot of similarities between the DC motor and the vector-controlled induction motor drive systems. The speed
controller is similar to the DC motor drive system. The torque control is performed by the quadrature com-
ponent of the stator current space vector isq, whereas in the DC motor it is performed by the armature current
ia. The field control is performed by the direct component of the stator current space vector isd, whereas in the
DC motor it is performed by the field current if in the separately excited winding. One should note that in
the DC motor drive system, the field control portion and the armature control portion are inherently decou-
pled. On the other hand, in the case of the induction motor there is no such inherent decoupling. The decou-
pling in the induction motor is achieved by aligning the arbitrary reference frame along the rotor flux linkage
Chapter 10.indd 624 3/18/2009 1:13:11 PM

3f AC
AC to DC
Field
controller
if
ifref − Modulator Field Vf
for field Field
+ converter if circuit
control
wmref iaref Modulator Armature Va DC

for torque power ia motor
+ +
− − control converter
Speed ia Torque/current
wm controller controller
wm
Tacho
Figure 10.66 Speed control of separately excited DC motor.
space vector as explained earlier. One should also note that in the control portion of the drive for the induc-
tion motor, the variables are in the rotor flux reference frame which is rotating synchronously with the rotor
flux linkage space vector. Therefore, all the variables are DC quantities. Thus, a vector-controlled induction
motor drive system is similar to a DC motor drive system. Further, the vector-controlled induction motor
drive system will give dynamic performances similar to that of DC motor drive system.
For vector control operation of the induction motor, the arbitrary reference frame must be aligned along
the rotor flux linkage space vector (or any other flux linkage space vector of interest) at every instant. It is
therefore essential that the position of the rotor flux linkage space vector r be accurately known at every
instant. This knowledge of the rotor flux linkage space vector position can be acquired either by measuring
the flux directly or by estimating the flux from terminal variables (i.e., by indirect means). This leads to two
possible control techniques of induction motor:
1. direct field-oriented control;
2. indirect field-oriented control.
Direct Field-Oriented Control The block schematic of a direct field-oriented control (DFOC) of the
induction motor is shown in Figure 10.68. The flux measurement can be made using either the Hall sensors
Chapter 10.indd 625 3/18/2009 1:13:11 PM

3f AC
AC
to
DC
Variables
Field
are DC isd
controller
−
isdref
+
a 2f
e jr to PWM VSI IM
isqref
3f
w mref b
+ +
− −
Speed r
wm isq Torque/current Variables are AC
controller controller
isa
isd
2f
isq e−jr to
i
3f sb
isc
r
wm
Tacho
Figure 10.67 Field-oriented vector control of induction motor.
or the stator search (sense) coils. If the stator sense coils are used, then the voltage sensed from the coils will
have to be integrated to obtain the air-gap flux linkages. The measured air-gap flux linkage components are
used to calculate the required (rotor, stator or air gap) flux linkage space vector magnitude and position r.
The value of r thus computed is used to align the arbitrary axis along the flux vector to achieve decoupled
control of the torque- and flux-producing components of the stator current space vector.
The flux-sensing devices are placed in the air gap of the machine, which will determine the air-gap flux
space vector. Any other flux space vector can be calculated as it has an algebraic relationship with the air-gap
flux space vector. The air-gap flux sensed by either Hall-effect devices or stator search coils suffers from the
disadvantage that a specially constructed induction motor is required. Further, Hall sensors are very sensitive
to temperature and mechanical vibrations and the flux signal is distorted by large slot harmonics that cannot
be filtered effectively because their frequency varies with motor speed.
In the case of stator search (sense) coils, they are placed in the wedges close to the stator slots to sense the
rate of change of air-gap flux. The induced voltage in the search coil is proportional to the rate of change of
flux. This induced voltage has to be integrated to obtain the air-gap flux. At speeds below 1 Hz, the induced
voltage will be significantly low which would give rise to inaccurate flux sensing due to presence of comparable
amplitudes of noise and disturbances in a practical system.
Chapter 10.indd 626 3/18/2009 1:13:11 PM

3f AC
AC to DC
Field
isd controller
−
isdref
Voltage source
+
Inverter
2f to 3f
a
PWM IM
e jr
wmref isqref
b
+ +
− −
wm Speed Torque/current r
isq
controller controller
isa
isd
α
3f to 2f
e−jr isb
isq
b isc
r
From stator
sense coils
Flux vector ∫
positon
r 3
calculation
From Hall sensers
3f air gap
flux components wm
Tacho
Figure 10.68 Speed control of direct field-oriented controlled induction motor drive system.
Indirect Field-Oriented Control A flux estimator is used to estimate the required flux linkage space
vector magnitude and angular position r. The shaft position is usually needed for estimating the flux linkage
space vector position. If the shaft transducer is a position encoder, then the position information ε can be
directly used. But if the shaft transducer is a speed transducer like a tacho, then the speed has to be inte-
grated to obtain the shaft position. In the case of shaft transducer being a position encoder, the speed feed-
back is obtained by differentiating the shaft position information.
Chapter 10.indd 627 3/18/2009 1:13:11 PM

Indirect sensing of flux space vectors give a more versatile drive system that can be used with standard
commercial motors, but this approach would generally result in a more complex control system. Since it is
generally desirable to have a scheme which is applicable for all induction motors, the indirect field-oriented
control has emerged as the more popular method. In the indirect method of field orientation, the flux link-
age space vector is estimated from the motor model. As a consequence, all indirect methods are sensitive to
variations in some machine parameter such as the stator or rotor time constants. For example, in the rotor
flux-oriented control, the indirect rotor flux estimator is sensitive to the rotor time constant tr of the motor.
In the case of the stator flux-oriented control, the indirect stator flux estimator is sensitive to the stator time
constant of the motor. In the air-gap flux-oriented control, the indirect air-gap flux estimator is sensitive to
both the stator and the rotor time constants. Therefore, if the value of the motor parameter varies, the
desired decoupling of the flux and the torque components of the stator current space vector is not achieved
and this leads to deterioration in the dynamic behavior of the drive system. As a consequence, in indirect
field-oriented control, parameter adaptation is essential to overcome the undesirable effects of the parameter
variations due to changes in temperature. However, indirect field orientation can be implemented down to
zero speed and is therefore suitable for servo drives also.
Flux Estimators for Indirect Field-Oriented Control

The flux estimator is model from the state space representation of the induction motor as discussed in the section
on estimators. However, by the knowledge of the plant, specific reduced order estimators can be synthesized. For
a rotor field-oriented control, the rotor flux model is deduced from the motor model of Eq. (8.128a).
dimr
τr + imr = isd
dt
dρ isq
= ωs = + ωm (10.92)
dt τ r imr
where imr = ψ r / M is defined as the equivalent rotor magnetizing current.
In the practical implementation of the field-oriented control, the induction motor is fed from a voltage
source inverter with fast current control loops. This approach is used in high-performance induction motor
servo drives for machine tool applications and industrial robots.
In field-oriented control system, the induction motor behaves like a DC machine under both steady-
state and transient conditions. Consequently, similar drive control strategies can be employed. Below base
speed, the magnetizing current of the induction motor representing the rotor flux magnitude is maintained
constant at its maximum possible value to achieve constant torque operation. Above base speed, the flux is
reduced, thereby giving the field weakening region or the constant horsepower region of operation.
A typical rotor field-oriented drive system is shown in Figure 10.69. The system consists of an induction
motor which is driven by a voltage source inverter. The DC-link to the inverter is obtained from the output
of an AC–DC converter which is fed from the three-phase mains. The inverter switching is controlled by the
field-oriented controller.
AC–DC converter is the first stage of the power-converter system of the induction motor drive. It con-
sists of a three-phase bridge rectifier followed by a capacitor filter. The output of the capacitor filter acts as
the DC-link voltage for the following inverter stage of the power-converter system. This type of AC–DC
converter would allow power flow only in one direction, that is, from the mains to the load. In the case of
the induction motor drive system, when the induction motor is in the generating mode, the power will flow
from the induction motor to the inverter DC-link source. Thus, suitable bi-directional front-end converters
which put back the power from the DC-link to the mains have to be used.
Chapter 10.indd 628 3/18/2009 1:13:12 PM

3f AC
AC to DC
isd
−
isdref
Voltage source
(flux +
Decoupling
Modulator
2f to 3f
network
inverter
reference)
d-controller
e jr
isqref
wmref
+ +
− −
Speed q-controller
wm controller isq
r isa
isd isα
3f to 2f
isb
e −jr
imr Flux isc
estimator isq isβ
wm
IM
Tacho
Figure 10.69 Block schematic of indirect field-oriented control of the induction motor.
The inverter normally consists of six power switches (thyristors, BJTs, MOSFETs or IGBTs) which
are connected in the classical bridge configuration. The control signals for the inverter switches are
obtained from the modulator block. The six power switch inverter configuration leads to a two-level
inverter. Three-level inverter configurations are used for higher powers wherein 12 switches have to be
used, four in each inverter leg. Its outputs are three-phase PWM voltages which are fed to the induction
motor.
Chapter 10.indd 629 3/18/2009 1:13:12 PM

Three-Phase to Two-Phase Transformation (a, b, c to a , b) The three-phase stator currents isa, isb and
isc that are measured, are first transformed to an equivalent two-phase system (isa and is b ) because the induc-
tion motor is represented as an equivalent two-phase machine. The three-phase to two-phase transformation
(3f – 2f) is carried out in the stator reference frame. The transformation equation is same as Eq. (10.83a).
This transformation is a general transformation that can be applied to any variable of the induction motor
like the stator voltages, stator currents, flux linkages, etc.
Stator to Synchronous Reference Frame Transformation (a , b to d, q) The two-phase stator currents
that are in the stator reference frame are transformed to a synchronous reference frame. The choice of the syn-
chronous reference frame is dependent on the flux along which the orientation is to be performed. If the arbitrary
axis is to be oriented along the rotor flux linkage space vector, then the synchronous reference frame would be
the rotor flux reference frame and if the arbitrary axis is to be oriented along the stator flux linkage space vector,
then the synchronous reference frame would be the stator flux reference frame. If the angle r represents the
instantaneous position of the synchronous reference frame along which the arbitrary axis is aligned, then the
transformation from the stator to the synchronous reference frame is given by Eq. (10.83b). The inputs to this
block are isa, isb and the rotor flux position r. The outputs of this block are isd and isq.
Flux Estimator As already discussed, the magnitude and the position of the flux linkage space vector has
to be estimated for indirect field-oriented control. For a rotor flux estimator model, this block has isd, isq and
wm as inputs and r and imr as outputs. The position information r is used to perform the e–jr and e jr trans-
formations [Eqs. (10.83b) and (10.84b)] as shown in Figure 10.69.
Speed and Current Controllers The speed reference wmref is compared with the fed-back speed wm, which
is obtained from either a speed transducer or a position encoder or a servo resolver. The speed error thus
obtained is passed through a zero steady-state error controller like a PI controller to obtain the command
value for the quadrature component of the stator current isqref (i.e., torque reference) in the synchronous
reference frame.
The reference for the direct component isdref of the stator current space vector in the case of the rotor
field-oriented control can be a constant value upto base speed operation of the motor. For the operation
of the motor above the base speed, the isdref decreases in such a manner so as to maintain the power constant,
that is, by weakening the field. The command values isdref and isqref are compared with the fed back values of
the stator currents isd and isq in the synchronous reference frame. The current errors thus obtained are passed
through PI controllers which form the current controllers of the drive system.
Decoupling Network Under rotor field orientation, the stator current components isd and isq are decou-
pled, and hence the outputs of the current controllers can be used as command values for the current source
inverter. However, in the case of a voltage source inverter, the stator voltage command values –Vsd and Vsq
are not decoupled as is clear from the d–q model shown in Figure 8.36. Hence decoupling networks are
necessary to generate Vsdref and Vsqref in the synchronous reference frame if a voltage source inverter is used.
In the present work, a voltage source inverter is used to drive the induction motor. Therefore, suitable
decoupling terms will have to be incorporated to the outputs of the current controllers. Referring to the
stator circuit of the d–q axes equivalent circuit of the induction motor shown in Figure 8.36, it is observed
that the d-axis stator circuit loop has a coupling term (waysq) from the quadrature axis and the q-axis stator
circuit loop has a coupling term (–waysd) from the direct axis. If these coupling terms are not compensated,
then the torque and the flux components of the stator current will not be decoupled. Thus feed-forward
terms, Vsd0 for d-axis voltage compensation and Vsq0 for q-axis voltage compensation, must be added to the
output of the current controllers. Vsd0 and Vsq0 are given by
Chapter 10.indd 630 3/18/2009 1:13:12 PM

Vsd0 = −ω s Lσ isq
Vsq0 = ω s Lσ isd + ω s ( Lss − Lσ )imr (10.93)
Equations (10.93) gives the decoupling terms that must be added to the outputs of the corresponding current
controller outputs to ensure proper decoupling between the direct and the quadrature axis voltages.
PWM Modulator The voltage source inverter operates from a constant DC-link voltage. The variable
frequency and variable voltage three-phase output is obtained by suitable pulse-width modulation. A large
variety of methods for pulse-width modulation have been developed over the past few decades. Their
implementation in the design of an AC drive system depends on the machine type, power level and the
semiconductor devices used in the inverter.
Pulse-width modulation strategies for induction motor drives also influence the structure of the current
control loops. The current control is performed in the stator reference frame. The references for the d and q
axes components (isdref and isqref ) of the stator current space vector are generated in the synchronous reference
frame. They are then transformed to the three-phase stator reference frame as isaref , isbref and iscref . These three-
phase stator current references are compared with the measured three-phase stator current components isa, isb
and isc and the corresponding three-phase current errors are passed through a suitable current controller and
pulse-width modulator to obtain the switching pattern for the inverter switches. The current controller and
the PWM generator can be current hysteresis type or carrier based. The implementation algorithm of
three-phase PWM is discussed in Chapter 12.
10.13 Output Feedback
T ill now the applications were considered wherein the controller can be the PI or PID type controller or
the full-state feedback controller. There is lot of intuition used in the design with PI type controller.
However, in the case of full-state feedback controller, once the desired closed-loop pole locations are decided,
the controller design is algorithmic. The output feedback method considers only measurable state feedback
instead of full-state feedback. All the state components are seldom available for feedback purposes in a prac-
tical design problem. As a consequence, the non-available state components are estimated from the available
states and the system model, using a reduced order observer. But unfortunately, the state estimates will
depend on the accuracy with which the system parameters are known. In the output feedback approach,
only the measurable or the available state components of the system are used for feedback control. Further,
unlike the full-state feedback, the output feedback control law can be used to design compensators with any
desired dynamical structure, thereby regaining much of the intuition of classical controls design.
In many control design applications, it is necessary to have some experience and knowledge which can
give indications about the type of the compensator dynamics that would be suitable for the specific applica-
tion. For example, it may be necessary to augment some feed-forward paths with integrators to obtain a
steady-state error as exactly zero. Again, low-pass filters may be required to filter measurements that are noisy.
In output feedback control, a dynamic compensator of a prescribed structure may be incorporated into the
system.
Consider Figure 10.70 which gives the control structure of a tracker with output feedback wherein the
plant is described by
xk +1 = Axk + Buk (10.94a)
yk = Cxk (10.94b)
Chapter 10.indd 631 3/18/2009 1:13:12 PM

Performance output z (t )
H
−
Dynamic − y(t)
compensator, L Plant C
r (t ) + e(t ) v u x
w (t ) − state vector
−
K
Measured output
Figure 10.70 Output feedback control structure.
with state x, control input u and measured output y available for feedback. In addition
z k = Hxk
is a performance output or the controlled output which must track the given reference input r. It is not generally
the same as y in all cases.
The dynamic compensator has the form
wk +1 = Fwk + Gek (10.95a)
vk = Jwk + Mek (10.95b)

with state w, output v and the input equal to the tracking error given by
ek = rk − z k (10.95c)
Also, F, G, J and M are known matrices chosen to include the desired structure in the compensator. The
allowed form for the plant control input is
uk = − Kyk − Lvk (10.96)
where the constant gain matrices K and L are to be chosen in the controls design step to result in satisfactory
tracking of r. This formulation allows for both feedback and the dynamic compensator dynamics.
Equation (10.96) is the control law for the output feedback structure shown in Figure 10.70. The augmented
system including the plant and the dynamic compensator can be represented using Eqs. (10.94a)–(10.96) as
⎡x ⎤ ⎡ A 0 ⎤ ⎡ x ⎤ ⎡B ⎤ ⎡0⎤
⎢w ⎥ = ⎢−GH ⎥ ⎢ ⎥ + ⎢ ⎥ uk + ⎢ ⎥ rk
F ⎦ ⎣w ⎦ k ⎣ 0 ⎦
(10.97)
⎣ ⎦k +1 ⎣ ⎣G ⎦
⎡ y⎤ ⎡ C 0⎤⎡x ⎤ ⎡ 0 ⎤
⎢ v ⎥ = ⎢− MH + r
J ⎥⎦ ⎢⎣w ⎥⎦k ⎢⎣ M ⎥⎦ k
(10.98)
⎣ ⎦k ⎣
⎡x ⎤
z = [H 0] ⎢ ⎥ (10.99)
⎣w ⎦ k
Chapter 10.indd 632 3/18/2009 1:13:12 PM

and the control input may be expressed as

⎡x ⎤
uk = [ − K −L] ⎢ ⎥ (10.100)
⎣ v ⎦k
The augmented plant description of Eqs. (10.97)–(10.99) can be rewritten by redefining the state, the
measured output and the matrix variables such that
xk +1 = Axk + Buk + Erk (10.101a)
yk = Cxk + Fr k (10.101b)
z k = Hxk (10.101c)
with zk the performance output which is required to track the reference input r. In terms of the new variables,
the control law for output feedback is of the form
uk = − Kyk = − KCxk − KFrk (10.102)
with the controller gain K to be determined. Using the control law of Eq. (10.68) in Eq. (10.67a), the
closed-loop system is
xk +1 = ( A − BKC )xk + ( E − BKF )rk = Ac xk + Bc r k (10.103)
Note that in the above formulation of the tracker problem with output feedback, the structure of the control-
ler can be fixed in terms of classical intuition of the controls design. Since the performance specifications of
most control systems are given in terms of time-domain criteria and these criteria are closely related to the
step response, the reference input r will be considered as a step command with magnitude R. The resulting
tracker will work for any arbitrary reference command r.
The dynamics of the augmented system is dictated by A – BKC. The design now amounts to the selection
of gain K for suitable performance. Since K appears in the feedback and the feed-forward portions of the
control input given in Eq. (10.102), this approach amounts to determining K so that both the closed-loop
poles and zeros are placed within the unit circle in the z-plane. The Ackerman’s formula can be used for deter-
mining KC, that is, KC = acker (A, B, Pc), where Pc is the closed-loop pole locations. As the C matrix is
known, K can now be obtained by LU decomposition of KC.
In the steady state, xk +1 = xk = x ∞, where x ∞ is the state under steady-state operating conditions given by
x ∞ = ( I − Ac )−1 Bc r∞
where r∞ is the reference input under steady-state operating conditions.
The output feedback approach can be understood by applying the structure to an induction motor plant
which is discussed in the following section.
10.14 Induction Motor Control with Output Feedback
T he output feedback approach will be used for control of induction motor drive system. The induction
motor can be split into two dynamical systems as shown in Figure 10.71. One of the systems consists
of the speed dynamics. A speed controller is used in the outer loop of the induction motor drive system to
account for the speed dynamics. The other system consists of the torque dynamics which is governed by the
rotor flux state model of the induction motor. The inner current control loops (i.e., the d-controller and the
Chapter 10.indd 633 3/18/2009 1:13:12 PM

Torque dynamics
isd
Vsd
(d –q model)
Torque isq
yrd
dynamics of
the induction motor
Vsq
yrq
tm = J/B
Drive torque Td + wm
generation
−
TL
Speed dynamics
Figure 10.71 Induction motor plant split as torque and speed dynamical systems.
q-controller) are designed to take care of the torque dynamics. The torque dynamics can be analyzed and the
current controllers can be synthesized using only the torque model. As the mechanical time constant of the
machine is generally very large compared to the electrical time constant of the induction motor, a reduced
order model involving only the speed dynamics can be used to synthesize the speed controller by ignoring
the current control loop dynamics.
Two digital current controllers, the d-axis and the q-axis controllers, can be synthesized simultaneously.
In other words, the d-axis and the q-axis current control loops will be closed simultaneously. Later the
speed controller can be synthesized using a reduced order model of the induction motor. In this way,
output feedback can be used for synthesizing the current and speed controllers to obtain the controller
gains. The torque model is a two-input, two-output system for the current controllers and a SISO system
for the speed controller. The controller gains are obtained with the added advantage that classical intuition
can still be brought forth in the choice of the dynamic control structure for the current- and the speed-
controlled system.
In building the system model of the induction motor drive system with output feedback, one has to consider
the dynamics of the compensator and other delays that are encountered in the control loop. Referring to
Figure 10.71, the dynamics of the induction motor model that is transformed to the discrete domain and the
dynamics of the d-axis and q-axis compensators form a part of the system formulation. The delay due to com-
putation and the inverter delays will also have to be taken in account. Rather than exactly calculating the
Chapter 10.indd 634 3/18/2009 1:13:12 PM

computation and the inverter delays, one could include all delays in the control loop into one sample time.
A sample time delay can be introduced in the d-axis and the q-axis loops to account for the various delays in the
loops as shown in Figure 10.71. To facilitate modeling of the induction motor drive system to conform to the
output feedback control structure of Figure 10.70, the inverters before the ZOH blocks have been introduced.
Since the field-oriented control is performed using a digital processor, the analog stator currents should
be filtered to remove anti-aliasing effects after sampling. Also because the anti-alias filter is in the control
path, the dynamics of the anti-alias filter should also be included in the system formulation. As the two-
phase to three-phase and three-phase to two-phase transformations are algebraic operations, the plant that is
considered is an equivalent two-phase induction motor that is described by Eq. (8.128a).
From the knowledge and experience with induction motor, one can arrive at a suitable compensator
structure. The compensator shown in Figure 10.72 has an integrator and a proportional component
(proportional–integral compensator). It should be noted that high dynamic performance is expected from
the induction motor (i.e., the plant) when the arbitrary axis is oriented along one of the flux linkages of the
motor. In the case of rotor field-oriented control, along with the use of the decoupling network the motor
model reduces to a second-order system. This is because under rotor field orientation, the yrq components
vanish and the yrd component is maintained constant. Each of the current control loops for isd and isq now
see only a first-order plant. On the other hand, when the field is not oriented, then each current control loop
sees a fourth-order plant.
It is important to note that the performance requirements are met only under the condition of rotor
field orientation. But, if there exists some field disorientation, then the performance criteria like speed of
a
− isda s +a
isdref e1d T(z + 1) e2d +
z −1 K id
+ ed 2(z − 1) Anti-alias
Integrator + filter
Delay i-gain
K pd Vsd
ZOH
−1 Plant
isd
p-gain
isq
ZOH
Gp(s)
−1
Vsq
isqref eq e1q T(z + 1) e2q + a

z −1 K iq
+ 2(z − 1) s+a
isqa
− +
Anti-alias
filter
K pq
Figure 10.72 Controller design block diagram.
Chapter 10.indd 635 3/18/2009 1:13:12 PM

response, overshoots, undershoots, etc. will not be met. However, the objective is to synthesize controllers
wherein the stability and performance criteria are met when the arbitrary axis is oriented along the rotor
field and at least stability criteria is met when the arbitrary axis is not oriented along the rotor field (wherein
the plant is a fourth-order system). Therefore, it is important that the stability of the system is maintained
even though there may be loss of performance due to field disorientation.
Plant Dynamics
The induction motor dynamics for torque or current control is given by Eq. (8.128a). Here, the state variables
and the input variables (isd, isq, yrd, yrq, Vsd and Vsq) are re-defined to represent deviations in the neighborhood
of an equilibrium point. The slip is not considered as an input but the variations in the slip of the induction
motor are equivalently considered as being caused due to the uncertainty in the rotor time constant. Robustness
to such uncertainties in the parameters of the induction motor will be treated in the next chapter. The plant Gp
is described by matrices Ap, Bp, Cp and Dp which are given in the continuous domain by
⎡ − 1 / τσ ωm + ωsl ν r / Lσ τ r ν r ωm / Lσ ⎤
⎢−(ω + ω ) − 1 / τσ ν r ωm / Lσ ν r / Lσ τ r ⎥⎥
Ap = ⎢
m sl
⎢ Rrν r 0 −1 / τ r ωsl ⎥
⎢ ⎥
⎣ 0 Rrν r −ωsl −1 / τ r ⎦
⎡1 / Lσ 0 ⎤
⎢ 0 1/ L ⎥
Bp = ⎢
σ⎥
⎢ 0 0 ⎥
⎢ ⎥
⎣ 0 0 ⎦
⎡1 0 0 0 ⎤
Cp = ⎢ ⎥
⎣0 1 0 0 ⎦
⎡0 0 ⎤
Dp = ⎢ ⎥ (10.104)
⎣0 0 ⎦
where the plant state is given by

x p = [isd , isq , ψ rd , ψ rq ]T
The input to the plant is given by
up = [Vsd , Vsq ]T
The output of the plant is given by
y p = [isd , isq ]T
Anti-Alias Filter Dynamics and Augmentation

The plant Gp(s) (i.e., the induction motor) is a low-pass filter. As long as the sampling frequency is selected
at least twice as large as the plant cut-off frequency, the effects of aliasing will be small. However, if current
sensors are used to measure the stator currents, then the sensors will introduce measurement noise. The
high-frequency measurement noise may be aliased down to lower frequencies that are within the plant
Chapter 10.indd 636 3/18/2009 1:13:12 PM

bandwidth and thus have a detrimental effect on the system performance. To avoid this, a low-pass anti-
aliasing filter of the form
a
Ha = (10.105)
s +a
is inserted after the measuring devices and before the samplers.
If the cut-off frequency of the anti-aliasing filter is not much higher than the plant cut-off frequency,
then the filter will affect the closed-loop performance due to the finite attenuation and phase shift intro-
duced to the measured signals. Therefore, the filter dynamics should be appended to the plant at the design
stage itself so that the controller is designed taking it into account. As there are two measured outputs
(isd and isq), there will be two anti-aliasing filters of the form indicated in Eq. (10.105), one for filtering
isd and the other for filtering isq. Therefore, if Ga represents the anti-alias filter for both the axes, then the
corresponding matrices, Aa, Ba, Ca and Da, are given by
⎡−a 0 ⎤ ⎡a 0⎤
Aa = ⎢ ⎥ ; Ba = ⎢
⎣0 − a ⎦ ⎣0 a ⎥⎦
(10.106)
⎡1 0 ⎤ ⎡0 0⎤
Ca = ⎢ ⎥; Da = ⎢
⎣ 0 1⎦ ⎣0 0 ⎥⎦
where the anti-alias filter state is given by

xa = [isda , isqa ]T
The input to the anti-alias filter is given by
ua = [isd , isq ]T
The output of anti-alias filter is given by
ya = [isda , isqa ]T
The induction motor plant Gp and the anti-alias filter Ga are in the continuous domain (s-domain) whereas
the controllers are in the discrete domain (z-domain). The discrete and continuous domains are interfaced
using ZOH and samplers as shown in Figure 10.72. The plant-anti-alias filter augmented system Gpa is
described by the matrices Apa, Bpa, Cpa and Dpa which are given by
⎡ apa11 apa12 ⎤
Apa = ⎢ ⎥ (6 × 6 matrix) (10.107a)
⎢⎣ apa21 apa22 ⎥⎦
where
apa11 = Ap (4 × 4 matrix)
apa12 = 0 (4 × 2 matrix)
⎡a 0 0 0 ⎤
apa21 = ⎢ ⎥
⎣0 a 0 0 ⎦
Chapter 10.indd 637 3/18/2009 1:13:12 PM

apa22 = Aa (2 × 2 matrix)
⎡bpa11 ⎤
Bpa = ⎢ ⎥ (6 × 2 matrix) (10.107b)
⎢⎣bpa21 ⎥⎦
where
bpa11 = Bp (4 × 2 matrix)
bpa21 = 0 (2 × 2 matrix)
The state of Gpa is given by
x pa = [ x p , xa ]T
The input to Gpa is given by
upa = up
The output of Gpa is given by
y pa = ya
The augmented plant Gpa(s) is transformed to the discrete domain using the ZOH equivalent transforma-
tion to obtain Gpa(z).
Controller Dynamics and Augmentation

Referring to Figure 10.72, the unit delay term is used to include all the computational and loop delays that
may occur. This unit delay is shifted before the proportional–integral control structure to make the controller
formulation easier without loss of generality.
Considering first the delay dynamics, one obtains
⎡ε1d ⎤ ⎡−isda ⎤ ⎡isdref ⎤
⎢ε ⎥ = ⎢−i ⎥ + ⎢i ⎥ (10.108)
⎣ 1q ⎦k +1 ⎣ sqa ⎦k ⎣ sqref ⎦k
Considering now the proportional–integral dynamics, one obtains
⎡ε 2d ⎤ ⎡(T / 2)ε1d ε 2d ⎤ ⎡(T / 2)ε1d ⎤
⎢ε ⎥ = ⎢(T / 2)ε ε ⎥ + ⎢(T / 2)ε ⎥ (10.109)
⎣ 2q ⎦k +1 ⎣ 1q 2q ⎦k ⎣ 1q ⎦k +1
where T is the sampling time of the analog signals. Substituting Eq. (10.108) in Eq. (10.109), the controller
dynamics is given by
⎡0 0 0 0⎤ ⎡−1 0 1 0 ⎤
⎡ε1d ⎤ ⎢0 0 0 0 ⎥ ⎡ ε1d ⎤ ⎢
0 − 1 0 1 ⎥⎥ ⎡ sda ⎤
i
⎢ε ⎥ ⎢ ⎥ ⎢ε ⎥ ⎢ ⎢i ⎥
⎢ 1q ⎥ = ⎢T ⎥ ⎢ 1q ⎥ ⎢ T T ⎥ ⎢ sqa ⎥
⎢ε 2d ⎥ ⎢ 0 1 0 ⎥ ⎢ ⎥ + ⎢− 0 0⎥ ⎢ (10.110)
ε 2d i ⎥
⎢ ⎥ ⎢2 ⎥⎢ ⎥ ⎢ 2 2 ⎥ ⎢ sdref ⎥
⎢⎣ε 2q ⎥⎦k +1 ⎢0 T ε
0 1⎥⎥ ⎢⎣ 2q ⎥⎦k ⎢⎢0 −
T
0
T ⎥ ⎢⎣isqref ⎥⎦
k
⎢⎣ 2 ⎦ ⎣ 2 2 ⎥⎦
Chapter 10.indd 638 3/18/2009 1:13:13 PM

where
⎡0 0 0 0⎤
⎢0 0 0 0⎥
⎢ ⎥
Ac = ⎢⎢T 0 1 0⎥
⎥
⎢2 ⎥
⎢0 T
0 1 ⎥⎥
⎢⎣ 2 ⎦
⎡ −1 0 1 0 ⎤
⎢0 −1 0 1 ⎥
⎢ ⎥
⎢
Bc = ⎢ − T T ⎥
0 0⎥
⎢ 2 2 ⎥
⎢0 − T 0 T ⎥
⎢⎣ 2 2 ⎦⎥
The state of Gc is given by
x c = [ε1d , ε1q , ε 2d , ε 2q ]T
The input to Gc is given by
uc = [isda , isqa , isdref , isqref ]T
The output of Gc is given by
yc = xc
Augmenting the controller Gc to the plant-filter dynamics Gpa, the total system dynamics G is obtained.
If the total system (plant-filter and controller) is described by the matrices A, B, C, E and F as described in
Eqs. (10.101a), (10.101b), (10.101c), one obtains
⎡a11 a12 ⎤
A=⎢ ⎥ (10 × 10 matrix) (10.111)
⎣a21 a22 ⎦
where
a11 = Apa (6 × 6 matrix)
a12 = 0 (6 × 4 matrix)
⎡0 0 0 0 −1 0 ⎤
⎢0 0 0 0 0 −1 ⎥
⎢ ⎥
a21 = ⎢⎢0 0 0 0
−T
0 ⎥
⎥
⎢ 2 ⎥
⎢0 −T ⎥
⎢⎣ 0 0 0 0
2 ⎥⎦
a22 = Ac (4 × 4 matrix)
Chapter 10.indd 639 3/18/2009 1:13:13 PM

⎡b11 ⎤
B=⎢ ⎥ (10 × 2 matrix) (10.112)
⎣b 21⎦
where
b11 = Bpa (6 × 2 matrix)
b21 = 0 (4 × 2 matrix)
⎡e11 ⎤
E =⎢ ⎥ (10 × 2 matrix) (10.113)
⎣e 21 ⎦
where
e11 = 0 (6 × 2 matrix)
⎡1 0⎤
⎢0 1⎥
⎢ ⎥
e 21 = ⎢⎢T 0⎥
⎥
⎢2 ⎥
⎢0 T⎥
⎢⎣ 2 ⎥⎦
⎡0 0 0 0 0 0 1 0 0 0⎤
⎢0 0 0 0 0 0 0 1 0 0 ⎥⎥
C =⎢ (10.114)
⎢0 0 0 0 0 0 0 0 1 0⎥
⎢ ⎥
⎣0 0 0 0 0 0 0 0 0 1⎦
F=0 (4 × 2 matrix) (10.115)

For the augmented total system shown in Figure 10.73, the state of G is given by
isdref
r=
isqref isda
A B E Z(t ) =
x(t ) H isqa
(10 × 10) (10 × 2) (10 × 2)
(2 × 10)
Vsd
u(t ) = Ax + Bu + Er
Vsq
−K
(2 × 4)
C
(4 × 10)
e1d
e1q
y (t ) = e
2d
e2q
Figure 10.73 Augmented system of the current control system for the induction
motor with output feedback.
Chapter 10.indd 640 3/18/2009 1:13:13 PM

x = [ x pa , x c ]T (totally there are 10 state variables)

The input to G is given by
u = upa = up
The reference input of G is given by
r = [isdref , isqref ]T
The output of G is given by

y = yc
The performance output for the system is
z = [isda , isqa ]T = Hx
and the performance output matrix H is given by
⎡0 0 0 0 1 0 0 0 0 0 ⎤
H =⎢ ⎥
⎣0 0 0 0 0 1 0 0 0 0 ⎦
Referring to Figure 10.72, the control input to the plant-filter upa is given by
⎡ε1d ⎤
⎡Vsd ⎤ ⎡ K pd 0 K id 0 ⎤ ⎢⎢ε1q ⎥⎥
upa = ⎢ ⎥ = − ⎢ ⎥ (10.116)
⎣Vsq ⎦ ⎢⎣0 K pq 0 K iq ⎥⎦ ⎢ε 2d ⎥
⎢ ⎥
⎢⎣ε 2q ⎥⎦
which is of the accepted control law u = –Ky where
⎡ K pd 0 K id 0 ⎤
K =⎢ ⎥
⎢⎣0 K pq 0 K iq ⎥⎦
From Eqs. (10.111)–(10.116), the induction motor is now formulated as an output feedback structure as
given in Eq. (10.101). The dynamics of the system is determined by A – BKC which is used to design the
value of K as discussed in the previous section.
Controller design is one of the aspects of a closed-loop especially is very popular for stable non-minimum
system that is most essential for the system perfor- phase systems. However, the Bode diagram has its
mance but least understood from the implementation limitations as discussed in this chapter. The root
perspective. This chapter endeavors to give controller locus method is definitely more complete in the
design insights through illustrative power electronic sense that it can handle any type of system as long as
system examples. it is linear and time invariant.
The classical controller design methods are still The state space methods are very powerful. The
popular but for smaller systems. The Bode diagram only uncertainty is in modeling of the physical system.
Chapter 10.indd 641 3/18/2009 1:13:13 PM

The quality of the performance is directly related to insight about the system and also reduce the number
the accuracy of the model of the physical system. of physical sensors that one needs to use. Consequently,
Once the modeling process is successfully accom- the cost and the reliability of the overall system will
plished, the state space methods can give closed-form improve. The controller design is also an art. Experi-
solutions with guaranteed stability. Compare and ence and intuition can lead to quick and better
contrast this with the classical methods wherein there controllers especially using the classical methods or
are a number of uncertainties in the controller design the output feedback methods. The next section gives
process. The modeling of the system, the choice of the some tutorial exercises on the three primary DC–DC
controller pole–zero structure and the choice of converters. These should be studied in the light of the
the controller gain are open aspects leading to many discussions in this chapter, using a computer-aided
trials before obtaining a successful controller. programming environment for faster iteration and
The estimator design is another important aspect removing the drudgery of manual computation and
of the state space approach. This can bring in greater fatigue.
The following tutorial exercises can be studied in 3. The small-signal dynamic transfer function for
either MATLAB or SciLAB environment. the boost converter is given below. The control
input is the duty cycle input pin and the con-
1. Implement a PID module in a simulation envi-
trolled output is the output voltage Vo. The
ronment with anti-windup feature.
main performance objective is to attain zero
2. The small-signal dynamic transfer function for steady-state error. Choose the sampling fre-
the buck converter is given below. The control quency to be same as the switching frequency.
input is the duty cycle input pin and the con-
trolled output is the output voltage Vo. The main ⎛ L ⎞
performance objective is to attain zero steady- ⎜⎜ 1 − s ⎟⎟
R (1 − D )
2
state error. Choose the sampling frequency to be vô Vi ⎝ ⎠
(s ) =
same as the switching frequency. dˆ (1 − D )2 ⎛ L LC ⎞
⎜⎜ 1 + s + s2 ⎟⎟
vô Vi ⎝ R (1 − D ) 2
(1 − D)
2
⎠
(s ) =
dˆ ⎛ L 2 ⎞
⎜ 1 + s + s LC ⎟
⎝ R ⎠ where Vi = 10 V, L = 1 mH, C = 100 μF,
where Vi = 10 V; L = 10 mH; C = 100 μF; R = 20 Ω; D = 0.4 and switching frequency is
R = 10 Ω and switching frequency is 10 kHz. 10 kHz. Vi is the input source voltage, D is the
Vi is the input source voltage, d is the operating nominal operating duty cycle, R is the load
duty cycle, R is the load resistance and Vo is the resistance, Vo is the output voltage and C is the
output voltage and C is the output capacitor. output capacitor.
Design digital controller for the above system, (a) Design digital controller for the above
using both the root locus approach and the system using full-state feedback approach
full-state feedback approach. (Note: capacitor with and without full-order estimator and
voltage is a state variable and the inductor simulate the complete system.
current is a state variable.)
Chapter 10.indd 642 3/18/2009 1:13:13 PM

(b) For a reference input tracking problem, where Vi = 10 V, L = 1 mH, C = 100 uF, R = 20 Ω,
design K, Nx, Nu and G (full-order prediction D = 0.4 and switching frequency is 10 kHz.
observer) and simulate the complete system.
Vi is the input source voltage, D is the nominal
4. The small-signal dynamic transfer function operating duty cycle, R is the load resistance,
for the buck–boost converter is given below. Vo is the output voltage and C is the output
The control input is the duty cycle input pin capacitor.
and the controlled output is the output voltage
(a) Design digital controller for the above
Vo. The main performance objective is to attain
system using full-state feedback approach
zero steady-state error. Choose the sampling fre-
with and without full-order estimator and
quency to be same as the switching frequency.
simulate the complete system.
⎛ DL ⎞ (b) For a reference input tracking problem,
⎜⎜ 1 − s ⎟⎟
vô −Vi ⎝ R ( − D )2
1 ⎠ design K, Nx, Nu and G (full-order predic-
(s ) = tion observer) and simulate the complete
dˆ (1 − D )2 ⎛ L LC ⎞
⎜⎜ 1 + s + s2 ⎟⎟ system.
R (1 − D ) (1 − D )2
2
⎝ ⎠

1. The controller design addresses the three perfor- 9. Noise is dominant at frequencies.
mance criteria, , and .
10. The order of a practical PID controller is
2. By Bode diagram approach, it is important to .
maintain the slope of the amplitude–log fre-
11. The root locus method is a graphical technique
quency curve in the vicinity of the gain cross-
for determining the poles of a system
over frequency at a slope of .
as a function of the .
3. Bode diagram method should not be applied
12. The transfer function assumes con-
for phase systems.
ditions.
4. In Bode diagram the magnitude is given as
13. The transfer function is applicable only for
and the phase angle is .
linear time-invariant input single
5. In Bode diagram, the two parameters used for systems.
the controller design are and .
14. The state space approach is applicable for
6. Both gain and phase margins are measures of input output systems.
.
15. A system is controllable if there exists a
7. The steady-state error is addressed by the sequence of such that it transfers the
of the PID controller. system from any state to any final
state in time.
8. The proportional and derivative parts of the
PID controller address the criteria.
Chapter 10.indd 643 3/18/2009 1:13:13 PM

16. A system is said to be provided that 23. The region beyond base frequency is called the
any initial state can be calculated from finite region.
number of .
24. To compensate for the drop, a small
17. The controller for a full-state feedback system boost to the stator terminal voltage is provided
is a matrix of . at low frequencies.
18. In full-state feedback control, the closed-loop 25. The torque producing and the flux producing
poles of the system can be placed components of the stator current are
within the unit circle. in field-oriented control of induction motor.
19. The controller design and the estimator gain 26. The output feedback method considers only
design can be done and of state feedback instead of full-state
each other. feedback.
20. To ensure that the power calculated in the sta- 27. The output feedback control law can be used to
tionary axis is the same as the power calculated design compensators with any desired
in the rotating reference axis, a factor of structure, thereby regaining much of the intu-
is used as the power balance factor. ition of classical controls design.
21. V/f control of induction motor falls in the class 28. In output feedback the controller gain is deter-
of control. mined so that both the closed-loop
and are placed within the unit circle
22. In stator voltage control if the stator voltage is
in the z-plane.
reduced, the flux .
1. What are the classical and modern methods of 9. What is the transfer function of the PID
controller design? controller?
2. Write short notes on: (a) Bode’s theorem I and 10. How are the constants Ki, Kp and Kd for a PID
(b) Bode’s theorem II. controller selected?
3. What is a non-minimum phase system? Give 11. Write short notes on: (a) Integrator windup,
an example. (b) anti-windup.
4. Discuss Bode diagram construction. 12. What is a root locus? Explain.
5. Write short notes on: (a) Gain margin, (b) Phase 13. Discuss the design steps for designing a
margin. controller using the root locus technique.
6. Discuss the design steps in the controller design 14. Write short notes on: (a) continuous state
with Bode diagram. equation, (b) discrete state equation.
7. Discuss the PID controller. 15. Write short notes on: (a) controllability and
(b) observability.
8. Explain the step response of a closed-loop
system that has a PID as its controller. 16. What is full-state feedback?
Chapter 10.indd 644 3/18/2009 1:13:13 PM

17. Distinguish between regulator problem and 36. Discuss the tracker problem? Is the controller
tracking problem? design same as that of the regulator problem?
18. What is the regulator control law? Explain. 37. How are Nx and Nu matrices designed for a
specific system?
19. Discuss regulator design by pole placement.
38. How is the output voltage fed back to the
20. What is an estimator? Is there a difference
controller side of the DC–DC converter with
between an estimator and an observer?
opto-isolation? How is the feedback opto-
21. Discuss the full-order prediction estimator. coupler’s temperature drift compensated?
22. What is the estimator state equation for a 39. What are the different ways of regulating the out-
full-order prediction estimator? puts of multiple-output DC–DC converters?
23. Discuss the design steps to obtain the prediction 40. Write short notes on: (a) coupled inductor,
estimator gain matrix. (b) magnetic amplifier.
24. How can the same Ackermann’s formula be 41. What are the drawbacks of the coupled induc-
used for obtaining the prediction estimator tor approach to multiple-output regulation?
gain matrix?
42. In current control of forward converter, why
25. What is separation principle? Explain. should the duty cycle be limited to 0.5?
26. What is the dynamics of the controller and 43. What is the principle behind slope compensa-
estimator together? tion in current programmed converters?
27. Discuss the full-order current estimator. 44. Explain the operation of the boost converter as
a unity power factor converter?
28. What is the estimator state equation for a
full-order current estimator? 45. Discuss the reference current generation for
single-phase unidirectional unity power factor
29. Explain the concept of time update and
converter?
measurement update in current estimator.
46. How is the inverter voltage reference generated
30. Discuss the design steps to obtain the current
for single-phase front-end converter in order to
estimator gain matrix.
improve the power factor?
31. How can the same Ackermann’s formula be
47. Discuss the single-phase front-end converter
used for obtaining the current estimator gain
operation and how it can be extended to three-
matrix?
phase systems?
32. Why is reduced order estimator needed? What
48. Explain the vector control of a three-phase
are its benefits as compared to a full-order
front-end converter for unity power factor at
estimator?
the grid.
33. Explain reduced order prediction estimator?
49. What is power balance factor in vector-
34. Explain reduced order current estimator? controlled systems?
35. How can the same Ackermann’s formula be 50. Discuss the steady-state equivalent circuit of an
used for obtaining the reduced order estimator induction motor.
gain matrix?
Chapter 10.indd 645 3/18/2009 1:13:13 PM

51. Why should a stator voltage boost be provided 56. Write short notes on: (a) direct field-oriented
at low applied frequencies? control, (b) indirect field-oriented control.
52. Discuss the frequency versus applied voltage 57. Discuss the block schematic of the direct and
characteristics that is required for V/f control? indirect vector control of induction motor.
Compare with the frequency versus back emf
58. What is decoupling network in vector control
characteristics.
of induction motor? Discuss with respect to
53. Distinguish between open-loop V/f and closed- the rotor field-oriented control.
loop V/f operation.
59. What is the control law for output feedback?
54. What are the main conceptual differences
60. What are the benefits and drawbacks of output
between V/f control and vector control of
feedback vis-à-vis the full-state feedback?
induction motor?
61. How can the Ackermann’s formula be used to
55. What is achieved by aligning the synchronously
determine the output feedback controller gain
rotating reference frame along the rotor flux
matrix?
axis?
| PROBLEMS
1. A unity feedback control system has an open- Plot the root loci. Is the root loci entirely
loop transfer function given as within the stable region of the s-plane? At what
value of the controller gain, does the system
10
G(s ) = become unstable?
s (1 + 0.2 s )(1 + 0.5s )
4. A unity feedback control system has an open-
Plot the Bode diagram. Find out the gain cross- loop transfer function given as
over frequency. What is the gain margin and
z ( z + 1)( z + 0.2)
phase margin? G (z ) =
( z + 0.3)( z 2 − 0.5z + 0.5)
Plot the root loci. Is the root loci entirely
loop transfer function given as
within the stable region of the s-plane? At what
z ( z + 1)( z + 0.2) value of the controller gain, does the system
G (z ) = become unstable?
( z + 0.3)( z 2 − 0.5z + 0.5)
5. Derive Ackermann’s formula for controller gain
Plot the Bode diagram. Find out the gain cross- calculation.
over frequency. What is the gain margin and
phase margin? 6. Derive the error dynamics for a full-order pre-
diction estimator?
loop transfer function given as 7. Derive the error dynamics for a full-order cur-
rent estimator?
10
G(s ) = 8. Suggest methods other than opto-isolation, to
s (1 + 0.2 s )(1 + 0.5s ) feedback the isolated output voltage to the
Chapter 10.indd 646 3/18/2009 1:13:14 PM

controller side of the DC–DC converter with- domain using zero-order hold. From the dis-
out sacrificing galvanic isolation. crete transfer function obtain the difference
equations giving the discrete state equations of
9. Show that the flux in the induction machine is
the system.
constant if the ratio of back emf to the applied
frequency is maintained constant. 15. For Problem 14, design the following:
10. Deduce the rotor flux estimate model for rotor a. Full-state feedback controller
field-oriented control.
b. Dead beat controller
11. Deduce the stator flux estimate model for rotor
c. Full-order prediction estimator
field-oriented control.
d. Full-order current estimator
12. A plant transfer function in the continuous
domain is given as e. Reduced order prediction estimator
f. Reduced order current estimator
K g ( s + b )ω r2
Gp ( s ) = g. Nx and Nu matrices for the tracking problem
s 2 ( s 2 + 2ζω r s + ω r2 )
16. Using the root locus technique, design control-
where the sampling rate is 0.1 ms, b = 0.0001, lers for the following plants for zero steady-state
ζ = 0.005, wr = 500, Kg =0.5.
error and fastest response with zero overshoot:
Plot the Bode diagram and find out the gain 1
and phase margins. Design a controller for a a.
z ( z − 1)
phase margin of atleast 60o and a gain margin
of atleast 6 dB. 1
b.
13. For Problem 12, plot the root loci and design a z ( z − 0.8)
controller using the root locus technique. ( z − 4)
c.
14. Consider servomotor transfer function given as ( z − 1)
10/[s(s + 1)] wherein the output is the position 1
of the shaft. The state variables of the system d.
( z − 1)2
can be considered to be the shaft position and
speed. Using a sampling rate of 0.1 s, convert
e. z2
the continuous transfer function to discrete ( z − 1)2
| ANSWERS
Fill in the Blanks
1. stability; steady-state accu- 5. gain margin; phase margin 10. two
racy; speed of response 6. relative stability 11. closed loop; gain
2. −20 dB/decade 7. integral or I-part 12. zero initial
3. non-minimum 8. speed of response 13. single; output
4. 20 ⋅ log10 G ( jω) ; ∠G ( jω) 9. higher 14. multiple; multiple
Chapter 10.indd 647 3/18/2009 1:13:14 PM

15. inputs; initial; finite 20. 2/3 25. decoupled

16. observable; measurements 21. scalar 26. measurable
17. gains 22. decreases 27. dynamical
18. anywhere 23. field weakening 28. poles; zeros
19. separately; independent 24. stator resistance
Chapter 10.indd 648 3/18/2009 1:13:14 PM

Optimal and Robust Controller Design
Learning Objectives
CHAPTER
11
understand and apply optimal controller concepts.
apply least squares and weighted least squares algorithm.
apply recursive algorithms such as least mean square and Kalman filter algorithms.
design optimal controller and estimator gains.
understand robust controller basics.
B ased on the discussions in Chapter 10, it may be noted that there are significant advantages in using the
state space approach for the design of controllers. They may broadly be classified as full-state feedback
and output feedback. In full-state feedback all the state variables of the system are required to be fed back to
the controller to take appropriate control action. As all the state variables are fed back, the entire infor-
mation about the system is available in these state variables. Hence only scalar gains are needed to appro-
priately scale these state variables to obtain the necessary control performance. Thus in full-state feedback
approach the introduction of the controller does not increase the overall order of the system with respect to
the uncontrolled open-loop system.
On the other hand, in output feedback, only the measurable states are fed back to the controller.
Hence the controller does not have the entire information about the state of the system as in the case of a
full-state feedback. As a consequence, dynamical controllers are needed to introduce additional degrees of
freedom in the synthesis of the controllers to achieve the stated performance. Thus, the overall system will
be of a higher order with respect to the order of the uncontrolled open-loop system. This is much like the
single-input single-output (SISO) case wherein only one state variable is fed back to the controller with
only partial information about the state of the system. The output feedback brings in the intuitive insights
and experience that one uses in the classical SISO controller designs.
Further, in the case of the full-state feedback approach, many a times all the state variables are not available
to be fed back due to either inaccessibility or the sensors may be too expensive. To overcome this problem,
the unavailable states are estimated by using estimators or observers that make use of the information in the
measurable states to estimate the state variable values of the unavailable states. This way the controller is fed
with all the state variables for the purposes of control. Therefore, in the design of control systems based on
the full-state feedback approach, two major blocks need to be synthesized: (a) controller; (b) estimator or
observer. This implies that the controller gain K and the estimator gain G need to be designed. The previous
Chapter 11.indd 649 3/17/2009 8:24:19 PM

chapter discussed various methods of designing the controller gain K and the estimator gain G to meet the
primary performance criteria (SSS), that is, the stability, steady-state accuracy and speed of response.
Suppose one would like to design K and G to meet the primary SSS performance criteria under some
constraints like minimum input energy or minimum state energy. Such controllers and estimators that
have been designed to meet the performance criteria under some constraints are called optimal controller
and optimal estimator, respectively. Kalman gain is one such optimal controller that achieves the control
performance criteria under the constraints of minimum input and state energies in the presence of system
noise. Kalman filter is an optimal estimator that estimates the unavailable states of a system in the presence
of noise. The optimal controllers and the optimal estimators operate on the linear quadratic principles
which will be the focus of discussion in this chapter. The topic will be developed first by starting with the
estimation and later leading to the discussion on the optimal control aspects.
11.1 Least Squares Principle

Consider a system that is represented as a state space model given as
x = Ax + Bu (11.1)
y = Cx + Du (11.2)
Referring to the output equation [Eq. (11.2)], in most practical situations, there is no direct feed through
from the input to the output. Hence D is a null matrix. Thus in majority of the situations,
y = Cx (11.3)
where y is the output that is measured, x is the state vector and C gives the relationship between the output
and the state vector. Let y be the output value that is given by Eq. (11.3) and yact represent the actual value
of the output. As the mathematical representation of the system [Eqs. (11.1)–(11.3)] will never be the exact
system representation, in practice there will always be an error e between the model value y and the actual
value yact. Thus,
Cx = y ≠ yact
The error is defined as
e = yact − y
= yact − Cx
If yact − Cx = 0 then the solution x is trivial and is given by C –1yact. However, in practice, yact − Cx is never
equal to zero. There will always exist an error between the actual or measured values and the model values as
given by Cx. In such a situation, there is no trivial solution. One will have to find the value of x that will
minimize the error yact − Cx. Such a solution for x is the least square solution. The following discussion will
evolve the least square solution for yact − Cx) = e.
Consider now a series of measurements of the actual value of the output, that is, yact1, yact2, …. as shown
in Figure11.1. The equation y = Cx gives the best estimate of the actual measured points of the output as
indicated in Figure 11.1. The central function of an estimator is to provide the value of x which will minimize
the errors. If the objective is to minimize the error, one should not consider the sum of the errors as the func-
tion to be minimized because the positive and negative errors may cancel each other and give a wrong notion
about the goodness of the fit. Alternately one may consider minimizing the sum of the error squares (e 2 ).
Chapter 11.indd 650 3/17/2009 8:24:20 PM

Optimal and Robust Controller Design 651
yact5
y
y = Cx
yact3 y4
yact1
yact4
yact2
y1
x1 x4 x
Figure 11.1 Actual values and the model values.
Referring to Figure 11.1, one has

e1 = yact1 − y1
e 2 = yact2 − y 2
(11.4)

en = yactn − yn
Let the function to be minimized be

n
f ( x ) = ∑ ei2
i =1
This can be written elegantly in vector form as
⎡ e1 ⎤
⎢e ⎥
f ( x ) = [ e1 e 2 … en ] ⎢ ⎥ = e T e
2
(11.5)
⎢ ⎥
⎢ ⎥
⎣e n ⎦
Equation (11.5) is the function that is to be minimized by choosing proper values of x. The value of x that gives
minimum value of f (x) is the best estimate of the state vector. The solution x of Eq. (11.3) is that value of x
which will provide the minimum value of f (x) and is called the least squares solution. Before obtaining the
solution for Eq. (11.3) based on the constraint that the solution should minimize Eq. (11.5), it is appropriate
at this point to slightly deviate to understand the quadratics in matrix notation called the quadratic forms.
11.2 Quadratic Forms

A quadratic form in two variables x1 and x2 is an expression that can be written as
f = ax12 + 2bx1x2 + cx22
Chapter 11.indd 651 3/17/2009 8:24:20 PM

where f is called the quadratic function. The function f can be written in matrix form as
⎡a b ⎤ ⎡ x1 ⎤
f = [ x1 x2 ] ⎢ ⎥⎢ ⎥ (11.6)
⎣b d ⎦ ⎣ x 2 ⎦
= x T Cx
Equation (11.6) is a quadratic of two dimensions. In general, a quadratic form in “n” variables x1, x2, …, xn
is defined to be an expression that can be written as
⎡ x1 ⎤
⎢x ⎥
[ x1 x2 … xn ] C ⎢ ⎥ = x T Cx
⎢⎥
2
(11.7)
⎢ ⎥
⎣ xn ⎦
where C is an n × n matrix. Equation (11.7) is called a quadratic of n dimensions. Consider the function
f (x1, x2) = ax12 + 2 bx1 x2 + cx22
where a, b and c are non-zero. It can be seen that f is zero only when x1 = x2 = 0, that is, when x = 0, a
null matrix. Now the question is when x ≠ 0 then is f > 0 or is f < 0? For the moment, consider that C in
Eq. (11.6) is an identity matrix. Then,
f = xTx
This results in f being a sum of all square terms which cannot be negative whatever be the non-zero value of
x. Therefore, if C is identity matrix, then f is always positive (i.e., f > 0). This implies that f being positive
or negative is solely determined by the character of the C matrix which contains the coefficients of the qua-
dratic function f. Thus the following cases arise:
1. If f is positive ( f > 0) for all non-zero values of x, implying xTCx > 0, such a matrix C is called positive
definite.
2. If f is negative ( f < 0) for all non-zero values of x, implying xTCx < 0, such a matrix C is called negative
definite.
3. If f is zero or positive ( f ≥ 0) for all non-zero values of x, implying xTCx ≥ 0, such a matrix C is called
positive semi-definite.
4. If f is zero or negative ( f ≤ 0) for all non-zero values of x, implying xTCx ≤ 0, such a matrix C is called
negative semi-definite.
5. If f is zero or positive or negative for all non-zero values of x, implying xTCx can either be zero, positive
or negative, such a matrix C is called indefinite.
Figure 11.2 illustrates the concept of positive definiteness, negative definiteness and indefiniteness for a
two-dimensional quadratic form, that is, the case of a function f of two variables x1 and x2. In Figure 11.2,
x1 and x2 are the two orthogonal axes in a plane and f is the axis that is orthogonal to the plane formed by
x1, x 2.
Figure 11.2(a) indicates a cup-shaped paraboloid wherein f is always positive whatever be the non-zero
value of x = [x1, x2]. This means that C is positive definite. It also indicates that at x = [0, 0], f is zero implying
that the minimum of the function f in this case is at x = 0. Figure 11.2(b) indicates an inverted paraboloid
wherein f is always negative whatever be the non-zero value of x = [x1, x2]. This means that C is negative
definite. It also shows that at x = [0, 0], f is zero implying that the maximum of the function f in this case
Chapter 11.indd 652 3/17/2009 8:24:21 PM

f f f
Inverted Saddle
cup
Cup
x1 x1
o x1
x2 x2 x2
(a) (b) (c)
Figure 11.2 Two-dimensional quadratic form: (a) Positive definiteness; (b) negative definiteness;
(c) indefiniteness.
is at x = 0. In the case shown in Figure 11.2(c), the function f is shaped like a saddle. Here f takes both
positive and negative values for non-zero values of x. The function f is zero not only at x = 0 but also at
other non-zero values of x. In such cases, the matrix C is called indefinite. There is neither a unique minimum
nor a maximum. In general,
1. If C is positive definite, there always exists a minimum.
2. If C is negative definite, there always exists a maximum.
Now the question arises as to what is the practical significance of the quadratic forms? In physical systems,
the energy is a quadratic form which is proportional to square of voltage or square of current or square of
velocity or square of angular speed, etc. In general, the square of the effort or the square of the flow in a
system represents the energy which is a quadratic form. The error square that was discussed earlier is also a
quadratic form. Many situations arise in practice where one needs to minimize either the square of the error
or the energy. Therefore, understanding the quadratic forms and the various definiteness of C is essential to
the process of optimization.
11.3 Minimum Energy Principle
A ll physical systems will reach a stable state, also called the equilibrium state, where the system will be at
its minimum energy state. As examples, a stick held vertically, will ultimately try to reach its minimum
energy state which is the horizontal position; a child sliding down a slide will reach the minimum energy
state on having reached the bottom; a capacitor connected to a load will discharge and reach the zero energy
state; and so on. Nature sees to it that maximum amount of energy is removed from the system so that it
may reach the equilibrium or the stable state.
Consider an RC circuit as shown in Figure 11.3. On closing the switch S the circuit will tend to reach
an equilibrium state. A current i will flow in the circuit which will charge up the capacitor towards Vg. As
the capacitor exponentially charges up to Vg, the current i will exponentially die down to zero. In the process
of charging up the capacitor so that the circuit reaches an equilibrium state, nature utilizes some energy
which is dissipated in the resistor R. Maximum energy will be removed from this circuit to reach equili-
brium or minimum energy state. This is true in the case of any physical system. This is called the minimum
energy principle.
Chapter 11.indd 653 3/17/2009 8:24:21 PM

v, i
i
s R
Vg
vc
Vg C vc
0 t
i
Switch
closed
(a) (b)
Figure 11.3 (a) RC circuit; (b) current and voltage waveforms of capacitor.
Referring to the RC circuit of Figure 11.3, on closing the switch S, by applying the Kirchhoff ’s voltage law
(KVL) around the loop, one has
Vg = Ri + v c
Multiplying the above equation by i and integrating, one obtains the following energy equation:
∫ Vgidt = ∫
Ri dt + ∫ v c idt
2
(11.8)
Energy lost
Equation (11.8) gives the energy equation for the RC circuit of Figure 11.3. The energy that is dissipated is
the energy that is required by nature to establish equilibrium in the circuit. Let the energy lost in the resistor
be denoted by ER. Then Eq. (11.8) can be re-written as
E R = ∫ Vgidt − ∫ v c idt (11.9)
From circuit theory, the current i through the RC circuit is given as

dv c
i =C
dt
Substituting for i in Eq. (11.9), one obtains
dv c dv
E R = ∫ VgC dt − ∫ v cC c dt
dt dt
which can be simplified as
1 (11.10)
E R = VgCv c − Cv c2
2
It is evident that Eq. (11.10) is a quadratic and the energy ER is a parabola with respect to vc. As nature
prefers to bring the circuit to equilibrium or minimum energy state, maximum amount of energy is removed
from the circuit by way of loss in R. One should find the maximum of the parabola of Eq. (11.10).
Differentiating ER with respect to vc and equating to zero, one obtains the condition for maximum value of
ER. This is given by
dE R
= VgC − Cv c = 0
dv c
vc = Vg
Chapter 11.indd 654 3/17/2009 8:24:21 PM

Applying the above condition to the loss energy equation [Eq. (11.10)], one obtains
1 1
E R = CVg2 − CVg2 = CVg2
2 2
This implies that whatever be the value of the resistor R, nature needs to remove the maximum amount of
energy equivalent to (1 / 2)CVg2 through the resistor to ensure that the circuit reaches equilibrium or the
minimum energy state. The energy equation [Eq. (11.10)] can be generalized for any physical system.
In general if EN is the energy of any system then EN has a minimum and the energy removed ER has a
maximum. The generalized energy equation is given as
1 2
EN (x ) = Ax − bx
2
dE N
= Ax − b (11.11)
dx
where x is any state variable and A and b are parameters of the system. Here again the derivative is zero at
x = b/A. If A is positive, the parabola opens towards the positive axis and this equilibrium point would be
stable. If A is negative, the parabola opens towards the negative axis and this point x = b/A would be a maxi-
mum and would not therefore be a stable point.
In the above case, x was a single-state variable, that is, a scalar quantity. However, when many state
variables are considered together, then x is a vector, A is a matrix and the parabola becomes a paraboloid.
However, the minimum still occurs where Ax = b if A is positive definite.
In the n-dimensional case, the energy equation is written as
1 T
EN = x Ax − x T b (11.12)
2
and if A is positive definite, the minimum occurs at
x = A −1b (11.13)
11.4 Least Squares Solution
N ow reverting back to the earlier problem of error minimization where the error quadratic is as given in
Eq. (11.5) and is repeated here for clarity.
⎡ e1 ⎤
⎢e ⎥
f ( x ) = [ e1 e 2 … en ] ⎢ ⎥ = e T e
2
⎢ ⎥
⎢ ⎥
⎣e n ⎦
The error in terms of the state vector is given as
e1 = yact1 − y1 = yact1 − Cx
e 2 = yact2 − y 2 = yact2 − Cx

en = yactn − yn = yactn − Cx
Chapter 11.indd 655 3/17/2009 8:24:21 PM

⎡ e1 ⎤ ⎡ yact1 ⎤ ⎡C ⎤
⎢e ⎥ ⎢ y ⎥ ⎢ ⎥
⎢ 2 ⎥ = ⎢ act2 ⎥ − ⎢C ⎥ x
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣en ⎦ ⎣ yactn ⎦ ⎣C ⎦
E = Yact − x (11.14)
where Yact is the vector of the actual or measured values of the output and x is the state vector. The quadratic
error function f can be written as
f = ETE = (Yact − Cx)T(Yact − Cx)

= (Cx − Yact)T(Cx − Yact)
= xT CT Cx − xT CTYact − YactT Cx + YactTYact
From the above, it is evident that

xT CTYact = (YactT Cx)T
Therefore, the function f can be equivalently written as
f = xT CT Cx − 2YactT Cx + YactTYact = xT CT Cx − 2xT CTYact − YactTYact
From the above, it is evident that f has two components, J1(x) which is a function of x and J2 which is not a
function of x. Thus,
f (x) = J1(x) − J2
where
J1(x) = xT CT Cx − 2xT CTYact and J2 = YactTYact
Evidently, minimizing f with respect to x means that one minimizes J1(x) with respect to x as J2 is not a
function of x. Therefore, it is sufficient to only consider J1(x) for purposes of minimization. Thus,
J1 1 T T
J= = x x − x T T Yact (11.15)
2 2
Therefore, minimizing J implies that f is minimized. Now compare Eq. (11.15) with the energy equation
[Eq. (11.12)]. J and EN are of similar form wherein
x→x
CT C→A
CTYact → b
A minimum for J would exist if CT C is positive definite. Equation (11.13) gives the solution of x such
that EN is minimum. Applying the analogous equivalence as shown above to the solution x = A −1b that
gives the minimum, one obtains
x = ( T )−1 T Yact (11.16)
Chapter 11.indd 656 3/17/2009 8:24:22 PM

Thus, Eq. (11.16) is called the least squares solution of E = Yact − x . This means that the value of the state
vector x as decided by Eq. (11.16) will give the smallest or the minimum square error when used in the
output equation model y = Cx .
Problem 11.1
Consider the following deterministic system:
(1 + b1z–1)yk = (a0 + a1z–1)uk–1
Formulate an estimator to estimate a0, a1 and b1 using the least square algorithm for minimizing the error.
Solution
Problem formulation
yk = a0uk −1 + a1uk − 2 − b1 yk −1
⎡a0 ⎤
yk = [uk −1 uk − 2 ⎢ ⎥
− yk −1 ] ⎢a1 ⎥
⎢⎣b1 ⎥⎦
y = Cx
where
C = [uk −1 uk − 2 − yk −1 ]
⎡a0 ⎤
x = ⎢⎢a1 ⎥⎥
⎢⎣b1 ⎥⎦
Problem 11.2
Consider the system
(1 + b1z–1 + b2z–2)yk = (a0 + a1z–1 + a2z–2)uk–1
Formulate the above system as a least square estimation problem to estimate the parameters a0, a1, a2, b1 and b2.
Solution
yk = a0uk −1 + a1uk − 2 + a2uk −3 − b1 yk −1 − b2 yk − 2
⎡a0 ⎤
⎢a ⎥
⎢ 1⎥
yk = [uk −1 uk − 2 uk −3 − yk −1 − k − 2 ⎢ a2 ⎥
y ]
⎢ ⎥
⎢b1 ⎥
⎢b2 ⎥
⎣ ⎦
y = Cx
Chapter 11.indd 657 3/17/2009 8:24:22 PM

where
C = [uk −1 uk − 2 uk −3 − yk −1 − yk − 2 ]
⎡a0 ⎤
⎢a ⎥
⎢ 1⎥
x = ⎢ a2 ⎥
⎢ ⎥
⎢b1 ⎥
⎢b2 ⎥
⎣ ⎦
11.5 Weighted Least Squares
I t is seen in the previous section that for a set of measured values represented by the vector yact, the best
estimate of the state vector x is given by the least square solution as given in Eq. (11.16). If now fresh
values of the output are measured, then yact vector will get updated. The best estimate of x should again
be recalculated. It can be seen from Eq. (11.16) that the accuracy of the best estimate of x is solely depen-
dent on the accuracy with which the output is measured or observed. This means that x is as good as one
can trust the values of yact. As measurements continue to arrive, the estimate of x is obtained from the
data by a straight line fit. For each new data point there will be a change in the best line. To obtain the
best estimate of x in the face of measurement and observation errors, one must include a trust factor or
weighting factor for the error due to measurements. The trust factor or the weighting factor for a batch of
measurements can be decided qualitatively on the basis of various factors such as sensing methods, tempe-
rature changes during sensing, operator fatigue, etc. Thus, if e is the error vector, then from Eq. (11.4),
one has
⎡ e1 ⎤
⎢e ⎥
e = ( y act − x ) = ⎢ ⎥
2
(11.17)
⎢ ⎥
⎢ ⎥
⎣e n ⎦
Let w1, w2, …, wn be the weighting factors or trust factors for the errors e1, e2, …, en respectively. Then the
weighted errors are represented as
⎡ w1e1 ⎤ ⎡w1 0 … 0 ⎤ ⎡ e1 ⎤
⎢w e ⎥ ⎢ 0 w … 0 ⎥ ⎢e 2 ⎥
⎢ 2 2⎥ = ⎢ 2 ⎥ ⎢ ⎥ = We (11.18)
⎢ ⎥ ⎢0 0 … 0 ⎥⎢ ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣w n e n ⎦ ⎣ 0 0 … w n ⎦ ⎣e n ⎦
where W is a diagonal matrix containing the trust factors as its diagonal elements and e is the error vector.
Pre-multiplying Eq. (11.17) by the weight matrix W, one obtains
We = ( Wy act − Wx ) (11.19)
It is now needed to find a solution to the above equation such that the weighted square error is minimized.
This means that (We)T(We) should be minimized. The least square solution of e = y act − x is as given
Chapter 11.indd 658 3/17/2009 8:24:22 PM

in Eq. (11.16). Comparing Eq. (11.19) with e = y act − x , the least square solution can be obtained by the
following analogous equivalence:
e → We
yact → Wyact
C→W C
eT → eTWT
CT → CTWT
Using the above equivalence and applying them in Eq. (11.16), one obtains
x = ( T W T W)−1 T W T Wy act (11.20)
Equation (11.20) is the weighted least square solution for Eq. (11.19). The term W T W that occurs in the
weighted least square solution of Eq. (11.19) can be written as
⎡w12 0 0 0 ⎤
⎢ ⎥
⎢ 0 w22 0 0 ⎥
W W=⎢
T
(11.21)
⎢ 0 0 … 0 ⎥⎥
⎢ 0 0 0 wn 2 ⎥⎦
⎣
As the measured data arrive on each measurement time interval, the variance also varies according to the
trust. If the measured data are wild, then the variance is large and vice-versa. If wi is set equal to 1/si2 where
si2 is the variance at the ith measurement interval, then the weighting factor gives a measure of the quality
of the observed values. If the variance is large then wi is small and hence that particular observed value is
given less importance. Therefore, variance can be used for deciding the weighting factors or trust factors.
11.6 Recursive Least Squares
A s the measurements arrive and the straight line fit is continuously updated. For each new measurement
or observation of yact, the new best line has to be found out. This means that the best value of vector x
has to be estimated for every new update in the measurement such that the latest solution of x is the least
square solution. This implies that for every new update, the least square solution for x has to be recalculated.
There are two major drawbacks in this approach: (a) the yact vector size continuously increases as more and
more data arrive and (b) the least square solution for x needs to be calculated for every measurement update
which will become more and more time consuming as yact vector length grows.
Therefore to overcome the above drawbacks, recursive techniques are used. In this approach, the newest
best estimate of vector x is dependent on the previous best estimate and the latest measurement. This can be
represented as
x (k + 1) = x (k ) + Correction (11.22)
where x(k + 1) is the estimate of state vector at (k + 1)th iteration; x(k) is the estimate of the state vector at
the kth iteration and the correction is dependent on the newest error due to the latest measurement yact.
At this point it is now appropriate to introduce the following changes in notation to distinguish between
certain variables. Therefore,
Chapter 11.indd 659 3/17/2009 8:24:22 PM

1. xˆ (k ) = xˆk is the estimate of the state vector at kth iteration. The actual state vector at kth iteration in
the physical system is denoted by x (k ) or x k .
2. y (k ) = y k = y act is the measurement or observed value of the output at the kth iteration.
Now the general recursive equation [Eq. (11.22)] can be re-written as
xˆ k +1 = xˆ k + Correctionk (11.23)
The value of Correctionk is proportional to the error due to the latest measurement which will take correc-
tive action on the value of the previous best estimate of the state vector to give the latest best estimate of the
state vector. Thus, the recursive estimate equation becomes
xˆ k +1 = xˆ k + Gek = xˆ k + G ( yk − Cxˆ k ) (11.24)
where ek is the error at the kth iteration. Equation (11.24) is the general recursive estimator equation. G is
called the correction gain. In fact, the various algorithms differ in the way in which G is chosen. There are
many algorithms that use the above recursive structure; however, this section will discuss two of the more
popular algorithms: the least mean square (LMS) algorithm and the Kalman filter algorithm. The LMS
algorithm chooses G in such a way that the minimum mean square error is reached by descending to the
bottom of the parabola with the steepest possible slope. The Kalman filter algorithm chooses G in such a
way that the error in the presence of system noise is minimized. In all the cases, the result is the best estimate
of the state vector that will minimize some cost function J.
LMS Algorithm
The least mean square algorithm is also called LMS algorithm. This is also very popular in adaptive filter-
ing applications such as noise cancellation apart from its application in estimators. In the previous sec-
tions, the optimization was done with respect to the square of the sum of errors or the quadratic of the
error vector, that is, eTe. Here, as the name suggests, the optimization is done with respect to the mean of
the square of sum of errors or mean/expectation of the quadratic of the error vector, that is, E(eTe). Let J
be the function which is equal to the mean/expectation of the error quadratic. Then one has to find a
value of the state vector x̂ such that J is a minimum. Such an x will be the optimal solution or the opti-
mal estimate of the state vector. The function to minimize J is shown in Figure 11.4 as a parabola for the
case of a one-dimensional state vector x. Here the state variable value xmin corresponds to the minimum
value of J, that is, Jmin.
The sequence of finding the state variable value for the one-dimensional case as shown in Figure 11.4 is
illustrated in Figure 11.5. Consider that at an arbitrary iteration k, the state variable is xk and the corre-
sponding cost function value is Jk. The Jk point is shown on the parabola in Figure 11.5(a). At the point Jk
on the parabola, grad Jk or ∇J k is the slope of Jk with respect to x. This slope is used as the correction term
for the next update of the state variable. To approach the minimum of J, one would like to take the opposite
direction to descend quickly or in other words the steepest descent. This is done by subtracting the present
value of the state variable with the present scaled slope of J to obtain the next value of the state variable.
Thus the recursive estimate equation is
xˆk +1 = xˆk + μ( −∇J k ) (11.25)
Equation (11.25) is the algorithm that will iteratively find the best estimate of the state vector such that the
cost function J is a minimum. The negative sign for ∇J k in Eq. (11.25) indicates that the opposite direc-
tion to the slope at the point is taken to provide fastest descent to the minimum.
Chapter 11.indd 660 3/17/2009 8:24:22 PM

J min
0 x min x
Figure 11.4 One-dimensional state vector versus J.
Jk,∇Jk
Jk + 1, ∇Jk + 1
Jk + 1, ∇Jk + 1
∧ ∧ ∧ ∧ x
xk xk + 1 x xk + 2 xk + 1
∧
= xk − m∇Jk = xk + 1 − m∇Jk + 1
(a) (b)
Jk + 2 , ∇Jk + 2
∧ ∧ x ∧ x
xk + 2 xk + 3 xmin
∧ ∧ ∧
= x k + 2 − m ∇Jk + 2 (xk + m ≅ xk + m + 1)
(c) (d)
Figure 11.5 Steepest descent technique for the LMS algorithm.
The variable m in Eq. (11.25) is a positive number called the slope scaling factor. This is also called the learn-
ing rate. For now, let m be appropriately chosen such that J converges. Referring to Figure 11.5(a), the slope
at Jk is negative. For minimizing one should move in the opposite direction. Therefore, a scaled positive
slope value is added to the previous value of the state estimate. The next state variable estimate is always to
the right of the previous state variable estimate if the previous slope of J is negative. x̂k +1 is the new state
variable estimate. The cost function value now is Jk+1 and the slope at this point is ∇J k +1 which is now
Chapter 11.indd 661 3/17/2009 8:24:22 PM

positive. The negative of this positive slope is scaled by m and added to the present state variable estimate to
obtain the next state variable estimate x̂k +2 . The next state variable estimate is always to the left of the pres-
ent state variable estimate if the present slope of J is positive. In this way, after a few iterations as illustrated
in Figure 11.5(d), the cost function will converge to its minimum value Jmin at which point the state variable
estimate is xmin which is the optimal solution. At the minimum, the previous and present state variable esti-
mates will be almost equal when the iteration can be stopped.
It should be noted that the choice of the slope scaling factor or the learning rate m is important to ensure
stability of the minimization of the process (Figure 11.6). If the value of m is higher than a critical value mcri,
then Jk+1 will be higher than Jk and therefore will not converge. This is an unstable situation wherein the
minimum cannot be found. If m is lower than the value of mcri then Jk+1 will be lower than Jk and will eventu-
ally converge to the minimum. This is the stable situation. One should also note that if m is very much lower
than mcri, then the convergence is very slow, that is, the learning rate is slow. Therefore, many versions of the
LMS algorithm exist strictly by the choice of the learning rate. However, as a simple practical approach, one
may start the minimization process with a very low value of m and then tune m by increasing it to improve
the learning rate.
Equation (11.25) is not an easy equation to implement as ∇J k is difficult to calculate in every iteration.
Therefore, it is necessary to obtain an approximation for ∇J k that is an easily implementable measure of the
slope of J at iteration k. For the one-dimensional case, it is known that
J k = ( yk − Cxˆk )
2
∂J k
∇J k =
≈ ( −C )2( yk − Cxˆk )
∂xˆk
where the above relationship is shown as an approximation. This is because even though yk is a measured
value, it is dependent on x̂k through the controller which will affect the measurements. However, in the
above relationship this dependency is ignored. Thus,
∇J k ≈ −2Cek (11.26)
mdiv > mcri > mconv

Unstable
mdiv
Jk mcri Critical
mconv Jk + 1
Stable
∧ ∧
Unstable
Critical
Stable
xk x
∧
xk + 1
Figure 11.6 Choice of slope scaling factor or learning rate m.
Chapter 11.indd 662 3/17/2009 8:24:23 PM


xˆk +1 = xˆk + 2μCek (11.27)
Equation (11.27) is the implementable LMS recursive algorithm for the one-dimensional case. Comparing
with Eq. (11.24), it is evident that the estimator gain G is equal to 2μC for the LMS recursive algorithm.
The LMS algorithm for the one-dimensional case can be extended to the multi-dimensional case along-
similar lines. Thus for the multi-dimensional case,
xˆ k +1 = xˆ k + 2μC T ek (11.28)
where ek is the error at the kth iteration.
Problem 11.3
Consider the following deterministic system,
(1 + b1z–1)yk = (a0 + a1z–1)uk–1
Formulate an estimator to estimate a0, a1 and b1 using the LMS algorithm for minimizing the mean square
error.
Solution
Formulation of the LMS problem:
yk = a0uk −1 + a1uk − 2 − b1 yk −1
⎡a0 ⎤
yk = [uk −1 uk − 2 ⎢ ⎥
− yk −1 ] ⎢a1 ⎥
⎢⎣b1 ⎥⎦
yk = Cx
where
C = [uk −1 uk − 2 − yk −1 ]
and
⎡a0 ⎤
x = ⎢⎢a1 ⎥⎥
⎢⎣b1 ⎥⎦
Here x has to be estimated. The estimate equation is given as
xˆ k +1 = xˆ k + 2μC T ek
where m is the learning rate and ek = yact − Cx̂ k .
The above formulation is simulated in MATLAB and the tracking of the estimate vector x for different
learning rates are plotted in Figures 11.7–11.9.
It can be observed that high learning rate requires lesser number of iterations to estimate the states.
Chapter 11.indd 663 3/17/2009 8:24:23 PM

1.2 0.8
1 0.6
0.8 0.4
0.6 0.2
0.4 0
0.2 −0.2
0 −0.4
−0.2 −0.6
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400 450 500
Plot of paramters for rate = 0.1 Plot of error over time for rate = 0.1
Figure 11.7 Plot of x and e for learning rate m = 0.1 − xT = [0.6 –0.2 0.9].
Kalman Filter
Kalman filter is an optimal estimator. It is also a recursive algorithm that is used for the optimal estimation
of the states. This algorithm gives the optimal state estimates taking into account the noise characteristics of
the system also. Till now, the state space model of the physical systems did not consider the noise of the
system in the state equation and the noise of the measuring or sensing devices in the output equation. If
noise is also considered then the system state space model is written as
x k +1 = Ax k + Bu k + w k (11.29)
y k = Cx k + Du k + v k (11.30)
1.2 1
0.8
1
0.6
0.8
0.4
0.6 0.2
0
0.4
−0.2
0.2
−0.4
0 −0.6
−0.2 −0.8
0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 1400
Plot of a0, a1, b1 for rate = 0.05 Plot of error for rate = 0.05
Chapter 11.indd 664 3/17/2009 8:24:24 PM

1 0.6
0.8 0.4
0.6 0.2
0.4 0
0.2 −0.2
0 −0.4
−0.2 −0.6
−0.4 −0.8
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Plot of error for rate = 0.5 Plot of error for rate = 0.5
where wk is the plant noise and vk is the sensor or the measurement noise. In LMS algorithm, it is seen that
the next state update is given by a one-step time update equation [Eq. (11.28)]. However, in the case of the
Kalman filter, there are two steps involved. First the time update is done and then a measurement update is
performed on the time updated state variable. The recursion steps for the Kalman filter are as follows:
Step 1: Time update
x k +1 = Axˆ k + Bu k (11.31)
where x k +1 is an intermediate state variable estimate which is obtained from the system state equation.
Step 2: Measurement update
xˆ k +1 = x k +1 + Gk +1( y k +1 − Cx k +1 ) (11.32)
where Gk + 1 is the estimator gain or the gain of the Kalman filter.

Observe that the Kalman filter is based on the current estimator concept of a time update followed by a
measurement update. During the time update, the estimate of the kth iteration is used to determine an
intermediate state vector of the (k + 1)th iteration, x k +1 . During the measurement update, the update
action is taken on variables that are all in the same time iteration level. The latest measurement yk + 1 is used
to take correction action to obtain the final estimate x̂ k +1 at the (k + 1)th iteration.
To implement the Kalman filter algorithm, there are a few stochastic terms that need to be described due
to the presence of noise terms in the model. The noises need to be characterized by two measures: (a) the
mean or the expected value and (b) the co-variance.
The noises wk and vk have a mean or average value that is zero. This means that the expected values of
wk and vk are zero. Thus,
E (w k ) = E (v k ) = 0
Chapter 11.indd 665 3/17/2009 8:24:24 PM

The co-variance is the expected or the mean of the quadratic. The co-variance gives a measure of the
energy content in the noise. Thus, E ( w kT w k ) is called the co-variance of noise wk. Likewise, E ( vTk v k ) is
the co-variance of noise vk. In general, any signal is characterized as
signal → {mean, co-variance}
Both the noise measures will have to be found out by repeated measurements on the physical system. For the
system or plant noise wk, the mean is zero and the co-variance is some value Q that is greater than or equal
to zero. Then,
wk → {0, Q } qualifies the plant noise
For the measurement noise vk, the mean is again zero and the co-variance is some value R that is greater than
zero. Then, vk is characterized as
vk → {0, R} which qualifies the sensor or measurement noise, vk
Similarly, the state vector can also be characterized in terms of the mean and the co-variances. However, the
mean value for the state vector is not zero. Let the mean value of the state vector be designated as x k which
is the mean value in a given interval kT and the co-variances are taken about the mean value. Thus, one can
define
Pk = E {( x k − x k )T ( x k − x k )}
which is the co-variance of the state variable value taken about the mean or in other words, it can be called
the error co-variance as it indicates the co-variance of the deviation of the state about the mean value in that
interval. Thus, the state vector can be characterized as
x k → { x k , Pk }
It should be noted that the cross co-variances between wk, vk and the state vector are zero. This means that
wk, vk and the state vector x occur independent of each other. Thus,
E (w T x) = E (w T x) = E (v T x) = 0
With this brief background discussed above in characterizing signals based on means and co-variances, the
Kalman filter algorithm can be understood and implemented. The state vector error co-variance Pk is the
important parameter of concern in this algorithm. Pk is the quadratic of the deviation of the state vector
value about its expected value. Thus Pk is the state error quadratic or the state error co-variance. Here Pk
needs to be minimized. Therefore, the propagation of Pk across each iteration step is critical. Pk should ulti-
mately settle down to a minimum. At the (zero)th iteration, the state vector is characterized as x 0 → { x 0 , P0 }
with some initial value of mean of state and error co-variance. Figure 11.10 illustrates the propagation of the
various variables across two arbitrary iteration steps k and k + 1.
Figure 11.10 illustrates the process of Kalman filtering. The ultimate goal of Kalman filter is to obtain
the next estimate of the state vector such that the error co-variance converges to a minimum. The associated
error correction gain Gk + 1 is called the Kalman filter gain or the Kalman estimator gain. Three iteration steps
are shown in Figure 11.10. At the end of the kth iteration, the state error co-variance Pk and the state
estimate x̂ k are available. First the state error co-variance is updated in time called the time updated value
which is an intermediate variable Pk +1 . Without going into the derivations, as these derivations are out of
the scope of the book, the time updated state error co-variance is given as
Pk +1 = APk AT + Q (11.33)
Chapter 11.indd 666 3/17/2009 8:24:24 PM

k th iteration (k + 1)th iteration (k + 2)th iteration
Measurement
Time update ~ Pk + 1
Pk Pk + 1 update
∧ Time update ~ ∧
Xk Xk + 1 Gk + 1 Xk + 1
Measurement
update
Kalman estimator gain
(or Kalman filter gain)
Figure 11.10 Time update and measurement update in Kalman filter algorithm.
where A is the system matrix as given in the system state equation and Q is the system or plant noise co-variance
as described earlier.
The measurement noise will also affect the state error. The measurement noise co-variance R will be used
to update the time updated state error co-variance. This is called the measurement update for the state error
co-variance which will result in the state error co-variance value in the (k + 1)th iteration. Thus the measure-
ment update of the state error co-variance is given by
Pk +1 = Pk +1 − Pk +1C T (CPk +1C T + R)−1 CPk +1 (11.34)
Note that the measurement update is dependent on only Pk +1 , C and R . The state error co-variance of the
(k + 1)th iteration as given in Eq. (11.34) is used for calculating the (k + 1)th iteration Kalman filter gain
Gk+1, which is given as
Gk +1 = Pk +1C T R −1 (11.35a)
The time update for the state vector using the state vector estimate of the kth iteration is given in Eq. (11.31)
which is repeated here for clarity:
x k +1 = Axˆ k + Bu k
where x k +1 is a intermediate time updated state vector estimate which is obtained from the system state equa-
tion. Now using the latest measurements obtained in the (k + 1)th iteration, the latest correction can be applied
to the time updated state vector estimate to obtain the state vector estimate of the (k + 1)th iteration. The
measurement update of the state vector estimate is given by Eq. (11.32) which is repeated here for clarity:
xˆ k +1 = x k +1 + Gk +1( y k +1 − Cx k +1 )
It should be noted that Eq. (11.35a) is not the only equation for Kalman filter gain. There are other variants
also.
Steady-State Kalman Fitler

Under steady-state conditions, Pk +1 = Pk +1 = Pk = Pk = P. Therefore, the Kalman gain from Eq. (11.35a) is
given as
Chapter 11.indd 667 3/17/2009 8:24:25 PM

G = PC T R −1 (11.35b)
The gain obtained from Eq. (11.35b) is called the steady-state Kalman gain. This will, however, give sub-
optimal estimator performance.
The implementation algorithm for the Kalman filter is discussed next.
Choose Po = I and xo = 0 and evaluate the estimator equations in the following sequence:
1. Pk +1 = APk A T + Q
2. Pk +1 = Pk +1 − Pk +1C T (CPk +1C T + R)−1 CPk +1
3. Gk +1 = Pk +1C T R −1
4. x k +1 = Axˆ k + Bu k
5. xˆ k +1 = Ax k +1 + Gk +1( y k +1 − Cx k +1 )
11.7 Optimal Control: Linear Quadratic
R eferring to the system block diagram as shown in Figure 11.11, the optimal estimation has been the
focus of discussion until now. The principle of minimum energy is used to obtain the optimal estima-
tor gain G such that a cost function based on the square of the error quadratic is minimized. This very same
principle of minimum energy can be used to obtain the optimal controller gain K such that a specific cost
function is minimized. The previous chapter discusses the concepts and principles of the regulator problem
and the tracker problem including the design of controllers and estimators for such structures. Here the
algorithm for obtaining the optimal controller gain K will be addressed.
System or plant Replace integrator by z −1

for discrete block diagram
(Ax + Bu)
r u Bu x y
+ K B + ∫ C
− +
Controller
r = 0 for regulator Ax
gain
r ≠ 0 for tracker ∧
A
x
Estimator
gain
∧
x e +
Estimator G
−
C ∧
Cx
Figure 11.11 Controller and estimator for a given system using state space control methods.
Chapter 11.indd 668 3/17/2009 8:24:25 PM

In the case of the regulator problem where the reference input is zero, all the states will ultimately settle
to zero in full-state feedback configuration. This means that the input which is proportional to the state
values will also settle down to zero in the steady state. On the occurrence of a disturbance one would like the
states to quickly settle down back to zero. One could minimize the area under the square of the state devia-
tion from zero and/or the area under the square of the input deviation from zero. This is illustrated in Figure
11.12. Figure 11.12 shows the one-dimensional case for x and u. The time evolution of x and u on the
occurrence of a disturbance is shown. Ultimately x and u will settle to zero. The time evolutions of x2 and u2
are also shown. The area under the x2 and u2 curves should be minimized so that the response will match the
reference as quickly as possible. A possible candidate cost function J can be taken as the mean of the area
under the x2 and u2 curves. Thus,
area u + area x
J=
2
Referring again to Figure 11.12, and substituting for the areau and areax , one obtains
∞
1
2 ∫0
J= (u 2 +x 2 )dt
The above cost function does not give much design flexibility as the input quadratic and the state quadratic
are not weighted. One may weigh the input quadratic by an appropriate weight R and the state quadratic by
another appropriate weight Q. Then the modified cost function is given as
∞
1
2 ∫0
J= ( Ru 2 + Q x 2 )dt
Q and R can be used as the design parameters in the minimization process to tune the value of the con-
troller gain K. By adjusting the values of Q and R, one can obtain the controller gain values such that the
response is faster or slower.
For the multi-dimensional case where x and u are vectors, then Q and R become weighting matrices,
respectively. Thus, for the general multi-dimensional system, the cost function is given as
x x2 ∫x 2dt = areax
0 t 0 t
∫u 2dt = areau
u u2
0 t 0 t
areau + areax
Minimize
2
J (Cost function or perfomance index)
Figure 11.12 Cost function for optimal control.
Chapter 11.indd 669 3/17/2009 8:24:26 PM

∞
1
2 ∫0
J= (u T Ru +x T Qx )dt (11.36)
where Q is a diagonal matrix containing the weighting factors for each state and R is a diagonal matrix con-
taining the weighting factors for each input. (Note that the Q and R used here are different from the noise
co-variance matrices Q and R used in the Kalman estimator.) J is a quadratic of both the input vector and
the state vector.
The cost function J in Eq. (11.36) consists of two quadratic forms. One is the quadratic of the state
vector and the other is the quadratic of the input vector. The objective is to design or select the controller
gains K such that the minimum sufficient input u is applied to obtain the specific performance objective.
This means that the controller gains should be selected such that the input quadratic or the input energy is
minimized. One can add one more constraint by including the minimization of the state energy. This would
imply that the controller gains be selected so that the input energy as well as the state energy is minimized.
Then such a choice of gain K is called the optimal controller. The states are in general controllable from the
input. In such a case, minimizing only the input energy is sufficient. Therefore, Q can be either zero or
positive, that is, Q should be positive semi-definite. However, R should be positive definite for a minimum
to exist.
It should be noted that the feedback gains are selected to minimize J that is given in Eq. (11.36). This
means that the integrand of Eq. (11.36) must vanish as t → ∞ . Thus with a suitable choice of Q, the state
x(t) will go to zero (which is the reference value for regulator action) with time and likewise for a suitable
choice of R, the input u(t) will also go to zero for regulator action.
Linear Quadratic Regulator

Let the plant or system by defined by the following state and output equations:
x = Ax + Bu (11.37a)
y = Cx (11.37b)
with x(t) as the n × 1 state vector, u(t) as the m × 1 control input and y(t) as the p × 1 measured output.
Referring to the previous chapter, the control law for full-state feedback is of the form
u = −Kx (11.38a)
where K is an m × n matrix of constant coefficients called the controller gains. For the regulation action, if
the controller K functions well for one specific constant reference, then it is as good for any other constant
references. Therefore, for analysis and synthesis the reference is set to zero for regulator action without loss
of generality. If one desires to feedback only the measurable output, then the control law for output feed-
back is of the form
u = −Ky (11.38b)
where K is an m × p matrix of constant feedback coefficients called the controller gains that is to be deter-
mined. Since the control input vector u(t) is only allowed to depend on the output vector y(t) and not on
the entire state vector x(t), this control law reflects the fact that only measurable quantities may be used to
determine u(t). This implies that in output feedback, one is dealing with reduced state information.
The performance objective of state regulation for the system may be attained by selecting the control input
u(t) to minimize a function J called the cost function or the performance index as given in Eq. (11.36).
Chapter 11.indd 670 3/17/2009 8:24:26 PM

Thus, for the linear quadratic regulator (LQR) problem, one can say that given a linear system as
described by the model of Eqs. (11.37a) and (11.37b), the controller matrix K must be selected or designed
for a regulator control input law as given in Eq. (11.38a) or Eq. (11.38b) such that the value of the per-
formance index J as given in Eq. (11.36) is minimized. Such a controller is called the linear quadratic
regulator.
By substituting the regulator control law of Eqs. (11.38a) and (11.38b) into the state equation [Eq.
(11.37)], the closed-loop system is given by
x = (A − BK)x = A c x (for full-state feedback) (11.39a)
x = (A − BKC)x = A c x (for output feedback) (11.39b)

For any admissible choice of the performance index weighting matrices Q and R, closed-loop stability is
guaranteed by this approach. Thus, Q and R may be treated as design parameters which are interactively
varied until a suitable control gain K is obtained. One should contrast this with the approach used in classi-
cal techniques which might involve directly varying the elements of the control gain K in an attempt to
obtain suitable closed-loop performance. In the classical approach, closed-loop stability is not even guaran-
teed for all values of K. Therefore it would be far more difficult to achieve an acceptable design. But here the
design for tuning the gains in K for good performance where closed-loop stability may not even hold has
been replaced by the problem of tuning the Q and R weighting matrices where closed-loop stability is at
least guaranteed. Generally, the starting value of Q is chosen as CTC and R is chosen as identity matrix. The
above discussions are valid even in the discrete domain. Without going into the derivations, the discrete
LQR design algorithm is given in the next sub-section.
Linear Quadratic Regulator: Full-State Feedback

Let the given system or plant be expressed in the discrete domain as
x k +1 = Ax k + Bu k
y k = Cx k + Du k (11.40)
For a full-state feedback control structure, the control law is given by

u k = −Kx k (11.41)
The performance index or the cost function that is to be minimized is given as
1 ∞ T
J= ∑ (u k Ru k + x kT Qx k )
2 k =0
(11.42)
Note that the integrator is replaced by the discrete accumulator in the above cost function equation.
Equations (11.36) and (11.42) are equivalent.
To find the controller gain matrix K that will minimize the cost function J as given in Eq. (11.42), the
optimal solution is given by
K = (R + B T PB )−1 B T PA (11.43)
where P is an intermediate matrix variable which is obtained from the solution of
A T PA − P + Q − A T PB(R + B T PB )−1 B T PA = 0 (11.44)
Chapter 11.indd 671 3/17/2009 8:24:26 PM

Equation (11.44) is popularly known as the Discrete Algebraic Riccati Equation (DARE). The solution
exists if P is positive definite. It may be cumbersome to find the solution to Eq. (11.44) manually. Therefore,
one may have to use the help of software packages like MATLAB or SciLAB for solving Eq. (11.44). If
DARE is a function that solves the DARE, one can find the P matrix using the command.
P = DARE(A,B,Q,R)
Equation (11.43) gives the LQR gain, K which is called the LQR gain and sometimes also called the Kalman
controller gain.
Linear Quadratic Gaussian

If the controller is optimal and the estimator is a Kalman filter, then the combined system is known as linear
quadratic Gaussian (LQG) system. In the LQG system, the estimation is optimal for Gaussian noise charac-
ter. Here stability is always guaranteed with the optimal performance. Such a system is robust in perfor-
mance also. In LQG scheme, both K and G are designed independently. K can be designed by LQR
algorithm and G can be designed by Kalman filter approach. This approach has the advantages of
1. better noise rejection;
2. optimal control input;
3. optimal state estimate;
4. guaranteed stability.
Linear Quadratic Tracker: Output Feedback

The output feedback formulation was discussed in the previous chapter. This section discusses the perfor-
mance index for the output feedback formulation given in Eqs. (10.101)–(10.103), reproduced here for easy
reference:
xk +1 = Axk + Buk + Erk
yk = Cxk + Frk
z k = Hxk
uk = − Kyk = − KCxk − KFrk
xk +1 = ( A − BKC )xk + ( E − BKF )rk = Ac xk + Bc rk
The input deviation is given by

ûk = uk − uk
uˆk = − KCxk − KFrk − ( − KC x − KF r ) = − KCxˆk (11.45)
where rk = r = r for a step reference. In the steady state, xk +1 = xk = x ∞ , where x ∞ is the state under steady-
state operating conditions given by
x ∞ = ( I − Ac )−1 Bc r∞ (11.45a)
Here r∞ is the reference input under steady-state operating conditions.
The steady-state error in response to a step of magnitude r is obtained by using the steady-state relationship
of Eq. (11.45a). It is given by
Chapter 11.indd 672 3/17/2009 8:24:26 PM

e = r − z = r − H x = [1 + H ( Ac − I )−1 Bc ] (11.46)
The error deviation is given by
eˆk = ek − e = (r − Hxk ) − (r − H x ) = − Hxˆk = − zˆ (11.47)
where
zˆk = z k − z = Hxk − H x = Hxˆk .
To make the tracking error, ek = eˆk + e small, one may have to weight both êk and e in a quadratic perfor-
mance index. Thus the following performance index is used:
1 ∞ T 1 1
J= ∑ (eˆk ⋅ eˆk + uˆkT ⋅ R ⋅ uˆk ) + 2 e T ⋅V ⋅ e + 2 ∑ ∑ ( g ij ⋅ kij2 )
2 k =0
(11.48)
i j
Using Eq. (11.47), one can rewrite the performance index of Eq. (11.48) in terms of the state deviations as
1 ∞ T 1 1
J= ∑ ( xˆk ⋅Q ⋅ xˆk + uˆkT ⋅ R ⋅ uˆk ) + 2 e T ⋅V ⋅ e + 2 ∑ ∑ ( g ij ⋅ kij2 )
2 k =0
(11.49)
i j
where Q = qH T H , R > 0 and V ≥ 0 which is the steady-state error weighting matrix. The scalar weight q
gives greater design flexibility in the choice of Q. It is generally acceptable to select R = rI and V = mI, where
r and m are scalar design parameters. If the dynamic compensator in each loop contains a discrete integrator
or a pole at z = 1, then the steady-state error will automatically be zero. In such cases, one can select V = 0
in the performance index function of Eq. (11.49). The usefulness of the above performance index is that
steady-state errors can be guaranteed to be small even when some of the loops do not have integrators or
poles at z = 1. The last term in the performance index weights the elements kij of the control gain matrix K.
The motivation is to allow for more structure in the control system wherein experience and intuition can be
used to suitably weight the elements of the control gain matrix K. Thus, if the error component of one loop
does not influence the control input of another loop, the appropriate element of K may be made as small as
desired by selecting its weight gij as a large value in the performance index. The first term in the performance
index is the weighted quadratic state and input deviations which should become zero as time tends to
infinity.
The question of concern to the control system designer is the selection of the weighting matrices Q and
R. This will be discussed in the following section while illustrating the induction motor example. One must
appreciate the fact that minimization of the performance index function as given in Eq. (11.49) may not
always be the true design objective. The problem, however, is that the true design objective often cannot be
expressed in mathematical terms and even when the design objective is expressible in mathematical terms, it
may be impossible to solve for the optimum control law.
Optimal Gain for Output Feedback

The output feedback gains are obtained by solving a set of coupled equations. This section will focus on the
design algorithm without going into the derivations for these coupled equations.
Design Equations
ACT PAC − P + Q + C T K T RKC = 0 (11.50)
AC SACT − S + X = 0 (11.51)
Chapter 11.indd 673 3/17/2009 8:24:26 PM

RKCSC T − B T PAC SC T + B T ( AC − I )− T ( P + H TVH )x ⋅ y T

(11.52)
− B T ( AC − I )− T H TVry T + g * K = 0
where rk is a step of magnitude r and g*K is a matrix with elements gijkij;
x = −( AC − I )−1 BC ; y = C x + Fr ; X = x ⋅xT
AC = A − BKC
BC = E − BKF
P is a positive definite constant matrix, that is, P > 0.
Performance Index Constraint The performance index given in Eq. (11.49) is the cost that needs to be
minimized. This performance index can be shown to be equivalent to the following function [given in Eq.
(11.53)]. The first term of Eq. (11.49) is replaced with an easily implementable equivalent given as the trace
of PX as indicated in Eq. (11.53).
1 1 1
J = tr( PX ) + e T ⋅V ⋅ e + ∑ ∑ ( g ij ⋅ kij2 ) (11.53)
2 2 2 i j
where e = [ I + H ( AC − I )−1 BC ]r
One should note that the optimal feedback gain K is obtained by solving the set of coupled equations
[Eqs. (11.50)–(11.52)] by minimizing the performance index given in Eq. (11.53). This minimization
problem may be solved by using either gradient-based algorithms or non-gradient-based algorithms like
the SIMPLEX algorithm. Many popular software packages provide subroutines for minimization algo-
rithms. If the SIMPLEX algorithm is used, then the optimal feedback gain K can be obtained by using
only Eq. (11.50) which is also called the Lyapunov equation. However, for gradient-based routines which
are faster than the SIMPLEX approach, all three equations [Eqs. (11.50)–(11.52)] are required. If the
system is controllable and observable, and the Q and R matrices are appropriately chosen, the minimiza-
tion of J will not fail to find a K which stabilizes the closed-loop system. This means that once the system
along with Q and R is specified, determination of K is a problem in numerical analysis. This does not
mean that minimizing J is a trivial problem. There are many standard algorithms used to find the global
minimum. Two of the standard methods used that are popular for finding the global minimum are:
1. Find the local minimum starting from widely varying starting values of the independent variables and
then pick the minimum among these.
2. Perturb a local minimum by taking a finite amplitude step away from it and then see if the routine
returns to a better point or always to the same one.
11.8 Induction Motor Example
T he induction motor has been formulated as an LQ tracker problem with output feedback in Section
10.14. The objective now is to determine the gain matrix K for the controller structure (which is a
proportional–integral structure in this case) so as to minimize a specified performance index. From the per-
formance index J, as given in Eq. (11.53), one can set V = 0 in this specific case, because the steady-state
error is zero due to the use of the proportional–integral structure which has a pole at z = 1.
From the gain matrix given in Eq. (10.116) (Chapter 10), it is clearly seen that some of the terms in the
K matrix are zero. Therefore, in the performance index function of Eq. (11.53), the corresponding matrix
Chapter 11.indd 674 3/17/2009 8:24:27 PM

elements kij can be weighted with a very large value of gij , so that after minimization the resulting gain values
of these kij elements will be very small and can be set to zero during implementation. This weighting of the
elements of K has to be done for minimization algorithms using the gradient approach. For the SIMPLEX
approach, those matrix elements which are zero can be set to zero and the remaining matrix elements can be
optimally obtained by minimization of the performance index given by
1
J = tr( PX ) (11.54)
2
which is obtained from Eq. (11.53). As a consequence, it would imply that with SIMPLEX approach, one
can fix any gain element of the gain matrix K and obtain the optimal gain values for the rest of the elements
of the gain matrix K by minimizing the performance index J, as given in Eq. (11.54). This in fact gives lots
of flexibility in arriving at stable optimal gains for the proportional–integral control structure of the
system.
The steps for obtaining the optimal gain values of the gain matrix K, once the system is formulated, as
given in Eqs. (10.101a)–(10.102), is as follows:
Step 1: The optimization process is an iterative process. Therefore, to start the optimization process, an
initial gain matrix K, which makes the closed-loop plant (i.e., A − BKC ) stable, has to be used. Such a
gain may or may not be easy to find. One reliable though tedious way to find a stabilizing gain is to use
discrete root locus techniques. The gain can then be optimized using the optimization of the perfor-
mance index J. However, as the induction motor is open-loop stable which is established from the
eigenvalues of the induction motor system matrix, the compensator and the induction motor system
combination is consequently open-loop stable. Therefore, all the open-loop poles lie within the unit
circle. If the gains are very small, then the closed-loop poles are very close to the open-loop poles and
hence, the closed-loop system is stable. As a consequence, the proportional gains can be set to zero and
very small values of integral gains (around 0.01) can be used. This will usually provide the starting
eigenvalues of A − BKC, such that they are within the unit circle.
Step 2: The state weighting matrix Q and the input weighting matrix R need to be chosen. From Eq.
(11.49) one obtains the state weighting matrix Q, which is equal to H TH. However, to tune the
response, a scalar weight q is used such that
Q = q ⋅ HT H
Regarding the input weighting matrix R, it is usual to choose R = rI, where r is a scalar. If a large value
of r is used, the resulting gain K leads to small inputs to the system, thereby slowing down the system
response. If one needs to have a slower or faster response of the system with respect to some input, then
the corresponding element in the R matrix can be suitably weighted higher or lower, respectively.
Step 3: The performance index J in Eq. (11.54) is minimized under the constraint of the Lyapunov
equation [Eq. (11.50)] using the SIMPLEX algorithm.
In the case of the induction motor, it is evident from Eq. (10.104) that the system matrix is dependent on
the rotor speed wm. As a consequence, one would obtain different optimal gains at different rotor speeds.
The plot of the Kpd , Kid , K pq and Kiq versus the rotor speed wm is shown in Figures 11.13(a) and (b). The
gains are calculated for eight different speeds as shown in Figure 11.13. For other speed values, the opti-
mum values of the gains are obtained by interpolation. A lookup table can be used in the digital controller
of the induction motor drive system such that the appropriate optimal gain values are used at a specific
speed. On the other hand, one may also use a fixed gain matrix K corresponding to the maximum rotor
Chapter 11.indd 675 3/17/2009 8:24:27 PM

0
Proportional gains, K pd and K pq
K pq
−0.5
K pd
−1
−1.5
0 20 40 60 80 100 120 140 160
wm (rad/s)
(a)
0
Integral gains, K id and K iq
−50 K iq
K id
−100
−150
0 20 40 60 80 100 120 140 160
wm (rad/s)
(b)
Figure 11.13 (a) Optimal proportional gains as a function of shaft speed; (b) optimal integral gains
as a function of shaft speed.
speed. This would lead to sub-optimal operation at other speed values. But nonetheless, stability is guaran-
teed at all speed ranges. The induction motor used for controller design has the parameters indicated in
Appendix V.
As lots of analytical and experimental experience is available on the induction motor, the designer should
put the intuition and experience to advantage while using the output-feedback approach. In fact, the pro-
portional–integral controller structure that was proposed while formulating the induction motor drive
system as an LQ tracker problem is based on the experience of induction motor operation in constant
Chapter 11.indd 676 3/17/2009 8:24:27 PM

flux operation schemes such as the V/f controls. It is evident that in this context, the output feedback is
appropriate and in view of the highly coupled dynamics of the current controllers, multi-variable design
techniques discussed till now become more or less necessary.
It is discussed in the previous chapter (sub-section “Vector Control” of Section 10.12) that good dynamic
performance is obtained only under field orientation. With the orientation of the arbitrary axis along the
rotor field and the use of the decoupling network, the induction motor plant as seen by each current control
loop reduces to a first-order system. However, when the rotor field is disoriented, then the induction motor
is a fourth-order system wherein the dynamic performance deteriorates compared to dynamic performance
under rotor field orientation. This is explicitly indicated in Figure 11.14 for the isd and isq currents.
The performance index J in Eq. (11.53) is also very flexible, which is conducive for intuitive manipula-
tion of the weights of the gains gij. The following strategy has been found useful in the context of the induc-
tion motor for arriving at the optimal gains.
1.5
(b)
1
isd
(a)
1.5
0
0 50 100 150
(a)
1.5
(b)
1
isq
(a)
1.5
0
0 50 100 150
No. of samples
1 sample = 1 ms
(b)
Figure 11.14 Step response for isd and isq: (a) Arbitrary axis is not aligned along the rotor field;
(b) arbitrary axis is aligned along the rotor field, that is, rotor field-orientated control.
(Q = 0.1HTH, R = diag[1, 20], Kgains = [–0.3, –62.1088, –0.3, –48.572].)
Chapter 11.indd 677 3/17/2009 8:24:28 PM

1. Start with the initial gain values of Kgains = [Kpd , Kid , Kpq , Kiq ] = [0, –0.01, 0, –0.01].
2. Set Q = HTH and R = diag(1,1) where the first element of the diagonal of R corresponds to the direct axis (d-
axis) input and the second element of the diagonal of R corresponds to the quadrature axis (q-axis) input.
3. Obtain the optimal gain matrix K by solving Eq. (11.50) (by SIMPLEX minimization method) under
the constraint of Eq. (11.53).
4. Check the step response of the closed-loop system.
5. If the step response is not proper, change the input weighting matrix R and the state weighting matrix
Q and obtain K till proper response is obtained.
6. If one has to slow down the system, decrease Q and/or increase R. If one has to speed up the system
then increase Q and/or decrease R, as shown in Figure 11.15.
2
a
b
c
1.5 d
e
i sd
1
f
0
0 50 100 150
(a)
1.5
a
b
c
d e
1
i sd
0.5 f
0
0 50 100 150
No. of samples
1 sample = 1 ms
(b)
Figure 11.15 Step response of isd and isq under rotor field-oriented control where a, b, c, d, e, f
responses are for the following gains, Q and R. Kgains = [Kpd Kid Kpq Kiq]; Q; R. (a) [–0.9404,
–146.1699, –0.4600, –99.7845]; Q = 10H TH; R = diag(0.1, 2).(b) [–0.7785, –141.9508,
–0.5872, –89.7422]; Q = H TH; R = diag(0.1, 2). (c) [–0.5457, –124.2175, –0.6775, –77.3730];
Q = 0.1H TH; R = diag(0.1, 2). (d) [–0.4231, –62.1088*, –0.7961, –47.9323]; Q = 0.1H TH;
R = diag(1, 20). (e) [–0.3*, –62.1088*, –0.3*, –48.5720]; Q = 0.1H TH; R = diag(1, 20).
( f ) [–0.0001*, –5*, –0.0001*, –15.6961]; Q = H TH; R = diag(1, 20).
Note: *These gains were fixed during optimization (constrained optimization).
Chapter 11.indd 678 3/17/2009 8:24:28 PM

7. If one has to speed up or slow down the system controller to a large extent, one could adopt the follow-
ing method which allows more structure to the control by way of increased intuitive interaction by the
designer. For example, if one has to slow down the response, fix the proportional gain values Kpd and Kpq
to a small value. Now set the input weight matrix R such that R = diag(rwd, 1) where rwd is a very large
value. While optimizing, if gradient algorithms are used, weight the Kpd and Kpq gains with large weights. If
the simplex algorithm is used, just fix the values of Kpd and Kpq to low values. After optimization, the
optimal Kid value obtained will be a small value. Now set R = diag(1, rwq ) where rwq is a large weight on
the q-input. Fix the Kid gain to the value obtained during the previous optimizing run. Now the gain
value Kiq obtained will also be low. Kid and Kiq can be iteratively fixed and optimized to obtain any
degree of slowness of speed of response.
Referring to Figure 11.15, it is evident that as Q increases, the response becomes faster. One should
also note that the speed of response of isd and isq need not be identical. This is because under rotor field-
oriented control, the isd and isq components are decoupled. From Figure 11.15, it is evident that response
(e) is preferable from the point of view of performance criteria. However, it will be shown in the next sec-
tion that response ( f ) is most appropriate because the controller gains corresponding to the response ( f )
give the best stability robustness compared to the other gain values. More importantly, under rotor field-
oriented control, upto base speed, the isd component is kept constant and the torque dynamics is deter-
mined only by the isq component. Therefore, it is sufficient that isq has good dynamic performance. In
the next section, it will be shown that the gains corresponding to response (e) can be used if the rotor
resistance adaptation is performed to improve stability robustness. In the following section, robustness
concepts are considered along with the design methodology using the induction motor example for illus-
tration.
11.9 Robust Controller Design
I n the previous section it is assumed that an exact state-variable description of the plant to be controlled is
available. However, in a practical situation, the actual plant and the model plant will never be identical.
Some of the parameters of the system matrix may be different from that of the actual system. In the case of
the induction motor example, parameters like the rotor resistance, the stator resistance or the magnetizing
inductance of the actual motor may be at variance from the corresponding parameters of the motor model,
which could lead to deterioration in performance and stability. The design of the current controllers con-
sidered in the context of robust controller design for induction motor accommodates for the uncertainty in
the system parameters.
Further, it is often important to account for disturbances and sensor measurement noises. Disturbances
may act to cause unsatisfactory performance in a system. Therefore, it is important to design controllers that
have stability robustness and performance robustness. Stability robustness is the ability to guarantee closed-
loop stability in spite of parameter variations. Performance robustness is the ability to guarantee acceptable
performance even though the system may be subject to disturbances.
Consider the standard closed-loop system shown in Figure 11.16. A few signals have been added to
characterize the uncertainties. The signal d(t) represents a disturbance acting on the system and n(t) repre-
sents the sensor or measurement noise. If S(z) is the system sensitivity and T(z) is the system co-sensitivity,
then referring to Figure 11.16, they are defined as
S ( z ) = ( I + Gpa Gc )−1 (11.55)
Chapter 11.indd 679 3/17/2009 8:24:28 PM

Disturbance
d (t )
+
r (t ) e(t ) − n (t ) u (t ) z (t )
Gc Gpa
+ +
− Compensator Plant
n (t )
e(t ) = r (t ) − z (t )
Measurement/sensor noise
Figure 11.16 Standard closed-loop system with disturbance input and measurement noise.
T ( z ) = ( I + GpaGc )−1 ⋅ GpaGc = GpaGc ( I + GpaGc )−1 (11.56)

As S(z) + T(z) = I, T(z) is called the complementary sensitivity or the co-sensitivity.
Referring to Figure 11.16, it can be shown that the performance output z(t) and the tracking error e(t)
are given in the z-domain by
Z(z) = T(z)[R(z) − N(z)] + S(z)D(z) (11.57)
E(z) = S(z)[R(z) − D(z)] + T(z)N(z) (11.58)
The disturbance d(t) occurs usually at low frequencies and the measurement noise n(t) is predominant at
high frequencies. On examining Eqs. (11.57) and (11.58), it is observed that at low frequencies where the
disturbances are predominant, the sensitivity S must be small, and at high frequencies where the measure-
ment noises are predominant, the co-sensitivity T must be small. To obtain a measure of the “smallness” of
the sensitivity S and the co-sensitivity T, multi-variable Bode plots are used.
The multi-variable Bode magnitude plot is nothing but the plot of the transfer function singular values
versus the frequency. MIMO generalizations of the Bode magnitude and phase plots have been reported in
the literature. However, the theory relating to the phase portion of the plot is more difficult to use in a prac-
tical design technique. In the case of robust control design for a discrete system, the magnitude plot is often
sufficient. If H(jw) is a square transfer function matrix (which is true in our case) at any frequency w, then
for any given input, the magnitude of the transfer function H(jw) at any given frequency w is bounded above by
its maximum singular value, denoted by σ (H(jw)) and bounded below by its minimum singular value, denoted
by σ (H(jw)). Therefore for design of the robust controller, one needs to consider only these two constraining
values of the magnitude.
When performing algebra with singular values, the following relations with respect to singular values are
useful:
For any matrices A, B, GK and non-singular matrix M.
σ (GK ) − 1 ≤ σ ( I + GK ) ≤ σ (GK ) + 1 (11.59)
Chapter 11.indd 680 3/17/2009 8:24:28 PM

1
σ(M ) = (11.60)
σ ( M −1 )
σ ( AB ) ≤ σ ( A )σ (B ) (11.61)
Frequency Domain Performance Specifications

It is desired that S(jw) be small at low frequencies and T(jw) be small at high frequencies. From Eq. (11.55)
and using relation in Eq. (11.60), one obtains
σ (S ) = σ[(1 + GpaGc )−1 ] = 1 / σ ( I + GpaGc ) ≈ 1 / σ (GpaGc ) (11.62)
From Eq. (11.56) and using relation in Eq. (11.60), one obtains
σ (T ) = σ[GpaGc ( I + GpaGc )−1 ] ≈ σ (GpaGc ) (11.63)
From Eq. (11.62) it is evident that a large value of σ (GpaGc) guarantees a small value of σ (S ) . From Eq.
(11.63), it is evident that a small value of σ (GpaGc ) will guarantee a small value of σ (T ) . Therefore, the
specifications that S(jw) should be small at low frequencies and T(jw) should be small at high frequencies
may well be formulated in terms of the minimum singular value, σ (GpaGc) being large at low frequencies and
the maximum singular value σ (GpaGc ) being small at high frequencies. Thus, all performance specifications
can be in terms of the minimum and maximum singular values of the loop gain, GpaGc. The practical signi-
ficance is that one needs to examine only the singular values and not S(jw) and T(jw). Further, software
packages give standard routines to compute the singular values of any transfer function. This makes the
robust controller design much simpler.
The sensor or measurement noise is generally appreciable at high frequencies. Therefore, to keep the
tracking error small in the face of the measurement noise, the co-sensitivity, T(jw) should be small at high
frequencies above a certain frequency, say wn. This may be guaranteed if
σ (GpaGc ) 1, for ω ≥ ωn (11.64)
Modeling Uncertainties
To guarantee stability robustness in the face of plant modeling uncertainty, the uncertainties may be
included into the model in one of the following two ways: (a) additive uncertainties or (b) multiplicative
uncertainties.
*
The actual plant model Gpa and the assumed plant model Gpa may differ by additive uncertainties such
that
*
Gpa = Gpa + ΔGpa (11.65)
where the singular value of the additive uncertainty satisfies a known bound
σ (ΔGpa ) < a(ω) (11.66)
*
with a(w) known for all w. On the other hand, the actual plant model Gpa and the assumed plant model
Gpa may differ by multiplicative uncertainties, such that
*
Gpa = [ I + M ( jω)]Gpa (11.67)
where the singular value of the multiplicative uncertainty satisfies a known bound
σ[ M ( jω)] < m(ω) (11.68)
Chapter 11.indd 681 3/17/2009 8:24:29 PM

with m(w) known for all w. Since Eq. (11.65) can be written as
−1
*
Gpa = [ I + ΔGpa Gpa ]Gpa = [1 + M ( jω)]Gpa (11.69)
one can confine the discussion to only multiplicative uncertainties. For stability robustness, the co-sensitivity
T( jw) should be bounded above by the reciprocal of the multiplicative modeling discrepancy bound m(w)
which is given by
σ[GpaGc ( I + GpaGc )−1 ] = σ (T ) < 1 / m(ω) (11.70)
If the worst case uncertainties in the system parameters are known, then the upper bound m(w) can be
found from Eq. (11.67) as follows:
−1
M ( jω) = (Gpa
*
− Gpa )Gpa (11.71)
m(ω) = σ[ M ( jω)] (11.72)
The robustness bound 1/m(w) and the maximum singular values of T(jw) for the system structure of
Figure 11.16 are plotted for an induction motor plant by using a computer program (Figure 11.17).
In Figure 11.17, the robustness bound 1/m(w), for rotor resistance, stator resistance and magnetizing
inductance parameter uncertainties (from 33% to 99%) are shown. In the case of the rotor and stator resis-
tance uncertainties as the actual parameters increase from the nominal values (which is the cold resistance
value used in the assumed model), the upper bound for T(jw) decreases. Though the magnetizing induc-
tance parameter uncertainty is shown upto 100% deviation from the nominal, in practice, the uncertainty
in the magnetizing inductance parameter will not be more than 10–20%. One can observe from Figure
11.17 that uncertainties in the magnetizing inductance will also lead to lowering the permissible system
bandwidth.
Controller Design Steps

The algorithm for robust controller design for the induction motor drive system which is formulated as an
LQ tracker problem with output feedback is as follows.
Step 1: Augment the entire drive system as an LQ tracker problem as discussed in Section 10.14 in the
previous chapter.
Step 2: Proper choices for Q and R matrices are made and the optimal gain matrix K is obtained using
either SIMPLEX minimization or gradient descent minimization techniques.
Step 3: Simulate the time responses of the closed-loop system to verify that they are satisfactory. If the
responses are not proper, go to Step (2).
Step 4: Determine the upper bound for stability robustness. Plot the maximum singular values of T(jw)
to verify that it is within the upper bound 1/m(w). If it is not satisfied, return to Step (2).
The robust stability bound and the maximum singular values of T(jw) for the response (e) of Figure
11.15 are plotted in Figure 11.18(a). It is seen that the upper bound for guaranteed robust stability is not
satisfied as the σ (T ) goes beyond the upper bound at certain frequencies. Therefore, to guarantee robust
stability one should reduce the controller bandwidth. In Figure 11.18(b), the robust stability bound and the
maximum singular values of T( jw) for the response ( f ) of Figure 11.18 is plotted. It is evident that the
gains corresponding to the response ( f ) of Figure 11.18 guarantees stability robustness. One should note
that stability robustness and speed of response are generally contradictory specifications. Hence, it turns out
Chapter 11.indd 682 3/17/2009 8:24:29 PM

a
20
b
1/m (w ) in dB
c
0 ∗
a − R r = (1.33)R r-nominal
b − R r∗ = (1.66)R r-nominal
c − R r∗ = (1.99)R r-nominal
−20
100 101
(a)
30
a
1/m (w ) in dB
b
20
a − R s∗ = (1.33)R s-nominal c
10
b − R s∗ = (1.66)R s-nominal
c − R s∗ = (1.66)R s-nominal
0
100 101
(b)
a
10
b
c
1/m (w) in dB
0
∗
a − M s = (1.33)M nominal
b − M s∗ = (1.66)M nominal
−10
c − M s∗ = (1.99)M nominal
−20 0
10 101 w (rad/s)
(c)
Figure 11.17 1/m(w) (robustness bounds) for: (a) Rotor resistance uncertainties; (b) stator resistance
uncertainties; (c) magnetizing inductance uncertainties, at wm = 157 rad/s.
as a classic case of compromise. It is evident that as the controller bandwidth is reduced, the system becomes
more and more robust. This same argument suggests that field-oriented control of an induction motor pro-
vides less stability robustness than scalar flux control techniques such as the V/f control strategies. However,
field-oriented control scores over V/f control when high dynamic performance is required.
Chapter 11.indd 683 3/17/2009 8:24:29 PM

1/m (w ), robustness bound

0
dB
s [T(jw)]
−50
10−1 100 101 102 103
(a)
1/m (w)
0
dB
s [T(jw)]
−50
−100
10−1 100 101 102 103
(b)
1/m (w)
0
dB
s [T(jw)]
−50
10−1 100 101 102 103
(c)
Figure 11.18 (a) Stability robustness with 100% uncertainty in rotor resistance, 100% uncertainty in
stator resistance and 10% uncertainty in magnetizing inductance {Q = 0.1HTH,
R = diag(1, 20), Kgains = [–0.3, –62.1088, –0.3, –48.572]}. (b) Stability robustness with
reduced bandwidth, but with 100% uncertainty in rotor resistance, 100% uncertainty in
stator resistance and 10% uncertainty in magnetizing inductance {Q = HTH,
R = diag(1, 20), Kgains = [–0.0001, –5, –0.0001, –15.6961]}. (c) Stability robustness
when the rotor resistance is adapted within 20% of the actual value even though there
is 100% uncertainty in stator resistance and 10% uncertainty in magnetizing inductance
{Q = 0.1HTH, R = diag(1, 20), Kgains = [–0.3, –62.1088, –0.3, –48.572]}.
Chapter 11.indd 684 3/17/2009 8:24:30 PM

An alternative to reducing the bandwidth of the controller to improve robustness would be to adapt the
parameters of the assumed model to that of the actual model. In Figure 11.18(c), it is shown that the stabil-
ity robustness of the gains corresponding to the response (e) of Figure 11.18 has been improved by adapting
the rotor resistance to within 20% of the actual value, that is, only partially adapted. In the case of the
induction motor plant with uncertainties in the rotor resistance, stator resistance and the magnetizing
inductance, it is evident from Figure 11.18(c) that the stability robustness property of the system improves
when the rotor resistance of the assumed model is adapted towards the actual plant parameter value. This is
because the system is more sensitive to uncertainties in the rotor resistance than the stator resistance. There-
fore, adaptation of the rotor resistance parameter improves not only the field orientation, which enhances
the dynamic performance, but also improves stability robustness.
Speed Controller Synthesis

The current control loop of the induction motor drive system is very fast compared to the speed control
loop. This is because the time constants of the stator currents are much faster than that of the mechanical
time constant of the system. Therefore, with respect to the speed loop, the plant can be approximated as a
first-order system with a lag corresponding to the mechanical time constant of the rotor. The control block
diagram for the speed control system is shown in Figure 11.19. The anti-alias filter is not shown in this case
because normally the speed is sensed using encoders which are directly compatible with digital
processors.
As this is a first-order system, one could adopt standard classical SISO techniques, like the root locus or
the Bode diagram techniques to design the speed controller. But one cannot get optimal gain values. There-
fore, using the same framework as discussed for the digital current controller synthesis, the speed control
model is also formulated as a LQ tracker problem with output feedback.
From Eq. (8.125), the plant Gpw dynamics is described by matrices Apw , Bpw , Cpw and Dpw , where
⎡ B⎤
Apω = ⎢− ⎥ (11.73)
⎣ J⎦
w mref
ew e 1w T(z + 1) e 2w +
+ z −1 kiw
2(z − 1)
− Delay i-gain +
Integrator
Plant
wm
Kpw wm
ZOH
−1
uw
p-gain
Gpw(s)
Figure 11.19 Control block schematic of the plant for the speed controller design.
Chapter 11.indd 685 3/17/2009 8:24:30 PM

(B is the friction coefficient; J the mechanical inertia, referred to the rotor shaft)
⎡1⎤
Bpω = ⎢ ⎥ (11.74)
⎣J⎦
where the state of Gpw is given by
T
⎡ω ⎤
x pω =⎢ m⎥
⎣ p ⎦
the input of Gpw is given by
upω = [uω ]T
the output of Gpw is given by

y pω = x pω
This is a special case of output feedback which is the full-state feedback. With full-state feedback, the LQ
tracker problem is much simpler and also the robustness and stability properties are better. However, follow-
ing the same underlying framework laid down for the current controller synthesis, the dynamics of the com-
pensator here is also a proportional–integral structure. One can then describe the controller Gcw by Acw , Bcw ,
Ccw and Dcw where
T
⎡0 0⎤
Acω = ⎢T ⎥ (11.75)
⎢ 1⎥
⎣2 ⎦
(where T is the sampling interval in the discrete controller);
⎡ −1 1⎤
Bcω =⎢ T T⎥ (11.76)
⎢− ⎥
⎣ 2 2⎦
where the state of Gcw , xcw is

x cω = [ε1ω , ε 2ω ]T
the input of Gcw , ucw is
T
⎡ω ω ⎤
ucω = ⎢ m , mref ⎥
⎣ p p ⎦
the output of Gcw , ycw is

y cω = x cω
The plant Gpw and the compensator Gcw are combined so that the speed control system is formulated as LQ
tracker problem for the augmented system given by Eqs. (10.101)–(10.102). The same steps used for the
Chapter 11.indd 686 3/17/2009 8:24:30 PM

current controllers are also used for the speed controller design. Figure 11.20 shows the step responses of
the shaft speed for various values of the state weighting matrix Q. It is seen that as Q increases, the speed
of response increases. The stability robustness for the speed control system for 100% uncertainty in param-
eters like J and B is shown in Figure 11.21 for the gains of response (a) of Figure 11.20. It can be seen that
the system can withstand large uncertainties in parameters J and B, without affecting the stability
robustness.
a
Shaft speed (rad/s)
b
1
1.5
0
0 50 100 150 200 250
No. of samples
1 sample = 1 ms
Figure 11.20 Step response for shaft speed: (a) Q = 100HTH, R = 0.2, [Kpw, Kiw ] = [–16.9404, –1.6134];
(b) Q = 10HTH, R = 0.2, [Kpw, Kiw ] = [–6.411, –0.6356]; (c) Q = HTH, R = 0.2, [Kpw, Kiw ] =
[–2.1645, –0.2163].
10
1/m (w), Robustness bound

0
s[T( jw)]
dB −10
−20
−30
10−1 100 101 102 103 104 105
w (rad/s)
Figure 11.21 Stability robustness with 100% uncertainty in J and 100% uncertainty in Q = 100HTH,
R = 0.2, [Kpw, Kiw ] = [–16.9404, –1.6134].
Chapter 11.indd 687 3/17/2009 8:24:30 PM

This chapter discusses some advanced aspects of problem of minimization of a cost function holds in
controller design that addresses optimality issues both cases.
and robustness issues. These controller design con- Most systems will have their system parameters
cepts are generally employed for complex and high varying during operation. This may be due to tem-
performance systems. However, building optimality perature drifts, change in operating points or heat-
and robustness into the controller and estimator ing due to load conditions. As a consequence the
designs is an added feature that one may incorporate parameters of the actual system are at variance with
to achieve better energy efficiencies, stability and respect to the system parameters used in the con-
performance quality. troller design. If the closed-loop pole locations are
It must be understood that higher the desired close to the stability boundary, then there is a pos-
speed of response, higher will be the instantaneous sibility that the system may drift into the unstable
power that will be drawn from the power source. zone or the performance may deteriorate. Robust
Optimal controllers will not change this principle, controllers address these issues. In the face of
but they would ensure that for a given speed of uncertainties, in parameters and disturbances,
response, the power drawn will be minimum if the robust controllers ensure that stability and perfor-
energy constraints are built into the cost function or mance are not compromised. This would actually
performance index. Optimal controllers generally result in controllers where the closed-loop system
operate on minimizing the state and input energy. poles are located close to the origin of the unit
On the other hand, the optimal estimators operate circle. In that sense, the dead beat controller is a
on minimizing the state error, though the generic very robust controller.
The following exercises can be studied using b. Formulate the problem for least square
MATLAB. solution.
1. A first-order system is given. This is in the form c. Generate input and output data for the
of a file sys_mod.p which is a MATLAB file. The system (recommendation − use random
file sys_mod.p is included in the accompanying inputs).
CD. The system has the following syntax: d. Estimate the parameters.
y = sys_mod(u) e. Plot the evolution of error and parameters
with time.
where u is the input and y is the output. The
After trying out the algorithm on the above
model of the system given in sys_mod.p is of
system, check if the algorithm works for
the form
another system which is of the form
(1 + b1z −1 ) yk = (a0 + a1z −1 )uk −1
(1 + b1z −1 + b2 z −2 ) yk = ( a0 + a1z −1 + a2 z −2 )uk −1
The objective is to estimate a0, a1 and b1 using
the least square algorithm for minimizing the wherein one has to estimate a0, a1, a2, b1 and
error. This involves the following tasks: b2. This system is given in sys_mod01.p file.
a. Write a general purpose least square algo- 2. A first-order system is given. This is in the form
rithm as an m-function. of a file sys_mod.p which is a MATLAB file.
Chapter 11.indd 688 3/17/2009 8:24:31 PM

L1 iL1 C1 i L2 L2 Vo
+ −
V c1
+
Vs S U2 D C R
−
U1
V GND = 0
Figure 11.22 Cuk converter circuit.
The file sys_mod.p is included in the accompa- xˆ = Axˆ + Buˆ

nying CD. The system has the following
syntax: yˆ = Cxˆ
where
y = sys_mod(u)
where u is the input and y is the output. The ⎡ −(1 − D ) ⎤
model of the system given in sys_mod.p is of ⎢ 0 0
L1
0 ⎥
⎢ ⎥
the form ⎢ −D −1 ⎥
⎢ 0 0
L2 ⎥
(1 + b1z −1 ) yk = (a0 + a1z −1 )uk −1 A=⎢
L2
⎥;
The objective is to estimate a0, a1 and b1 using ⎢1 − D D ⎥
⎢ C 0 0 ⎥
the least square algorithm for minimizing the C1
⎢ 1 ⎥
error. This involves the following tasks: ⎢ 1 −1 ⎥
⎢ 0 C2
0
RC 2 ⎥⎦
a. Write a general purpose LMS algorithm as ⎣
an m-function.
b. Formulate the problem for least square
solution. ⎡1 −V g ⎤
⎢ ⎥
c. Generate input and output data for the ⎢ L1 L1(1 − D ) ⎥
system using random inputs. ⎢ −V g ⎥
⎢0 ⎥
L2 (1 − D ) ⎥ ; C = [1 0 0 0]
d. Estimate the parameters.
B=⎢
e. Plot the evolution of error and parameters
⎢ Io ⎥
with time. ⎢0 ⎥
⎢ C1(1 − D ) ⎥
After trying out the algorithm on the above ⎢0 ⎥
system, check if the algorithm works for ⎣ 0 ⎦
another system which is of the form
(1 + b1z −1 + b2 z −2 ) yk = ( a0 + a1z −1 + a2 z −2 )uk −1 The converter is designed to meet the following
specification: Vg = 130 V, I0 = 1.5 A, V0 = 15 V,
wherein one has to estimate a0, a1, a2, b1 and
fs = 10 kHz, ΔVo = 1500 mV and D = 0.6. The
b2. This system is given in sys_mod01.p file.
sampling frequency is 5 kHz. To meet the
3. Consider the Cuk converter DC–DC con- above specification, the selection of compo-
verter shown in Figure 11.22. The small signal nents are: R = 10 Ω, C1 = 100 μF, C2 = 10 μF,
model of the converter of Figure 11.22 is L1 = 10 mH, L2 = 10 mH. The plant noise
given as co-variance matrix is
Chapter 11.indd 689 3/17/2009 8:24:31 PM

⎡0.1 0 0 0⎤ b. Design the linear quadratic regulator gain

⎢ 0 0.1 0 0⎥
for the above system.
⎢ ⎥
⎢0 0 0.1 0 ⎥ 4. Consider the buck converter DC–DC con-
⎢ ⎥ verter shown in Figure 11.23. This circuit
⎣0 0 0 0.1⎦
includes few non-idealities of the components
and the sensor noise co-variance is 0.1. of the circuit. The small signal state space
a. Design the Kalman estimator gain for the model for the above circuit is given as
above system. Eq. (11.77)
⎡ diˆ ⎤ ⎡ 1⎛ Rc R ⎞ 1 ⎛ R + 2 Rc ⎞⎤
⎢ ⎥ ⎢− ⎜ D( Rq + Rs ) + (1 − D )Rd + Rw − ⎟ − ⎜ ⎟⎥ ⎡ ˆ ⎤ ⎡ D Vg − Vd ⎤ vˆ
⎡ ⎤
⎢ L⎝ R + Rc ⎠ L ⎝ R + Rc ⎠⎥ i + ⎢ ⎥ g
⎢ dt ⎥ = L ⎥. ⎢ ˆ ⎥
⎢ dvˆc ⎥ ⎢ ⎥ ⎢⎢vˆc ⎥⎥ ⎢ L
⎢⎣ dt ⎥⎦ ⎢
R
−
1
⎥ ⎣ ⎦ ⎢⎣ 0 0 ⎥⎦ ⎢⎣d ⎥⎦
⎣⎢ c ( R + Rc ) c ( R + Rc ) ⎥⎦
C = [0 1]
The following parameters are given: Vg = 10 V; values. Obtain the transfer function of the
L = 1e–3; C = 100e–6; R = 2; V d = 0.7; D output with respect to the duty ratio. Com-
= 0.4; Rc = 1e–3; Rw = 10 mΩ; Rq = 10 mΩ; pute the multiplicative uncertainty,
Rs = 0.1; Rd = 0.01; Ts = 1/20 kHz. M( jw ).
d. Plot the reciprocal of the maximum singular
a. Design the Kalman estimator gain for the
value of M( jw ) as a function of frequency.
above system assuming the noise co-variances
This will provide the robustness upper
of the same order of magnitude as that for
bound.
the previous exercise.
e. Compute the maximum singular value of
b. Design the linear quadratic regulator gain
the co-sensitivity T( jw ) and plot it as a
for the above system.
function of frequency.
c. Assume 50% uncertainty in the value of C,
20% uncertainty in the value of L and 10% Note: Software packages like MATLAB give
uncertainty in the non-ideal component computation and plotting of singular values.
Rq
Q
L Rw
C
Rs
PWM
generator Vd R
Vg Rc
Rd
Buck converter
Figure 11.23 Buck converter circuit with non-idealities.
Chapter 11.indd 690 3/17/2009 8:24:31 PM


1. A controller that has been designed to meet the 14. For xTCx, if C is negative definite, then there
performance criteria under some constraints is always exists a .
called an controller.
15. All physical systems will reach a stable state also
2. An estimator that has been designed to meet called the state.
the performance criteria under some constraints 16. In the equilibrium state the system will always
is called an estimator. be at its minimum state.
3. Kalman gain is one in an controller. 17. Maximum energy will be removed from a system
4. Linear quadratic regulator achieves the control such that it reaches the or
performance criteria under the constraints of energy state.
minimum and energies. 18. If the measured data has a large variance, then
5. Kalman filter is an estimator that that data should be trusted .
estimates the unavailable states of a system in 19. If the measured data has a small variance, then
the presence of . that data should be trusted .
6. If x is the state vector and ei are errors at 20. The LMS algorithm is a least squares
different instants of time, then the value of x algorithm.
n
that gives minimum value of ∑ ei2 is the 21. The LMS algorithm chooses the correction
i =1 gain in such a way that the minimum is reached
solution. with the slope.
7. Least squares solution is the estimate 22. The Kalman filter is a estimator
of the state vector. algorithm.
8. For all non-zero values of x, if xTCx > 0, then 23. The Kalman filter algorithm chooses the correc-
C is called definite. tion gain in such a way that the error in the
presence of system is minimized.
9. For all non-zero values of x, if xTCx < 0, then
C is called definite. 24. The LMS algorithm belongs to the class of
estimator.
10. For all non-zero values of x, if xTCx is greater
than or equal to zero, then C is called 25. The Kalman filter belongs to the class of
-definite. estimator.
11. For all non-zero values of x, if xTCx is lesser 26. The steady-state Kalman gain will give
than or equal to zero, then C is called estimator performance.
-definite. 27. In LQR, the state weighting matrix should be
positive .
12. For all non-zero values of x, if xTCx is either
greater than, lesser than or equal to zero, then 28. In LQR, the input weighting matrix should be
C is called . positive for a minimum to exist.
13. For xTCx, if C is positive definite, then there 29. The actual plant and the model plant will never
always exists a . be .
Chapter 11.indd 691 3/17/2009 8:24:32 PM

30. Stability robustness is the ability to bounded above by its value and
closed-loop stability in spite of parameter bounded below by its value.
.
34. The design of the robust controller needs only
31. Performance robustness is the ability to the maximum and minimum values
acceptable performance even though of the transfer function.
the system may be subject to .
32. The multi-variable Bode magnitude plot is the 35. For stability robustness, the co-sensitivity
plot of the transfer function versus should be bounded above by the of
the frequency. the multi plicative modeling discrepancy
bound.
33. For any given input, the magnitude of the
transfer function at any given frequency is
1. Explain least squares principle. 11. Discuss how learning rate affects the stability
of the LMS estimator.
2. Discuss quadratic forms. Where and why are
quadratic forms used? 12. What are the features of the Kalman estimator?
3. What is the minimum energy principle for any 13. Explain the time update and measurement
given system? update in a Kalman estimator.
4. What is the least squares solution for y = Cx? 14. Discuss the expectation and co-variances of the
Give a graphical interpretation of the least plant and sensor noise.
squares solution.
15. What is state error co-variance? What is its
5. Explain the motivation for weighted least importance in the Kalman estimator?
squares.
16. Distinguish between Kalman gain and steady-
6. How are the weights of the weighting matrix state Kalman gain in a Kalman filter.
in the weighted least squares method selected?
17. Discuss the development of the cost function
7. What is recursive least squares? Discuss the for the linear quadratic regulator.
motivation for recursive least squares.
18. Write short notes on: (a) input energy weight-
8. Write a short note on learning rate. ing and (b) state energy weighting.
9. What is the cost function for the LMS 19. Discuss the cost function for the discrete linear
algorithm? quadratic regulator for full-state feedback.
10. What is the effect of the learning rate on the 20. Discuss the cost function for the discrete linear
convergence speed in the LMS estimator? quadratic tracker for output feedback.
Chapter 11.indd 692 3/17/2009 8:24:32 PM

21. What is linear quadratic Gaussian? Discuss. 26. Discuss the frequency domain performance speci-
fications for robust controller design in terms of
22. Write short notes on: (a) gradient descent min-
the singular values of the transfer functions.
imization, (b) SIMPLEX minimization.
23. What is the motivation for robust controller 27. How are modeling uncertainties incorporated
design? into the robust controller design?
24. Write short notes on: (a) sensitivity, (b) co- 28. Write short notes on: (a) additive uncertainties,
sensitivity. (b) multiplicative uncertainties.
25. Discuss the singular value algebra. 29. Discuss the robustness bound.
| PROBLEMS
1. Consider the following deterministic system: and the Kalman estimator gain. What is the
effect of the plant and sensor noise co-variances
(1 + b1z–1 + b2z–2)yk = (a0 + a1z–1 + a2z–2) uk
on the Kalman estimator gain?
Formulate an estimator to estimate a0, a1, a2,
b1 and b2 using the least square algorithm for 6. Derive the cost function for the least square
minimizing the error. regulator.
2. In a particular laboratory experiment, a multi- 7. Suggest generic methods for obtaining the
meter is used to measure the voltage at a parti- values of the state weighting matrix and the
cular node of a DC circuit. Three readings were input weighting matrix for a given system.
taken at different instants of the time. The first 8. For the following transfer functions compute
reading was 94 V, the second 97 V and the the multiplicative uncertainty for 20% varia-
third 89 V. What is the least squares estimate of tions in the pole and zero locations:
the voltage at the node (a) after two readings,
(b) after three readings? 10
a. Gp ( s ) =
3. Consider the following deterministic system s ( s + 1)
(1 + b1z–1 + b2z–2)yk = (a0 + a1z–1 + a2z–2)uk 10( s + 2)

b. Gp ( s ) =
( s + 0.5)( s + 1)( s + 3)
Formulate a LMS estimator to estimate a0, a1,
a2, b1 and b2. 9. For Problem 8, assume a sampling rate of 1 ms
and convert the transfer functions to the discrete
4. What are the dimensions of the plant noise
domain using zero-order hold. Now compute the
co-variance matrix and the sensor noise co-
multiplicative uncertainty for the discrete plant.
variance matrix?
5. For a first-order single-input single-output 10. For Problem 9, plot the robustness bound and
system, derive the steady-state error co-variance the co-sensitivity.
Chapter 11.indd 693 3/17/2009 8:24:32 PM

| ANSWERS
Fill in the Blanks
1. optimal 13. minimum 25. current
2. optimal 14. maximum 26. sub-optimal
3. optimal 15. equilibrium 27. semi-definite
4. input; state 16. energy 28. definite
5. optimal; noise 17. equilibrium; minimum 29. identical
6. least squares 18. less 30. guarantee; variations
7. best 19. more 31. guarantee; disturbances
8. positive 20. recursive 32. singular values
9. negative 21. steepest 33. maximum singular; minimum
10. positive; semi 22. recursive singular
11. negative; semi 23. noise 34. singular
12. indefinite 24. prediction 35. reciprocal
Chapter 11.indd 694 3/17/2009 8:24:32 PM

Discrete Computation Essentials
Learning Objectives
CHAPTER
12
select the appropriate numeric format for a specific application.
use the various arithmetic algorithms.
I t must be mentioned that most of the applications today are implemented with the microcontroller or
microprocessor as the controlling device. This implies that all the control algorithms will be implemented
within the digital domain. This would require the implementation of equations, generation of waveforms
and implementation of the controllers. This chapter discusses the systematic approach towards implementa-
tion of the numerical algorithms for the various applications.
The first step in a discrete domain implementation is the selection of an appropriate fixed-point numeric
format. The number of fractional bits and the number of integer bits are selected by applying engineering
logic for the specific application. One should also become familiar with the basic arithmetic like addition,
subtraction and multiplication for a specific numeric format.
The next important part of discrete domain implementation is the notion of normalization and scaling.
This is very important to the development of efficient and reliable working algorithms. Therefore, a proper
understanding of these concepts will go a long way in developing robust implementations. Implementation
of equations would involve operations like reciprocal, square root, exponential, sine and cosine functions
and logarithms. The implementation of these operations will be discussed. This will be followed by a discus-
sion on commonly implemented algorithms like PI controllers and PWM generation that are required for
most power electronic systems. The discussion of the algorithms in this chapter will be by use of pseudo-
codes. These pseudo-codes may be converted to the respective codes for a specific processor by appropriately
mapping the instruction set to the pseudo-codes.
12.1 Numeric Formats
I n the implementation of the control and signal generation algorithms for the various power electronic
systems, the governing equations are based on the real numbers. The real numbers are represented as .
The real number system has the following two distinctive features:
Chapter 12.indd 695 3/17/2009 8:23:03 PM

1. is infinite in range (both positive and negative). This implies that the real number line extends from
0 to +∞ and 0 to –∞ .
2. is infinitely dense. This implies that the interval between any two real numbers contains infinite
number of real numbers.
However, in a digital processor, due to the limitation of the length of the data bits, the discrete real number
has neither of the above two features. As a consequence, this results in two major problems:
1. Overflow or Out-of-Range Problem: As the data bit length is finite, the range of numbers that can be
represented is also finite. Any number on the real number line that is greater than the maximum number
that can be represented, will result in an overflow error.
2. Precision or Resolution Problem: Again as the data bit length is finite, the minimum number that can
be represented is also finite. Therefore, the interval between any two real numbers contains finite
number of real numbers. This implies that the resolution is finite.
Therefore, for discrete computation, one has to discuss and focus on the finite real number system that can
be applied and used for implementation in a digital processor. Broadly, there are two classes of
representations:
1. Fixed-Point Number System: It is the more popular format for low computational power and low-cost
digital processors.
2. Floating-Point Number System: It is used in the high computational power and high-cost processors.
Fixed-Point Numeric Format

The representation of numbers in a fixed-point format is a generalization of the familiar decimal representa-
tion of a number as a string of digits with a decimal point. In this notation, the digits to the left of the decimal
point represent the integer part of the number and the digits to the right of the decimal point represent the
fractional part of the number. Thus, in general, a real number system is represented in fixed-point format as
D(B, i, f )
where B is the base (generally 2 for binary numbers, 10 for decimal numbers); i the number of digits for the
integer part; f the number of digits for the fractional part.
The real number x can be written as
⎡ ⎛d d− f ⎞⎤
( ) d
x = ± ⎢ d 0 ⋅ B 0 + d1 ⋅ B 1 + + d i −1 ⋅ B i −1 + ⎜ −1 + −22 + + f ⎟⎥
⎣ ⎝ B B B ⎠⎦
EXAMPLE 12.1 For a 16-bit processor

D(2, 0, 16) implies base 2 system with 0 sign bit, 0 bit for integer digit, 16 bits for
the fractional digits
D(2, 0, 15) implies base 2 system with 1 sign bit, 0 bit for integer digit, 15 bits for
fractional digits.
D(2, 4, 12) implies base 2 system with no sign bit, 4 bits for integer digits, 12 bits
for fractional digits.
D(2, 3, 12) implies base 2 system with 1 sign bit, 3 bits for integer digits, 12 bits
for fractional digits.
Chapter 12.indd 696 3/17/2009 8:23:05 PM

Discrete Computation Essentials 697
which is written as
x = ± d i −1... d1d 0 . d −1d −2 d −3 ... d − f
where 0 ≤ d k < B for − f ≤ k < i.
Relative Error and Relative Density

The relative error is defined as
⎛ x − fix( x ) ⎞
rE = ⎜ ⎟ × 100 (12.1)
⎝ x ⎠
where rE is the percent relative error; x the real number; fix(x) the real number that is represented in the
specified digital domain fixed-point format that is fix(x) ∈ D(B, i, f ).
Consider a five-digit fixed-point system D(10, 3, 1). This is a base 10 system that consists of 1 sign digit,
3 integer digits and 1 fractional digit. With this system the range of numbers that can be represented is from
–999.9 to +999.9. A real number x in this range can be represented by
fix( x ) ∈ D(10, 3, 1)
Consider a representative real number towards the end of the range, say x = 865.54. Then fix( x ) = 865.5 .
The absolute error is x − fix( x ) and is equal to 0.04. From Eq. (12.1), relative error is given as
⎛ 865.54 − 865.5 ⎞
rE = ⎜ ⎟ × 100 ≈ 0.005%
⎝ 865.54 ⎠
If a representative real number near zero is selected, say x = 0.86554, then fix(x) = 0.9. The absolute error is
less than 0.05. The relative error is
⎛ 0.86554 − 0.9 ⎞
rE = ⎜ ⎟ × 100 ≈ −4%
⎝ 0.86554 ⎠
This shows that the relative error near zero is more than the relative error near the end of the range for simi-
lar absolute errors. This implies that the relative density of the fixed-point number system is not uniform
within the range. This, in fact, is the basic disadvantage of the fixed-point number system.
Problem 12.1
Consider an eight-bit digital processor. If one bit is used as sign bit, then what is the range of numbers for
(a) D(2, 3, 4)
(b) D(2, 1, 6)
Solution
(a) The fixed-point system D(2, 3, 4) is a binary system (i.e., base 2) with the fractional part having four
bits. Out of the remaining four bits, one bit is used as the sign bit. The remaining three bits are the
integer bits. A number x is represented as
Chapter 12.indd 697 3/17/2009 8:23:05 PM

⋅
3
sign(2 ) 2 2 2 2 1 0
2 −1 2 −2 2 −3 2 −4

Integer part Fractional part
The largest number is 0 111 1111. This is the representation of

⎛1 1 ⎞
( ) ( ⎝ 2 2
)
x = − sign ⋅ 23 + 22 + 21 + 20 + ⎜ + 2 + 3 + 4
1
2 2
1
⎟
⎠
= 0 + 7 + 0.9375 = 7.9375
Thus, (01111111)2 ⇔ (7 F )16 ⇔ (7.9375)10. This binary base 2 number can be equivalently repre-
sented in base 16 (or hexadecimal) as 7F or in base 10 (decimal) as 7.9375.
For the negative number, the 2’s complement is used. The 2’s complement of the largest number
gives the minimum value that the above system can represent. Thus, taking the 2’s complement of
the above number, one obtains
2’s complement of 0 111 1111 = 1 000 0001
which is the representation of the following base 10 number:
⎛0 0 0 1 ⎞
x = ( −sign ⋅ 23 ) + (0 ⋅ 22 + 0 ⋅ 21 + 0 ⋅ 20 ) + ⎜ + 2 + 3 + 4 ⎟
⎝2 2 2 2 ⎠
= −8 + 0 + 0.0625 = −7.9375
(b) The fixed-point system D(2, 1, 6) is a binary system with the fractional part having six bits. Out of
the remaining two bits, one bit is used as sign bit. The remaining one bit is the integer bit. A number
x is represented as
⋅
sign( 21 ) 20 2−1 2−2 2−3 2−4 2−5 2−6

Integer part Fractional part
The largest number is 0 1 111111. This is the representation of the following base 10 number:
⎛1 1 1 1 1 1 ⎞
x = ( −sign ⋅ 21 ) + ( 20 ) + ⎜ + 2 + 3 + 4 + 5 + 6 ⎟
⎝2 2 2 2 2 2 ⎠
= 0 + 1 + 0.984375 = 1.984375
The 2’s complement of the largest number gives the minimum value that the above system can
represent. Thus, taking the 2’s complement of the above number, one obtains
2’s complement of 0 1 111111 = 1 0 000001
which is the representation of the following base 10 number:
⎛0 0 0 0 0 1 ⎞
x = (−sign ⋅ 21 ) + (0 ⋅ 20 ) + ⎜ + 2 + 3 + 4 + 5 + 6 ⎟
⎝2 2 2 2 2 2 ⎠
= −2 + 0 + 0.015625 = −1.934375
Thus, (10000001)2 ⇔ (81)16 ⇔ (−1.934375)10 .
Chapter 12.indd 698 3/17/2009 8:23:05 PM

Floating-Point Numeric Format

The floating-point representation consists of a mantissa M which is multiplied by the exponential factor BE,
where B is the base and E is the exponent. A real number system in floating-point format is represented as
D(B, f, L, U )
where B is the number base; f the precision of the mantissa or length of the fractional part of the mantissa;
L the lower bound of the exponent E and U the upper bound of the exponent E.
A non-zero real number x is represented in the floating-point notation as
x = M ⋅BE (12.2)
where L ≤ E ≤ U.
The real number x can be written as
⎛d d d df ⎞
x = ± ⎜ 1 + 22 + 33 + + f ⎟ × B E
⎝B B B B ⎠
which is written as
x = ±0.d1d 2 d 3 ... d f × B E
where 1 ≤ d1 < B (implies that the first digit is non-zero); 0 ≤ d k < B for 2 ≤ k ≤ f.
The above representation where d1 is non-zero is called the normalized form. In the normalized form, the
absolute value of mantissa (M ) is
1
≤ M <1
B
However, in the normalized form, one cannot represent the number zero directly. It can be represented as
the subtraction of two equal normalized numbers. Normalized representation of the numbers in the floating-
point system improves the precision as can be observed from the following example.
EXAMPLE 12.2 Consider a floating-point system D(10, 4, –2, 3). With this system the range of
numbers that can be represented in the normalized form is from ±0.1 × 10−2 to
±0.9999 × 103 (i.e., −999.9 to −0.001 and +0.001 to +999.9 ).
Let x = 0.029411765 which can be represented in the following two forms:
1. Without normalization: 0.0294 × 100
2. With normalization: 0.2941 × 10−1
It can be observed that the precision in the normalized representation is more than
that without normalization. The digits are better utilized with normalization.
Relative Error and Relative Density

The relative error is defined as
⎛ x − float( x ) ⎞
rE = ⎜ ⎟ × 100 (12.3)
⎝ x ⎠
where rE is the percent relative error, x the real number, float(x) is the real number that is represented in the
specified digital domain floating-point format, that is, float(x) ∈ D(B, f, L, U ).
Chapter 12.indd 699 3/17/2009 8:23:06 PM

Consider a floating-point system D(10, 4, –2, 3). Let x = 865.54. Then float( x ) = 0.8655 × 103 . From
Eq. (12.3), the relative error is given as
⎛ 865.54 − 865.5 ⎞
rE = ⎜ ⎟ × 100 ≈ 0.005%
⎝ 865.54 ⎠
If a representative real number near zero is selected, say x = 0.86554, then float( x ) = 0.8655 × 100 . The rela-
tive error is
⎛ 0.86554 − 0.8655 ⎞
rE = ⎜ ⎟ × 100 ≈ 0.005%
⎝ 0.86554 ⎠
This shows that the relative error near zero is the same as the relative error near the end of the range for simi-
lar absolute errors. This implies that the relative density of the floating-point number system is uniform
within the range. Thus, the floating-point system has uniform relative density.
However, in the case of normalized floating-point system, there exists an underflow hole, that is, there is
a jump from minimum value to zero, which implies that values between the minimum value and zero
cannot be represented. The underflow hole is depicted in Figure 12.1(b). Figure 12.1(a) shows the pictorial
depiction of the fixed-point system and Figure 12.1(b) shows the pictorial depiction of the floating-point
system. In the fixed-point system the numbers are uniformly distributed over the range. This makes the
relative density non-uniform in the fixed-point system. However in the floating-point system, the numbers
have a higher density towards zero as compared to the numbers away from zero. This makes the relative
density uniform in the floating-point system.
In Figure 12.1(b), the underflow hole can be observed. To take care of the underflow hole problem, the
floating-point normalized representation is generally modified as
x = ±1.d1d 2 d 3 ... d f × B E
where 0 ≤ d k < B for 1 ≤ k ≤ f and 0 = 1.00 × B 0 − 1.00 × B 0 . In this representation, there is no restriction
on d1 to be non-zero. Therefore one can represent smaller numbers. For example, in the floating-point
−999.9 0 +999.9
Fixed point system
(a)
Underflow hole
−999.9 −0.001 +0.001 +999.9

0
Floating point system

Range
(b)
Figure 12.1 Pictorial depiction of: (a) Fixed-point system; (b) floating-point system.
Chapter 12.indd 700 3/17/2009 8:23:06 PM

system D(10, 4, –2, 3) the minimum in the previous normalized representation is 0.001. If x = 0.0001, then
it cannot be represented. In the modified normalized representation, x = 0.0001 is obtained as
float( x ) = 1.0002 × B 0 − 1.0001 × B 0
Problem 12.2
Consider an eight-bit digital processor. If one bit is used as sign bit, then what is the range of numbers for
(a) D(2, 4, 0, 7)
(b) D(2, 4, –3, 3)
Solution
(a) The floating-point system D(2, 4, 0, 7) is a binary system (i.e., base 2) with the mantissa having four bits.
Out of the remaining four bits, one bit is used as sign bit. The remaining three bits are the exponent bits.
It should be noted that the sign bit is for the mantissa part alone. A number x is represented as
sign 22 21 20 2−1 2−2 2−3 2−4

Exponent Mantissa
The smallest positive number is 0 000 1000. This is the representation of the following base 10 number,
x = 0.5 × 20 = 0.5
The largest positive number is 0 111 1111. This is the representation of the following base 10 number:
x = ⎛ + 2 + 3 + 4 ⎞ × 27 = 0.9375 × 27 = 120
1 1 1 1
⎝2 2 2 2 ⎠
For negative numbers, the 2’s complement of the mantissa alone is performed. For the number 0 000
1000, on taking the 2’s complement of the mantissa along with the sign bit, the bit pattern 1 000
0001 is obtained and this represents
⎡ ⎛0 0 0 1 ⎞⎤
x = ⎢(−sign ⋅ 20 ) + ⎜ + 2 + 3 + 4 ⎟⎥ × 27 = [ −1 + 0.0625] × 27 = −0.9375 × 27 = −120
⎣ ⎝2 2 2 2 ⎠⎦
However, it should be noticed that the mantissa in the above case is not normalized. In the normalized
form, the first digit of the mantissa should always be non-zero. Thus, the maximum negative number that
can be represented in the normalized form is 1 000 1001. This represents the following base 10 number,
⎡ ⎛1 0 0 1 ⎞⎤
x = ⎢(−sign ⋅ 20 ) + ⎜ + 2 + 3 + 4 ⎟⎥ × 27 = [ −1 + 0.5625] × 27 = −0.4375 × 27 = −56
⎣ ⎝2 2 2 2 ⎠⎦
As the mantissa is normalized, the sign bit value will be the next higher bit value with respect to the
mantissa and therefore the bit value of the sign bit is 20 as indicated in the above calculation.
(b) The floating-point system D(2, 4, –3, 3) is a binary system (i.e., base 2) with the mantissa having four
bits. Out of the remaining four bits, one bit is used as sign bit. The remaining three bits are the expo-
nent bits. Among the three exponent bits, one is used as sign bit for the exponent. This will leave only
two bits for the exponent amplitude. A number x is represented as
Chapter 12.indd 701 3/17/2009 8:23:06 PM

sign sign 21 20 2−1 2−2 2−3 2−4

Exponent Mantissa
The smallest positive number is 0 101 1000. Here the exponent part alone is the 2’s complement of
011 which is 101 or −3. This is the representation of x = 0.5 × 2−3 = 0.0625 .
The largest positive number is 0 011 1111. This is the representation of the following base 10 number:
⎛1 1 1 1 ⎞
x = ⎜ + 2 + 3 + 4 ⎟ × 23 = 0.9375 × 23 = 7.5
⎝2 2 2 2 ⎠
The bit pattern 1 011 1001 represents x = –3.5.
General Notes on Number System

1. The relative density is uniform in the floating-point number system whereas it is not uniform in the
fixed-point number system.
2. In the fixed-point system, one must keep track of the position of the point separating the integer and
fractional parts whereas there is no need to keep track of the decimal or binary point in the floating-
point system.
3. There is a limitation in the floating-point system in representing zero in the normalized format whereas
in the fixed-point system, the representation of zero is like natural numbers.
4. The floating-point system has the problem of underflow “hole” that limits the representation of very
small numbers.
5. In floating-point system, it can be observed from Figure 12.1(b) that the interval between numbers
increases as the numbers grow in magnitude. It is this characteristic that allows floating-point numbers to
preserve the significant digits. However, the quantizing errors are proportional to the magnitude of the
number, x being represented. For normalized floating-point numbers, the worst case quantization error
bounds can be expressed as
Δx
≤ ∈ if truncation is employed
x
Δx ∈
≤ if rounding is employed
x 2
where ∈ = B 1− f is called the machine epsilon and Δ x is the error due to quantization.
6. The computation overhead for the fixed-point system is much lower than that for the floating-point
number system as arithmetic operations must be performed on both mantissa and exponent. Therefore,
for a given calculation, the fixed-point system is faster on a given digital processor. This would imply
that for a given algorithm, a lower power digital processor can be used in a fixed-point system as com-
pared to the floating-point system. This would reduce the cost of the algorithm and the system cost.
Therefore the fixed-point system is the more popular choice. Floating-point system is used only when
the computations become very large and complex.
Chapter 12.indd 702 3/17/2009 8:23:07 PM

12.2 Tracking the Base Point in the Fixed-Point System
T he discussions in the remaining portion of the chapter will use the fixed-point number system as this is
the first choice for a given application. Only if it becomes extremely difficult or impossible to manage
with fixed-point number system should one shift to the floating-point number format. One of the major
problems that is faced in the fixed-point system is the need to keep track of the virtual base point (decimal
point in base 10 system or the binary point in the base 2 system). This section will discuss the primary arith-
metic operations and the method of tracking the base point in a systematic manner.
There are three primary operations based on which all arithmetic calculations are performed in the
digital processor. The three primary operations are:
1. addition with symbolic operator “+”;
2. subtraction with symbolic operator “–”;
3. multiplication with symbolic operator “*”.
Addition of Numbers
The addition can be performed with two numbers only if the two numbers are in the same fixed-point
numeric format. Thus, if x1 is in D(B, i, f ) and x2 is in D(B, i, f ), then x1 + x2 is also in D(B, i, f ). If x1 is
in D(B, i1, f1) and x2 is in D(B, i2, f2) where i1 ≠ i2 and f 1 ≠ f 2 then it is not possible to add x1 and x2.
Either x1 should be converted to D(B, i2, f2) or x2 should be converted to D(B, i1, f1) format before perform-
ing the addition. The conversion from one numeric format to another for the same base B is done by either
right shifting or left shifting.
Subtraction of Numbers
Like addition, subtraction also can be performed with two numbers only if the two numbers are in the same
fixed-point numeric format. Thus, if x1 is in D(B, i, f ) and x2 is in D(B, i, f ), then x1 – x2 is also in D(B, i,
f ). If x1 is in D(B, i1, f1) and x2 is in D(B, i2, f2) where i1 ≠ i2 and f 1 ≠ f 2 then it is not possible to subtract
x2 from x1 without converting one of the numbers into a compatible numeric format.
Multiplication of Numbers
Unlike addition or subtraction, multiplication can be performed on two numbers of different numeric
formats but with the same base B. Thus if x1 is in D(B, i1, f1) and x2 is in D(B, i2, f2), then x1 × x2 will be in
D(B, i1 + i2, f1 + f2).
EXAMPLE 12.3 Let x1 = +113.54 in D(10, 3, 2) and x2 = +008.34 in D(10, 3, 2). The product is
given as
x1 × x 2 = ( +113.54) × ( +008.34) = +000946.9236
Observe that the product is in D(10, 6, 4). If the product has to be converted to
D(10, 3, 2) numeric format, then there are two possible methods:
1. Shift Right: The product x1 × x2 in D(10, 6, 4) is shifted right by two digits.
This will cause the two lower significant digits to be removed and two zeros
will be included in the most significant bits (MSBs). The number now is
+00000946.92. The 5 least significant bits (LSBs) of this number are stored
in a new variable as +946.92 which is in D(10, 3, 2).
Chapter 12.indd 703 3/17/2009 8:23:07 PM

2. Shift Left: The product x1 × x2 in D(10, 6, 4) is shifted left by three digits.

This will result in +946.9236000. The 5 MSBs of the left-shifted number are
stored in a new variable as +946.92 which is in D(10, 3, 2).
In general if x1 is in D(B, i1, f1) and x2 is in D(B, i2, f2), then let the product x1 × x2
be in D(B, ip, fp). To convert the product to D(B, i1, f1), then shift right by f2
number of bits and extract the (i1 + f1) LSBs or shift left by i2 number of bits and
extract the (i1 + f1) MSBs to express the product in D(B, i1, f1). In order to express
the product in D(B, i2, f2), shift right by f1 number of bits and extract the (i2 + f2)
LSBs or shift left by i1 number of bits and extract the (i2 + f2) MSBs.
EXAMPLE 12.4 Let x1 = +1213.54 in D(10, 4, 2) and x2 = +04.347 in D(10, 2, 3). The product is
given as
x1 × x 2 = ( +1213.54) × ( +04.347 ) = +005275.25838
The product x1 × x2 is in D(10, 6, 5). If the product x1 × x2 is to be represented in
D(10, 4, 2) say, then i1 = 4, f1 = 2, i2 = 2 and f2 = 3. Shift right the product
number by f2 bits (three bits), that is, +000005275.25 and extract the (i1 + f1)
(4 + 2 = 6) LSBs, that is, +5275.25. Alternately left shift the product number by
i2 bits (two bits), that is, +5275.2583800 and extract the i1 + f1 (4 + 2 = 6) MSBs,
that is, +5275.25.
12.3 Normalization and Scaling

or an n-digit system, the numbers are represented in a numeric format D(B, i, f ). The resolution of the
F numeric format is 1/Bf. The precision of the numbers is highest if “f ” has the maximum number of
digits allotted to it out of the available n digits. In an n-digit system, f is maximum when
i = 0 and f = n, that is, D(B, 0, n)
For highest precision, the resolution is 1/Bn. However, in the digital domain with finite number of digits,
one has to compromise between the range and the precision. If the precision is to be high, then the number
of integer bits will become low and as a consequence the range will reduce. In the limiting case when the
fractional part f = n, the integer part i = 0. Thus for an n-digit system, at maximum precision, the range is
limited to 1.
If D(B, 0, n) numeric format is to be used, then all the variables in the processor for that specific applica-
tion should have a value less than 1 so that the values are within the range 0–1. It is not sufficient if the
values of the variables are within the range. The values resulting from the primary arithmetic operations
must also stay within the range. Thus if x1 and x2 are in D(B, 0, n), then
1. x1 + x2 < 1 if x1max = a then x2max < 1 − a and vice-versa where a < 1.
2. x1 – x2 < 1 if x1 > x2. If x2 > x1, then the result will be out of range as there is no sign bit in D(B, 0, n)
numeric format.
3. x1 × x2 < 1 if both x1 and x2 are less than 1.
Therefore, D(B, 0, n) is not a very practically useful numeric format. A popular numeric format that can cater to
many applications is D(B, 1, n – 2) for an n-digit system. This implies that there is 1 sign digit, 1 integer digit
Chapter 12.indd 704 3/17/2009 8:23:07 PM

and n – 2 fractional digits giving a precision of 1/B n–2. For a binary system, B is 2 and this numeric format has
a range from –1.9999… to +1.9999… (approximately, from –2 to +2). If x1 and x2 are in D(2, 1, n – 2), then
1. x1 + x2 should be < 2 if x1 < 1 and x2 < 1.
2. x1 – x2 should be < ±2 if x1 < 1 and x2 < 1.
3. x1 × x2 < 1 if both x1 and x2 are less than 1.
EXAMPLE 12.5 Consider a 16-bit system with D(2, 1, 14) numeric format. This implies a sign bit
also. The characteristic features of this 16-bit numeric format are
1. Range: (1000 0000 0000 0001)2 to (0111 1111 1111 1111)2 or (8001)16 to
(7FFF)16 or (–1.9999)10 to (+1.9999)10.
2. If x1 and x2 are in D(2, 1, 14), then x1 and x2 < (0011 1111 1111 1111)2 or
(3FFF)16 or (0.9999)10 so that x1 + x2 < 2, x1 − x2 < ±2 and x1 × x2 < 1 are
satisfied.
Thus, from the above example, it is evident that if the variables in the application are ensured to have a value
less than unity under all conditions, then all values including the result will be in the range of D(2, 1, 14).
To achieve this, the variables are divided by the worst case maximum value that the variables can assume in
a given applications. This will ensure that the variables will all be per unit values or normalized values. This
is called normalization. For the normalized system, 2 f will represent 1 or 100%. In the above example, 214
will represent 1.
Let x be a variable used within the digital processor. If xmax is the worst case maximum value that the
variable can assume, then
x
xn = (12.4)
x max
where xn is the normalized value of x. Consider a situation wherein an analog variable is to be discretized and
taken into the digital domain. The block schematic is shown in Figure 12.2. Let the range of analog input
to the analog-to-digital converter (ADC) be 0–VAD. The maximum possible input voltage to the ADC is
VAD. Define a base value, xbase = 1 per unit (pu) or 100%. The value of xbase is in the range 0 < xbase ≤ VAD.
The value xbase < VAD is used in applications wherein some portion of the signal amplitude space is
needed for representing the overloads. This is achieved by sacrificing precision under normal operating con-
ditions. If xbase = VAD, then it implies that there is no overload. The maximum value of the signal that can be
represented is the rated value or the base value itself.
The output of the ADC is a binary word of na-bits length. For DC signals, one can use all the na-bits to
represent the magnitude of the analog signal x. After ADC, the digital value is called xAD(k) at time instant
kT where T is the sample time. The output of the ADC is scaled by K to obtain the n-bit normalized data
xn(k) in D(2, i, f ) numeric format.
The relationship between the input and the output of the ADC is as follows:
1. 2na in digital domain at the output of the ADC represents VAD value in the analog domain at the input
of the ADC.
2. xAD(k) in the digital domain at the output of the ADC represents a value x AD (k ) ⋅ (V AD / 2na ) which is
the quantized value of the analog variable x.
Chapter 12.indd 705 3/17/2009 8:23:07 PM

Analog domain Digital domain
x ADC x AD(k) Scaling x n(k)

n a-bits K n -bits
(data)
x
V AD
0 t
Figure 12.2 Scaling in analog-to-digital domain conversion of variables.
The ADC output bits na is usually lesser than the number of data bits n. Therefore, the output of the ADC
is padded with (n – na) zero MSBs so that the ADC output is converted to n-bit data.
If xbase is the per unit value or rated value or the nominal value of the variable x, then to normalize,
divide the variable with the base value and multiply with 2 f to convert to D(2, i, f ) numeric format:
x
xn =
x base
x AD (k ) ⋅ (V AD / 2na ) f
x n (k ) = x n ⋅ 2 f = ⋅2 (12.5)
x base
From Eq. (12.5) it can be observed that when x = xbase, then xn = 2 f which represent 1 or 100%. Equation
(12.5) can be re-written as
⎛V 2f ⎞
xn (k ) = x AD (k ) ⎜ AD ⋅ n ⎟ = x AD (k ) ⋅ K (12.6)
⎝ x base 2 ⎠
a
The factor K in Eq. (12.6) is called the scaling factor.

For applications where x is an AC waveform, then x will swing both positive and negative. However,
the ADC in most cases will accept only unidirectional or DC voltages. Therefore, the AC waveform, x is
level shifted by VAD/2 such that the peak-to-peak amplitude of x is within VAD. In such cases, the input
to the ADC is x + (V AD / 2) . The ADC output xAD(k) will also be level shifted by 2na −1 which represents
VAD/2. In this case, one bit of the ADC, that is, the MSB is used as a sign bit. 2na −1 is subtracted from
xAD(k) to represent x. However, it should be noted that as one of the ADC is used as sign bit, only na – 1
bits are available for representing the amplitude of x. From Eq. (12.6), the normalized quantized value
of x is
Chapter 12.indd 706 3/17/2009 8:23:07 PM

⎛V / 2 2 f ⎞
xn (k ) = [ x AD (k ) − 2na −1 ] ⎜ AD ⋅ n −1 ⎟
⎝ x base 2 ⎠
a
⎛V 2 ⎞
f
= [ x AD (k ) − 2na −1 ] ⎜ AD ⋅ n ⎟ (12.7)
⎝ x base 2 ⎠
a
= [ x AD (k ) − 2na −1 ] K
Problem 12.3
Consider a 10-bit ADC that has an input range of 5 V. For 16-bit data to be represented in D(2, 3, 12)
numeric format, find the scaling factor K for
(a) DC input where xbase = 3 V
(b) AC input where xbase = 2.5 V is the peak value of the AC waveform
Solution
(a) For a DC input, VAD = 5 V, na = 10, f = 12, n = 16 and xbase = 3 V. From Eq. (12.6) one obtains
⎛V 2f ⎞ ⎛ 5 212 ⎞
xn (k ) = x AD (k ) ⎜ AD ⋅ n ⎟ = x AD (k ) ⎜ ⋅ 10 ⎟ = x AD (k ) ⋅ 6.67
⎝ x base 2 ⎠ ⎝3 2 ⎠
a
K = (6.67)10
Atleast three bits are required to represent the integer part of K; therefore K can be expressed in D(2,
3, 12) numeric format.
(b) For an AC input, VAD = 5 V, na = 10, f = 12, n = 16 and xbase = 2.5 V. From Eq. (12.7) one obtains
⎛ 5 212 ⎞
xn (k ) = ( x AD (k ) − 29 ) ⎜ ⋅ 10 ⎟ = ( x AD (k ) − 2 )⋅ 8
9
⎝ 2.5 2 ⎠
K = (8)10
Atleast four bits are required to represent the integer part of K; therefore K can be expressed in D(2, 4,
11) or D(2, 5 10) numeric format.
For power electronic applications, generally, the voltage and current overshoots up to twice rated are
possible in most cases for short times. This implies that the instantaneous powers are up to four times rated.
This implies that at least three bits are required to represent power up to four times rated. Thus, three integer
bits are normally used which can represent upto seven times the nominal value. One bit is needed for sign
bit. Therefore, for a 16-bit digital system, D(2, 3, 12) numeric format is most appropriate for the power
electronic applications.
Multiplication Algorithm
In some microcontrollers, only addition and subtraction are provided. In that case, a routine for multiplica-
tion has to be used. The multiplication uses two variables as inputs called the multiplicand and the multi-
plier. Let x1 in D(B, i1, f1) be the multiplier with m digits and x2 in D(B, i2, f2) be the multiplicand with n
digits. Then the multiplier operation is shown in the following template.
Chapter 12.indd 707 3/17/2009 8:23:07 PM

×
1 0
B n
B B Bm B0
x 2 − multiplicand x1 − multiplier
The digits of x1 are multiplied with the digits of x2 one by one. If the digit number i of x1 is multiplied with
digit number j of x2, then the digit product is added to the previous product after multiplying by Bi + j or
added to the previous product after shifting left by i + j digits. Thus,
x p = x1 × x 2 = x prev + (digit product ) × B i + j
The above equation has to be applied till all the digits of the multiplier have been multiplied with all the
digits of the multiplicand. Thus, the generic algorithm for multiplication is
set xprev = 0
Loop for i = 0 to m
Loop for j = 0 to n
xp = xprev + (dj × di)*Bi + j
end loop j
end loop i
Problem 12.4
Multiply (34)10 and (23)10
Solution
Step 1: xprev = 0; xp = 0 + (4 × 3)*100+0 = 12

Step 2: xprev = 12; xp =12 + (3 × 3)*100+1 = 102
Step 3: xprev = 102; xp = 102 + (4 × 2)*101+0 = 182
Step 4: xprev = 222; xp = 182 + (3 × 2)*101+1 = 782
If the base B = 2, then multiplication consists of only shift left and add. The above algorithm for binary
multiplication becomes
set xprev = 0
Loop for i = 0 to m
Loop for j = 0 to n
xp = xprev + (dj × di) * 2i + j
end loop j
end loop i
Chapter 12.indd 708 3/17/2009 8:23:08 PM

2 i + j is nothing but shift left operation. The digit product can be either 0 or 1 in the case of the binary system.
Therefore, only if the digit product is 1 then, shift left by (i + j) and add to the xprev.
12.4 Arithmetic Algorithms
M ost digital processors have only three primary arithmetic operators: addition, subtraction and multi-
plication. If multiplication instruction is not available, then a multiplication routine has to be used as
discussed in the previous section. Some advanced processors also have hardwired division that is in-built.
This section discusses the methods of computing the functions in an iterative manner using only these three
primary arithmetic operators.
Given a continuous function y = f (x), it is required to compute the value of the function for a given
value of the parameter x. If the function y = f (x) is difficult to compute, it can be re-written in implicit form
as F(x, y) = 0.
Let yk be the approximate value of the function. If the error in the function estimate is Δy then
∂F
F ( x , yk ) = Δy ⋅
∂y x , yk
∂F
= ( yk − yk +1 ) ⋅ (12.8)
∂y x , yk
and the convergence is ensured if ∂F/∂y and ∂2F/∂y 2 have same signs in the interval under consideration
wherein the root of y lies. Generally, in iterative algorithms, the initial value y0 is arbitrarily chosen such that
it is as close as possible to the final value of y. The process of iteration is continued till yk and yk–1 coincide or
differ by a very small value ∈.
From Eq. (12.8), the general iterative equation for a continuous function y = f (x) is
F
yk +1 = yk − (12.9)
(∂F / ∂y ) x , y
k
The advantage of the iterative algorithm is that the operations are of the same type for each iteration and
therefore are easy to program. It should, however, be noted that F(x, y) = 0 for the function y = f (x) may be
realized in many ways. This fact should be utilized to obtain an implicit function that leads to rapid
convergence.
Reciprocal
Let y = 1/ x where x > 0. Set
1
F (x, y) ≡ x − =0
y
∂F 1
= F′= 2
∂y y
⎛ x − (1 / yk ) ⎞
yk +1 = yk − ⎜ 2 ⎟
⎝ 1 / yk ⎠
Chapter 12.indd 709 3/17/2009 8:23:08 PM

Re-arranging the above, the iterative solution for the reciprocal of x is given as
yk +1 = yk ( 2 − x ⋅ yk ) where k = 0, 1, 2, … (12.10)
Equation (12.10) is the iterative equation for computing the reciprocal of x. The initial value y0 is selected as
follows.
1. Let x be an n-digit binary number in D(2, i, f ) numeric format. The binary number x can be written as
x = 2m ⋅ xm where m is an integer and (1 / 2) ≤ xm < 1
2. Check the bit number of the MSB of x. Let this be nmsb. Then
m = nmsb + 1
−m
3. Set y0 = 2 in order to have rapid convergence.
EXAMPLE 12.6 Consider a 16-bit binary number x = (3)10 in D(2, 3, 12) numeric format:
x = 0011.000000000000
(Note that the binary point is a virtual point.)
Now nmsb = 1 (corresponding to 21). Therefore
m = nmsb + 1 = 2
y0 = 2−2 = 0000.010000000000 = (0.25)10

y1 = 0.25(2 − 3 ⋅ 0.25) = (0.3125)10 = 0000.010100000000
y 2 = 0.3125( 2 − 3 ⋅ 0.3125) = ( 0.33203)10 = 0000.010101001111
…
The accuracy will improve with iterations and converge towards y = (0.33333)10
or 0000.0101 0101 0101.
Alternate Algorithm
The computation of the reciprocal can be considered as a negative feedback system. The converging portion
of the iteration is the transient portion and the steady-state portion is the converged value of the function.
Figure 12.3 gives the control block schematic of the reciprocal estimation.
1 y
+
− e K ∫
xy
X
Figure 12.3 Reciprocal algorithm based on control principles.
Chapter 12.indd 710 3/17/2009 8:23:08 PM

The function F(x, y) is re-defined as

F (x, y) ≡ 1− x ⋅ y = 0 (12.11)
From Figure 12.3 and Eq. (12.11), it can be observed that the reference is unity or 1. The output y is multi-
plied with x and given as feedback. The error is passed through a gain K which is passed through an integra-
tor to obtain the output y. The integrator ensures that the steady-state error is zero in which case xy = 1 or
the output y = 1/x. The integrator can be a digital integrator like the trapezoidal integrator with the gain K
tuned to achieve fast transient response.
Square Root
Let y = x where x > 0. Set
F (x, y) ≡ y 2 − x = 0
∂F
= F′ = 2y
∂y
⎛ y2 − x ⎞
yk +1 = yk − ⎜ k ⎟
⎝ 2 yk ⎠
Re-arranging the above, the iterative solution for the square root of x is given as
1⎛ x ⎞
yk +1 = ⎜ yk + ⎟ where k = 0, 1, 2, … (12.12)
2⎝ yk ⎠
Equation (12.12) is the iterative equation for computing the square root of x. The initial value y0 is selected
as follows:
m = nmsb + 1
3. Set y0 = 2trunc ( m / 2 ) in order to have rapid convergence, where trunc(m/2) truncates the number m/2 to
the lower integer value.
EXAMPLE 12.7 Consider a 16-bit binary number x = (5)10 in D(2, 3, 12) numeric format:
x = 0101.000000000000
Here nmsb = 2 (corresponding to 22). Therefore
m = nmsb + 1 = 3
y0 = 2trunc (3/ 2 ) = 21 = 0010.000000000000 = ( 2)10

y1 = 0.5( 2 + 5 / 2) = ( 2.25)10 = 0010.010000000000
y 2 = 0.5( 2.25 + 5 / 2.25) = ( 2.2361)10 = 0010.001111000111
…
Chapter 12.indd 711 3/17/2009 8:23:08 PM

Alternate Algorithm 1
The square root iteration algorithm given in Eq. (12.12) requires the reciprocal algorithm as the reciprocal
of yk has to be evaluated first before evaluating the square root. This is therefore not a very efficient algorithm
in terms of computational speed. Alternately, set
x
F (x, y) ≡ −1 = 0
y2
∂F 2x
= F′= − 3
∂y y
⎛ ( x / yk2 ) − 1 ⎞
yk +1 = yk + ⎜ 3 ⎟
⎝ 2 x / yk ⎠
Re-arranging the above, the iterative solution for the square root of x is given as
yk ⎛ yk2 ⎞
yk +1 = ⎜ 3 − ⎟ where k = 0, 1, 2, … (12.13)
2 ⎝ x ⎠
In the above equation, it can be observed that the reciprocal algorithm is still needed for x. However, the
reciprocal of x needs to be calculated only once whereas in the case of Eq. (12.12), the reciprocal of yk needs
to be calculated every iteration.
Alternate Algorithm 2
The computation of the square root can be considered as a negative feedback system. Figure 12.4 gives the
control block schematic of square root estimation.
The function F(x, y) is re-defined as
F (x, y) ≡ x − y 2 = 0 (12.14)
From Figure 12.4 and Eq. (12.14), it can be observed that the reference is x. The output y is multiplied with
itself and fed back. The error is passed through a gain K which is passed through an integrator to obtain the
output y. The integrator ensures that the steady-state error is zero in which case y2 = x or the output y = x .
The integrator can be a digital integrator like the trapezoidal integrator with the gain K tuned to achieve fast
transient response.
x
+
− e
K ∫ y
X
y2
Figure 12.4 Square root algorithm based on control principles.
Chapter 12.indd 712 3/17/2009 8:23:08 PM

Reciprocal of Square Root

Let y = 1/ x where x > 0.The function can be re-written as y = 1/ x . Applying the iteration algorithm
for square root as given in Eq. (12.13), one obtains
yk
yk +1 = (3 − x ⋅ yk2 ) where k = 0, 1, 2, … (12.15)
2
The initial value y0 is selected as follows:
m = nmsb + 1
− trunc ( m / 2 )
3. Set y0 = 2 in order to have rapid convergence, where trunc(m/2) truncates the number m/2
to the lower integer value.
Extension to Square Root Calculation

The above algorithm can be used to compute the square root of a number x using the reciprocal process.
Let y = x where x > 0. This function can be re-written as y = x = x ⋅ 1/ x where x > 0. Using
Eq. (12.15), 1/ x can be computed without any reciprocal operation. After computing 1/ x , multiply
the resulting output value with x to obtain the square root of x.
Hypotenuse
Let a and b be the sides of a right-angled triangle and let h be its hypotenuse. Then by Pythagoras theorem,
h = a2 + b2 (12.16)
Equation (12.16) can be computed using one of the square root algorithms. Alternately, a very fast open-
loop algorithm for calculating square roots can be used. It is based on the fact that for a triangle, the hypote-
nuse h is greater than either of the other two sides a and b. Therefore,
h = max( a, b ) + k ⋅ min( a, b ) (12.17)
where 0 ≤ k ≤ 1 . Here max(a, b) is the maximum value among a and b and min(a, b) is the minimum value
among a and b. Let
x = max( a, b ) (12.18)
y = min( a, b ) (12.19)
Substituting Eqs. (12.18) and (12.19) into Eq. (12.17), one obtains
h = x + ky (12.20)
Chapter 12.indd 713 3/17/2009 8:23:09 PM

From Eq. (12.14), one has

h = x + ky ≈ x 2 + y 2 (12.21)
The error in percent between the approximate algorithm and the actual value obtained by the Pythagoras
theorem is
x 2 + y 2 − ( x + ky )
e= × 100 (12.22)
x2 + y2
Using a computer program, x and y can be normalized and k can be varied from 0 to 1. The minimum error
obtainable is at k = 0.34 where the error is guaranteed to be less than 5.55%. However, from an implemen-
tation point of view, k = 0.5 is very desirable as scaling by 0.5 can be achieved by simply shifting the data
right by one bit. But at k = 0.5, it can be guaranteed that error will be less than 11.8%. If this maximum
possible error is acceptable, then this would be a very fast hypotenuse computation algorithm.
Polynomial
Most of the digital filters are polynomials. Therefore, evaluating polynomials for a specified value of the
parameter x is very common in digital processing applications. Consider a polynomial of degree n. Then,
P ( x ) = a0 x n + a1 x n −1 + + an −1 x + an (12.23)
where ai are real coefficients. If it is required to evaluate the polynomial P(x) at x = a, then
P (α ) = a0α n + a1α n −1 + + an −1α + an

(12.24)
= (...((( a0α + a1 ) ⋅ α + a2 ) ⋅ α + a3 )α + + an −1 )α + an
The numbers are computed successively as follows:

b0 = a0
b1 = a1 + b0α
b2 = a2 + b1α

bn = an + bn −1α
Then
P(a) = bn
Sine and Cosine

The sine and cosine of an angle are used in many algorithms in power electronics. Therefore, computing sine
or cosine of an angle is important. If the digital processor contains ample data memory space, then the best
approach from the point of view of speed is to pre-compute the sine values for angles varying from 0° to
360° and store them in the data memory space. Two pointers may be used: one for the sine and the other
for the cosine. The two pointers point to the sine table such that they are displaced in angle corresponding
to 90°. This means that if the sine pointer points to the memory location corresponding to sin(a), then the
cosine pointer points to the memory location corresponding to sin(a + 90). Alternately, one may compute
the sine and cosine values by using the series expansion method based on the following discussion.
Chapter 12.indd 714 3/17/2009 8:23:09 PM

The sine and cosine may be calculated by their series expansions. Let a and b be the angles such that
0 ≤ α ≤ π / 2 and β = (π / 2) − α . If 0 ≤ α ≤ π / 4 , then π / 4 ≤ β ≤ π / 2 . The sine and the cosine are
given as
∞
α 2n +1
cos β = sin α = ∑ ( −1)
n
(12.25)
n =0 (2n + 1)!
If π / 4 ≤ α ≤ π / 2 then 0 ≤ β ≤ π / 4 . The sine and the cosine are given as
∞
β 2n
sin α = cos β = ∑ ( −1)
n
(12.26)
n =0 (2n )!
Equation (12.25) can be computed by the summation of
cos β = sin α = s1 + s 2 + ... + sk + sk +1 + ... (12.27)
where sk are found by the following relationships:

s1 = α
α2
sk +1 = − sk (k = 1, 2, …)
2k ( 2k + 1)
The summation process can be terminated when sk ≤ ∈, where ∈ is a small number indicating the residual
error.
Similarly, Eq. (12.26) can also be computed by the summation of
sin α = cos β = c1 + c 2 + ... + ck + ck +1 + ... (12.28)
where ck are found by the following relationships:

c1 = 1
β2
ck +1 = − ck (k = 1, 2, …)
2k ( 2k − 1)
The summation process can be terminated when ck ≤ ∈, where ∈ is a small number indicating the residual
error.
Exponential
The exponential function e x can be expanded in series form as
x2 xn
e x = 1+ x + + + + (12.29)
2! n!
where −∞ < x < ∞. The number x can be written as
x=I+F
where I is the integer part and F is the fractional part wherein 0 ≤ F < 1 . Therefore, e k can be expressed as
ek = e I.e F (12.30)
Chapter 12.indd 715 3/17/2009 8:23:09 PM

In Eq. (12.30), the first factor is expressed as

e I = e ⋅ e ⋅ e e I times if I ≥ 0
or
1 1 1 1
eI = ⋅ ⋅ I times if I < 0
e e e e
where e = 2.7182818… and 1/e = 0.3678794….
In Eq. (12.30), the second factor is expressed as
F 2 F 3 ... ∞ F n
e F = 1+ F + + + =∑ where F < 0
2! 3! n =0 n !
This can be computed by summation of the series

e F = u1 + u2 + ... + uk + uk +1 + ...
where
u0 = 1
F
uk +1 = uk (k = 1, 2, ...)
k +1
The summation process can be terminated when uk ≤ ∈, where ∈ is a small number indicating the residual
error.
Logarithm
The series expansion of the natural logarithm of numbers is given as
x 2 x 3 x 4 ... xk
ln(1 + x ) = x − + − + + (−1)k −1 + ... (12.31)
2 3 4 k
where −1 < x ≤ 1 or 0 < 1 + x ≤ 2 . Equation (12.31) is very limited in its application as the range of num-
bers x is small and for values of x whose absolute value is close to unity, the convergence is very slow. To solve
this problem of range and convergence, the following modification to the series is performed. Substituting x
with –x in Eq. (12.31), one obtains
x 2 x 3 x 4 ... x k ...
ln(1 − x ) = − x − − − − − − (12.32)
2 3 4 k
Subtracting Eq. (12.32) from Eq. (12.31), one obtains
⎛ 1− x ⎞ ⎛ x3 x5 ⎞
ln ⎜ ⎟ = −2 ⎜ x + + + ... ⎟ (12.33)
⎝ 1+ x ⎠ ⎝ 3 5 ⎠
Set z = (1 − x ) / (1 + x ) . This means x = (1 − z ) / (1 + z ) . Substituting the value of x in Eq. (12.33), one has
⎡⎛ 1 − z ⎞ 1 ⎛ 1 − z ⎞3 1 ⎛ 1 − z ⎞5 ⎤
ln z = −2 ⎢⎜ ⎟+ ⎜ ⎟ + ⎜ ⎟ + ...⎥ (12.34)
⎢⎣⎝ 1 + z ⎠ 3 ⎝ 1 + z ⎠ 5 ⎝ 1 + z ⎠ ⎥⎦
Chapter 12.indd 716 3/17/2009 8:23:10 PM

For x in the range −1 < x ≤ 1 , z is in the range 0 < z < ∞. Equation (12.34) can now handle the full range
of the positive real numbers.
Let x be an n-digit binary number in D(2, i, f ) numeric format. The binary number x can be written as
1
x = 2m ⋅ xm where m is an integer and ≤ xm < 1
2
From Eq. (12.34),
ln x = m ln 2 + m ln xm
⎡⎛ 1 − x ⎞ 1 ⎛ 1 − x ⎞3 1 ⎛ 1 − x ⎞5 ⎤
ln x = m ln 2 − 2 ⎢⎜ m
⎟+ ⎜
m
⎟ + ⎜
m
⎟ + ...⎥ (12.35)
⎢⎣⎝ 1 + xm ⎠ 3 ⎝ 1 + xm ⎠ 5 ⎝ 1 + xm ⎠ ⎥⎦
Let α = (1 − xm ) / (1 + xm ) then 0 < α ≤ 1 / 3 and
⎛ α 3 α 5 ... α 2k −1 ... ⎞
ln x = m ln 2 − 2 ⎜ α + + + + + ⎟
⎝ 3 5 2k − 1 ⎠
This equation can be computed by series summation as
ln x = m ln 2 − 2 ( u1 + u2 + u3 + ... + uk + ...)
where
ln2 = 0.693147…
u1 = α
( 2k − 1)
uk +1 = uk (k = 1, 2, …)
( 2k + 1)
The summation process can be terminated when uk ≤ ∈, where ∈ is a small number indicating the residual
error.
Implementation Examples
This section will discuss some application examples that are common in power electronic systems. These
application examples are discussed in an algorithmic manner indicating the equations and the sequence in
which the equations are arranged for digital computation. The arithmetic operations for computing the
equations are based on the algorithms discussed in the previous sections.
PI Controller
The PI controller is discussed in detail in Chapter 10. The block schematic of the PI controller is shown in
Figure 12.5. It consists of a comparator that compares the reference xref and the feedback signal xf . The
error epi is fed to the proportional and the integral parts of the controller as shown in the figure. The inte-
grator of the controller is a limited integrator to prevent windup. The output of the PI part vcont is added
with a feed-forward signal if present and passed through a clamper to limit the controller output Vc to
within the band bounded by Vmax and Vmin. Together with the Figure 12.5, the algorithm given below is
self-explanatory.
Chapter 12.indd 717 3/17/2009 8:23:10 PM

Limited
V int
Ki ∫
Integrator + V max
+
+ +
x ref − e pi V cont + Vc
+ V min
Kp Clamp
xf x ff
Feedback Feed-forward
Figure 12.5 Implementational block schematic of a PI controller.
[vc] = pi_controller(xref, xf, xff)

Local Variables:
epi, vint, vcont
Constants:
//*** PI regulators parameters
Ki, Kp, Kclamp
//*** vc clamp limits
Vmin, Vmax
{
epi = xref − xf;
//integrator portion
if (vc < Vmax)& (vc > Vmin)
Kclamp=1;
else
Kclamp=0;
end
vint = vint + Kclamp*Ki*epi; //integrator output with anti-windup
//proportional plus integral portion
vcont = vint + Kp*epi;
//add feedforward term and clamp output
if (vcont + xff)>= Vmax
vc = Vmax;
elseif (vcont + xff) <= Vmin
vc = Vmin;
else
vc = vcont + xff;
end
}
Chapter 12.indd 718 3/17/2009 8:23:11 PM

Theta
Theta
(2p ) FF
7F
(0) 00
00 FF
7F
Vector table index
Figure 12.6 Sine table and the angle relationship.
Sine and Cosine

The sine and cosine of an angle are obtained here by browsing through a pre-computed sine table. Assume
that the data is 16-bit data, but say only 256 data words of memory storage are available in the digital pro-
cessor. To create a sine table, generate sine values from 0 to 2π in D(2, 0, 15) with one bit being used for
sign bit for the 2’s complement number. As the sine values are less than unity, no integer bits are required.
From the generated table (containing 216 16-bit data), extract 256 16-bit data by selecting only those
wherein the most significant 8 bits are distinct. This would form the sine table. The starting address of the
sine table (*sin_table) is also passed to the function along with the 16-bit angle data to compute the sines
and cosines.
The algorithm is simple and can be understood clearly from Figure 12.6. The angle Theta can take values
from 0 to 2π , that is from (0000)16 to (FFFF)16. The data table may be considered as a vector wherein the
length varies from 0 to 255, that is, (0000)16 to (00FF)16. In order to map the angle to the data table vector,
an angle of Theta = (00xx)16 maps to the sine table vector element (0000)16 and the angle of Theta =
(FFxx)16 maps to the sine table vector element (00FF)16. This is a many-to-one mapping where “x” is a
“don’t care” term. The starting address of the sine table, “*sin_table”, is used to offset the sine table vector
index. For obtaining the cosine of the angle Theta, an additional offset corresponding to one-fourth, the
length of the sine table has to be included in the index to obtain a 90° phase shift.
[sinTheta, cosTheta] = sincos(Theta, *sin_table)

Local Variables:
index, sine_table
{
//obtaining sine of the angle
index=MSByte(Theta);//Most signifcant byte of Theta is index of
//the sine table vector
index = index + *sin_table;//offset index with starting address of
// sine table
sinTheta = sine_table(index);
//obtaining cosine of the angle
Chapter 12.indd 719 3/17/2009 8:23:11 PM

index = MSByte(Theta);//Most signifcant byte of Theta is index of

//the sine table vector
index = index + *sin_table + (0040)16;//offset index with starting
//address of sine table.
//(0040)16 additional offset corresponds
//to a 90o phase shift.
cosTheta = sine_table(index);
}
Pulse-Width Modulation (PWM)

The PWM has been discussed in Chapter 6. However, this section presents an algorithm for imple-
menting the single-phase PWM. The input to the algorithm is the reference signal, Vref . The input ref-
erence signal is compared notionally with a triangular waveshape. Due to the linear structure of a
triangular waveform, the pulse widths are proportional to the voltages at the instant of comparison
with the triangular carrier. Therefore, if Vr is the voltage of the reference signal at the instant of com-
parison, then
⎛V ⎞T
tr = ⎜ r ⎟ s (12.36)
⎝ Vdc ⎠ 2
where Vdc is the maximum attainable peak voltage of the phase waveforms that corresponds to modulation
index of 1 and Ts is the switching time period. The value of Vr may be positive or negative depending on the
nature of the reference waveshape. If the reference waveshape is a DC signal then Vr is always positive and
hence tr is also positive. If the reference waveshape is an AC signal like in single-phase sinusoidal PWM,
then Vr can be either positive or negative. Both these cases will be discussed.
DC Reference
Figure 12.7 shows the DC reference wave compared with a triangular carrier. This is generally encountered
in DC–DC converters. The reference waveform is sampled at an arbitrary comparison instant. The expanded
time scale at the sample instant shows the comparison that spans a carrier period or switching period Ts. At
the comparison instant the sampled value is Vr. During the period Ts, the sampled value is compared with
the triangular waveform as shown:
T0 Ts
= − tr or T0 = Ts − 2t r (12.37)
2 2
where T0 is the time when the PWM output is zero.

Over-Modulation The maximum value of the converter output voltage is limited by the DC bus voltage.
To maintain linearity, the reference amplitude is maintained within the DC bus voltage. However, during
transients, the reference amplitudes may go beyond the DC bus. Such a situation is called over-modulation.
During over-modulation, tr becomes greater than Ts and T0 becomes negative. To compensate for this, the
times are multiplied by Ts/tr to clamp the average output to the DC bus. This is a simple over-modulation
technique sufficient for many applications.
Chapter 12.indd 720 3/17/2009 8:23:11 PM

T s /2 T s /2 Triangular
carrier
vr
Vr
0 t
Ts
Sample tr
instant T o /2
Time reference
Figure 12.7 PWM generation with DC reference.
The generalized algorithm for the PWM generation with DC reference is as follows:
[tr] = pwm(Vr)
Local Variables:
tr, toffset, Ts, Vdc, T0
{
tr=Ts*Vr/Vdc/2;
T0 = Ts – 2*tr;
//Overmodulation
if (T0 < 0),
tr = Ts/2;
end
}
The value of tr obtained from this algorithm should be pushed into a compare register of the digital processor
that makes comparison with the triangular waveform. The triangular carrier waveform is generated by program-
ming one of the internal general-purpose timers of the digital processor as an up–down counter to emulate a tri-
angle waveform having a period Ts. The output of the compare registers will produce the PWM waveforms.
AC Reference
Figure 12.8 shows an AC reference waveform. This is generally encountered in single-phase inverter applica-
tions. The reference waveform vr and an inverted vr as shown in Figure 12.8 are used to generate the PWM
for two-level two-arm inverters. The reference vr is applied to arm “a” of the inverter and is denoted as vra
and the inverted reference is applied to arm “b” of the inverter and is denoted as vrb. It is required to obtain
the ON-times taon and tbon for the two inverter arms. If Va and Vb are the sampled voltages of vra and vrb at a
specific sample instant, then
Chapter 12.indd 721 3/17/2009 8:23:11 PM

⎛V ⎞
t a = ⎜ a ⎟ Ts
⎝ Vdc ⎠ (12.38)
⎛V ⎞
t b = ⎜ b ⎟ Ts (12.39)
⎝ Vdc ⎠
index of 1 and Ts is the switching time period. The value tmax = max(ta, tb) gives the maximum value among
ta and tb; tmin = min(ta, tb) gives the minimum value among ta and tb.
Figure 12.8 shows the steps involved in generating the PWM waveform. The two outputs of phase refer-
ence waveforms are sampled at an arbitrary comparison instant. This comparison instant spans a period Ts.
The comparison instant is shown expanded wherein the two-phase sampled values are Va and Vb, as shown.
During the period Ts, the sampled values are compared with the triangular waveform as shown. At that
instant, the waveshape having the largest voltage will have a time tmax. In the figure, Va is shown to have a
maximum value and therefore tmax = ta in this case. As Vb is the minimum value at this instant, tmin = tb.
From Figure 12.8, it can be observed that tmax which is measured right from the central reference is
positive. tmin is measured left from the central reference and is a negative number as Vb is negative during the
sampled instant. The time value tmin cannot be negative. Therefore an offset has to be provided to the central
reference line so that the time values are positive. The new time reference is shifted from the central line to
the line at the beginning of the Ts interval as shown in the figure. The shift from the central line to the new
reference line is the time offset which is given as
T0
t offset = t min +
2
T0
t offset = −t min + (12.40)
2
where T0 is the time duration when the PWM output across the inverter bridge is zero in a Ts period.
Over-Modulation The maximum value of the inverter output voltage is limited by the DC bus voltage.
The reference amplitude is maintained within the DC bus voltage. However, during transients, the peak
reference amplitude may go beyond the DC bus value. Such a situation is called over-modulation. During
over-modulation, tmax – tmin becomes greater than Ts and T0 becomes negative. To compensate for this, the
ON-times are scaled to clamp the peak output to the DC bus. This is a simple over-modulation technique
sufficient for many applications.
The generalized algorithm for the PWM generation for AC reference is as follows.
[taon, tbon] = spwm(Va, Vb)

Local Variables:
tmin, tmax, toffset, Ts, Vdc, Teff, T0
{
//fnd ta, tb
ta=Ts*Va/Vdc;
tb=Ts*Vb/Vdc;
//fnd tmax and tmin
tmax = ta;
Chapter 12.indd 722 3/17/2009 8:23:12 PM

if (tb > tmax)

tmax = tb;
end
tmin = ta;
if (tb < tmin)
tmin = tb;
end
Teff = tmax – tmin;
T0 = Ts – Teff;
Toffset = (T0/2) − tmin; //offset time from the origin.
//the on-time for each of the PWM phases is given by:
taon = ta + toffset;
tbon = tb + toffset;
//Overmodulation
if (T0 < 0),
toffset = −tmin*Ts/Teff;
taon = ta*Ts/Teff + toffset;
tbon = tb*Ts/Teff + toffset;
end
}
Vr Ts Ts
V dc
Va
0 t
Vb
−V dc
Central
line
t min t max
Sample
instant
New time
reference
To
t a(on)
2
a arm
t b(on)
b arm
To T eff To
2 2
Figure 12.8 PWM generation with AC reference.
Chapter 12.indd 723 3/17/2009 8:23:12 PM

The values taon and tbon should be pushed into a compare register of the digital processor that makes
comparison with a triangular waveform. The triangular carrier waveform is generated by programming
one of the internal general-purpose timers of the digital processor as an up–down counter to emulate
a triangle waveform having a period 2Ts. The output of the compare registers will produce the PWM
waveforms.
Space Vector PWM

The space vector PWM and its generation have been discussed in Chapter 6. However, this section presents
an algorithm for implementing the space vector PWM that is much simpler and faster. The inputs to the
algorithm are the three-phase reference sine waves, varef, vbref and vcref. The input reference waveshapes are
compared notionally with a triangular waveshape. Due to the linear structure of a triangular waveform, the
pulse widths are proportional to the voltages at the instant of comparison with the triangular carrier.
Therefore, if Va, Vb and Vc are the voltages of the three-phase waveforms at the instant of comparison with
the triangular wave, then
⎛V ⎞
ta = ⎜ a ⎟ Ts (12.41)
⎝ Vdc ⎠
⎛V ⎞
tb = ⎜ b ⎟ Ts (12.42)
⎝ Vdc ⎠
⎛V ⎞
tc = ⎜ c ⎟ Ts (12.43)
⎝ Vdc ⎠
index of 1 and Ts is the switching time period.
Figure 12.9 shows the three-phase waveform and the steps in the generation of the corresponding timing
for each of the PWM time intervals T0, T1 and T2, respectively, where T0 is the time spent in zero space
vector, T1 is the time spent in the primary space vector of the sector and T2 is the time spent in the second-
ary space vector of the sector. The notations T0, T1 and T2 connote the same meanings as explained in
Chapter 6 in the discussion of space vector PWM. The value tmax = max(ta, tb, tc) gives the maximum value
among ta, tb and tc and tmin = min(ta, tb, tc) gives the minimum value among ta, tb and tc.
Figure 12.9 shows the steps involved in generating the PWM waveform. The three-phase reference
waveform is sampled at an arbitrary comparison instant. This comparison instant spans a period Ts. The
comparison instant is shown expanded wherein the three-phase sampled values are Va, Vb and Vc as shown.
During the period Ts, the sampled values are compared with the triangular waveform as shown. At that
instant, the waveshape having the largest voltage will have a time tmax. In the figure, Vb is shown to have a
maximum value and therefore tmax = tb in this case. As Vc is the minimum value at this instant, tmin = tc. Also,
Va is in between Vb and Vc and therefore, tmid = ta.
From Figure 12.9, it can be observed that tmax which is measured right from the central reference is
positive. Both tmin and tmid are measured left from the central reference and are negative as Va and Vc are
negative during the sampled instant. The time values tmin and tmid cannot be negative values. Therefore an
offset has to be provided to the central reference line so that the time values are positive. The new time refer-
ence is shifted from the central line to the line at the beginning of the Ts interval as shown in the figure. The
shift from the central line to the new reference line is the time offset and is given as
Chapter 12.indd 724 3/17/2009 8:23:12 PM

V aref V bref V cref Vb
0 0 Triangle ramp
0 Va
Vc
Ts
Comparison
instant Expanded Central
comparison reference
instant
t max
t mid
t min
New time
reference
a - phase
t
b - phase
t
c - phase
t
T0 T2 T1 T0
2 2
t bon
t aon
t con
Figure 12.9 Generation of SPWM waveform.
T0
t offset = t min +
2
T
t offset = −t min + 0 (12.44)
2
where T0 is the time duration when the system is in zero space vector state.
The time interval tmax – tmin is denoted as T1 and is the time interval for which the system is at the space
vector corresponding to the maximum phase value which is phase b in this case. The time interval tmid – tmin
is the time interval for which the system is at the space vector corresponding to the middle phase value
which is phase a phase in this case. The values of T0, T1 and T2 are obtained from the instantaneous values
of the waveform. For every alternate cycle, one must take the mirror image of this sequence as the triangular
waveform slope alternates between positive and negative slopes. For the case of Figure 12.9, phase b has the
maximum voltage and therefore it is ON till time tbon (tmax + toffset) from the new reference. The phase a has
the in-between voltage and therefore, it is ON till time taon [(T0/2) + tmid – tmin] from the new reference.
Phase c has the minimum voltage and therefore, it is ON for only tcon (T0/2) time.
Chapter 12.indd 725 3/17/2009 8:23:12 PM

Over-Modulation During over-modulation, tmax – tmin becomes greater than Ts and T0 becomes negative.
To compensate for this, the times are multiplied by Ts/(tmax – tmin) such that the resulting space vector is
clamped to the hexagon boundary.
The generalized algorithm for the PWM generation is as follows.
[taon, tbon, tcon] = svpwm(Va, Vb, Vc)

Local Variables:
tmin, tmax, toffset, Ts, Vdc, Teff, T0
{
//fnd ta, tb, tc
ta=Ts*Va/Vdc;
tb=Ts*Vb/Vdc;
tc=Ts*Vc/Vdc;
//fnd tmax and tmin
tmax = ta;
if (tb > tmax)
tmax = tb;
elseif (tc > tmax)
tmax = tc;
end
tmin = ta;
if (tb < tmin)
tmin = tb;
elseif (tc < tmin)
tmin = tc;
end
Teff = tmax – tmin;
T0 = Ts − Teff;
Toffset = (T0/2) − tmin; //offset time.
//the on-time for each of the PWM phases is given by:
taon = ta + toffset;
tbon = tb + toffset;
tcon = tc + toffset;
//Overmodulation
if (T0 < 0),
toffset = −tmin*Ts/Teff;
taon = ta*Ts/Teff + toffset;
tbon = tb*Ts/Teff + toffset;
tcon = tc*Ts/Teff + toffset;
end
}
Chapter 12.indd 726 3/17/2009 8:23:12 PM

Here taon, tbon and tcon should be pushed into compare registers of the digital processor that makes
comparison with the triangular waveform. The triangular carrier waveform can be generated by program-
ming one of the internal general-purpose timers of the digital processor as an up–down counter to emulate
a triangle waveform having a period 2Ts. The output of the compare registers will produce the PWM
waveforms.
Over-Modulation
The maximum value of the inverter output voltage is limited by the DC bus voltage. To maintain linearity,
the reference space vector amplitudes are maintained within the space vector hexagon. However, during
transients, the reference space vector amplitudes may go beyond the hexagon. Such a situation is called over-
modulation. Such reference space vector demand must be detected and appropriately limited. There are
many strategies existing in the literature to treat the reference space vectors during over-modulation.
However, the over-modulation strategies fall into the following three categories: (a) circular boundary limit,
(b) hexagonal boundary limit and (c) expanding switching period.
Circular Boundary Limit

In this method, the largest circle that can be inscribed within the space vector hexagon is considered as the
boundary of the space vector. Therefore, in over-modulation situations, the reference space vector amplitude
is limited to the circular boundary as shown in Figure 12.10.
Referring to Figure 12.10, Vr* is the reference space vector that is demanded. As Vr* is beyond the maxi-
mum circular boundary limit and is limited to the circular boundary Vrm. The circular boundary limit is
given by
3
Vrm = Vdc ⋅ cos 30 = Vdc .
2
The space vector Vr* can be considered to be resolved into Vsa* and Vsb* as indicated in Figure 12.10.
V ∗rβ V r∗
V rβ
V rm
q V ∗rα
0 V rα V dc
Figure 12.10 Circular boundary limit within the space vector hexagon.
Chapter 12.indd 727 3/17/2009 8:23:12 PM

Thus,
Vr* = Vr*α 2 + Vr*β 2
If Vr* = Vrα* 2 + Vr*β 2 > Vrm , then Vr* is limited to Vrm such that
Vrm
Vrα = ⋅Vr*α
Vrα* 2 + Vr*β2
Vrm
V rβ = ⋅Vr*β
Vrα* 2 + Vr*β2
Vr = Vrα 2 + Vrβ 2
Thus it can be observed that in over-modulation situation, this method applies the largest possible voltage
space vector that is oriented in the same direction as the demanded reference voltage space vector and is on
the maximum circular locus that can be inscribed within the space vector hexagon. The inverter switches
cannot respond to small pulse widths near the state vectors. Therefore, if it is required to eliminate pulse
dropping then the circular locus may be accordingly modified to a near circular locus to ensure that the
pulse widths near the state vectors do not become very small.
Hexagonal Boundary Limit

The circular boundary limit method requires the square root and reciprocal algorithms to be used. The hexa-
gonal boundary limit method does not require the square root or the reciprocal functions. Therefore, this
method is more processor-friendly than the circular boundary limit method. The space vector PWM imple-
mentation discussed in the previous sub-section actually uses this hexagonal boundary limit.
Here, during over-modulation
T1 + T2 > Ts
Teff = (tmax – tmin) > Ts
In this case the calculated times are scaled with Ts / (T1 + T2 ) or Ts/Teff to produce a realizable vector
on the space vector hexagon as shown in Figure 12.11. The realized space vector is again oriented in the
V2
Circular bound
Hexagonal
bound
Sector 1 V r∗
√3
V
2 dc
60° V rc V rn
V dc
Figure 12.11 Hexagonal boundary limit.
Chapter 12.indd 728 3/17/2009 8:23:13 PM

same direction as the demanded reference space vector but it now lies on the hexagonal boundary rather
than the circular the boundary. This algorithm is simple to implement on a digital processor but the
resulting line voltages contains low frequency distortions. However, in this method the voltage vectors
obtainable near the state vectors of the sector are larger than that of the circular boundary method as
Referring to Figure 12.11, Vr* is the demanded space vector, Vrc is the resulting space vector on using the
circular boundary limit method and Vrn is the resulting space vector on using the hexagonal boundary limit
method.
Expanding Switching Period

In the previous two methods, the switching period Ts is held constant. To better utilize the inverter espe-
cially for high-power drives, the inverter drive pulse should seamlessly move from the high-frequency carrier
PWM to the low-frequency six-step waveform as described in Chapter 6. For this the switching period Ts is
gradually expanded till six-step operation results. There are numerous algorithms in the literature that dis-
cuss the smooth transition from PWM to six-step each with their advantages and disadvantages. A simple
method of expanding the switching period is as follows.
During over-modulation situation, T1 + T2 > Ts or Teff > Ts. For a demanded reference space vector,
calculate T1 and T2 for the active state vectors. If T1 + T2 > Ts then T0 < 0. Set Ts = T1 + T2. This implies
that the period register of the processor’s general-purpose timer that is being used as an up–down counter
to emulate a triangular carrier is changed. With this changed period, T0 = 0 and the inverter system
spends T1 period in one active state vector of the sector and the remaining period (T2) is spent on the
other active state vector of the sector. Progressively, if the demanded reference space vector becomes larger
and larger, the carrier period for those space vectors is progressively increased till six-step operation is
reached.
This chapter on discrete computational essentials algorithm. The choice of the base values of variables
is a rather important one. The implementation of for normalization is another classic case of trade-off
the power electronic circuits can be completed between precision on one hand which has a bearing
only if the control algorithms are successfully on the minimum value of the variables and range on
incorporated within a digital processor. The con- the other hand to accommodate short-term over-
cept of the real number is much different when flows like over-voltage and over-currents in practical
viewed from within the digital domain due to the applications.
two primary limitations of the discrete real num- This chapter discusses various arithmetic algo-
bers: (a) range and (b) resolution. In this regard, it rithms that are essential for executing the linear and
becomes very important to select a numeric format non-linear models of the power electronic systems
for a particular application. Simple arithmetic pro- for the purposes of control. The framework of arith-
cesses like addition, subtraction and multiplication metic computation is based on iterative algorithms
can give wrong results if the numeric format is not in order to make the computations repeatable,
properly chosen. simple and self-terminating. The next section pro-
Normalization of the variables within the digital vides few tutorial exercises that will give the neces-
domain is another important consideration that sary practice for implementing even complex
needs to be addressed before implementing an applications.
Chapter 12.indd 729 3/17/2009 8:23:13 PM

1. Capture and convert analog voltage to discrete (b) Acquire the signal and let that represent
domain Vm in Vm sinwt.
(c) Include a sine table in the data memory
Mode of implementation:
space.
a. Microcontroller or DSP Kit
(d) Using pointers that are spaced an equiva-
Tasks for study: lent 90° apart, read out the sine and cosine
(a) Connect a potentiometer to a 5 V supply values from the table and multiply with
and connect the tap point to the ADC Vm. The read-out rate (frequency of the
connector input. (Keep the tap value to waveforms) can be fixed by the timer that
less than ADC pin voltage rating). interrupts the interrupt service routine
(b) Write an interrupt service routine (ISR) to (ISR).
acquire the analog signal and observe the (e) This will result in two waveforms: Vmsinwt
acquired value by transferring the value in and Vmcoswt. Let them be named as Va
the ADC FIFO register to a variable. and Vb , respectively.
Observe the variable value in the WATCH (f ) Now apply the following algorithm,
window of the debugger. i. Compare Va and Vb .
(c) Calculate the a priori value that should be ii. Evaluate Larger of [Va, Vb] + 0.5 times
obtained in the variable register and com- Smaller of [Va, Vb ].
pare with the result displayed in the debug- iii. Comment on the result.
ger WATCH window.
(d) Now transfer the variable to a DAC ports (g) In MATLAB compute the maximum error
and observe on the oscilloscope. that can occur between the following two
(e) In order to compute the time taken by the algorithms.
above interrupt service routine. i. sqrt(V 2a + V 2b ).
i. At the beginning of the ISR make an ii. Larger of [Va, Vb] + 0.5 times Smaller
I/O port high. of [Va, Vb ].
ii. At the end of the ISR make that I/O iii. Compare with results of step 6 and
port low. comment on the results
iii. Observe the waveform of that I/O port 3. Normalization and Scaling.
on the scope.
iv. Measure the ON-duration of the pulses Mode of implementation:
to obtain the ISR compute time. a. Microcontroller or DSP Kit.
2. Computation of a space vector using the open- Tasks for study:

loop hypotenuse algorithm. (a) Acquire the signal and let that represent
Vm in Vmsinwt.
Mode of implementation: (b) Include a sine table in the data memory
a. Microcontroller or DSP Kit space.
Tasks for study: (c) Use the following base values for normal-
(a) Connect a potentiometer to a 5 V supply ization and compute the scaling factor
and connect the tap point to the ADC i. Vbase = Vmax
connector input. (Keep the tap value less
ii. Vbase = 0.5Vmax
than the ADC input voltage rating).
iii. Vbase = 0.25Vmax
Chapter 12.indd 730 3/17/2009 8:23:13 PM

where Vmax is the maximum voltage

1⎛ x ⎞
that can be applied to the ADC i. yk +1 = ⎜ yk + ⎟
inputs. 2⎝ yk ⎠
(d) Convert the signal Vm and the resulting ii. if x = 2mx1 and 0.5 ≤ x1 < 1, then y0 =
sine signal into the D(2, 3, 12) format and 2truncate(m/2)
observe the sine wave on the oscilloscope. iii. Evaluate the above algorithm for speed of
convergence in the D(2, 7, 8) format for
4. Computation of reciprocal: To compute y = 1/x a given value of x in D(2, 7, 8) format.
where x > 0
(b) Algorithm 2:
Mode of implementation: y ⎛ y 2⎞
a. Microcontroller or DSP Kit i. y k +1 = k ⎜ 3 − k ⎟
2 ⎝ x ⎠
Tasks for study: ii. if x = 2 x1 and 0.5 ≤ x1 < 1, then y0 =
m
(a) Algorithm 1: 2truncate(m/2)

i. yk+1 = yk(2 – xyk) iii. Evaluate the above algorithm for speed of
ii. if x = 2mx1 and 0.5 ≤ x1 < 1, then y0 = convergence in the D(2, 7, 8) format for
2−m a given value of x in D(2, 7, 8) format.
iii. Evaluate the above algorithm for speed
of convergence in the D(2, 7, 8) format (c) Algorithm 3:
for a given value of x in D(2, 7, 8) i. y = sqrt(x) can be modeled as (x − y ⋅ y)
format. = 0 or (x − y2) = e where e is the
error.
(b) Algorithm 2: ii. A typical control system can be used
i. y = 1/x can be modeled as (1 − xy) = 0 wherein the reference is the value ‘x’ and
or (1 − xy) = e where e is the error. the feedback is y*y. The error is fed to an
ii. A typical control system can be used integrator through a gain K. An integral
wherein the reference is the value ‘1’ control is used to obtain zero error.
and the feedback is x × y. The error is iii. Simulate the above first in SIMU-
fed to an integrator through a gain K. LINK.
An integral control is used to obtain iv. Then implement the same in the DSP
zero error. using D(2, 7, 8) format.
iii. Simulate the above first in SIMU-
LINK. v. Find out the value of K which results
iv. Then implement the same in the DSP in the fastest convergence to stability.
using D(2, 7, 8) format. (d) Compare the speeds of convergence of the
v. Find out the value of K which results above algorithms using the same initial
in the fastest convergence to stability. starting values.
(c) Compare the speeds of convergence of the 6. Implement the following discrete transfer func-
above two algorithms. tion in a digital processor,
5. Computation of square roots: to compute y =
1 − e −τT
x where x > 0 H (z ) =
Mode of implementation:
(
τ z − e −τT )
a. Microcontroller or DSP Kit where T is the sampling time period in seconds
and t is any positive constant in D(2, 3, 12)
Tasks for study:
format.
(a) Algorithm 1:
Chapter 12.indd 731 3/17/2009 8:23:13 PM

Mode of implementation: (b) Comment on the nature of the output.

a. Microcontroller or DSP Kit (c) Vary t and observe the output. Comment
on the effect of t on the output.
Tasks for study:
(d) Increase the frequency of the input square
(a) Apply a very low-frequency square wave as
wave gradually. What happens to the
an input to this function and observe the
output?
output on the oscilloscope.

1. Natural real numbers have infinite 8. The computation overhead for the fixed-point
and infinite . system is much than that for the
floating-point number system.
2. In discrete real number finite range results in
error. 9. If x1 is in D(B, i, f ) and x2 is in D(B, i, f ),
then x1 + x2 is in .
3. Discrete real number has finite apart
from finite range. 10. If x1 is in D(B, i1, f1) and x2 is in D(B, i2, f2)
where i1 ≠ i2 and f 1 ≠ f 2 then it is
4. The relative density is in the floating-
to add x1 and x2.
point number system.
11. If x1 is in D(B, i, f ) and x2 is in D(B, i, f ),
5. In the fixed-point system, one must keep track
then x1 − x2 is in .
of the position of the point during
arithmetic operations. 12. If x1 is in D(B, i1, f1) and x2 is in D(B, i2, f2)
where i1 ≠ i2 and f 1 ≠ f 2 then it is
6. There is a limitation in the floating-point
to subtract x1 and x2.
system in representing the number
in the normalized format. 13. Multiplication can be performed on two num-
bers of numeric formats but with the
7. The floating-point system has the problem of
same base.
under flow that limits the representa-
tion of very numbers. 14. If x1 is in D(B, i1, f1) and x2 is in D(B, i2, f2),
then product will be in .
1. Discuss how the fixed-point number system is 5. How does the floating-point representation
represented. differ from the fixed-point representation?
2. For a 16-bit processor, what does the following 6. What is normalized floating-point?
mean? 7. How is zero represented in floating-point
D(2, 8, 8), D(2, 7, 8), D(2, 0, 16) and D(2, 0, number system?
15). 8. What is relative error and relative density with
3. Write short notes on: (a) relative error, (b) rela- respect to floating-point number system?
tive density. 9. Compare the relative errors and relative densities
4. Discuss floating-point numeric format. of fixed- and floating-point number systems.
Chapter 12.indd 732 3/17/2009 8:23:13 PM

10. What is underflow hole in floating-point 19. Discuss the normalization and scaling for con-
number system? verting an analog value to the digital number.
11. How can the underflow hole problem in a 20. For a 16-bit system, if the variables in the appli-
floating-point number system be solved? cation are ensured to have a value less than unity
Explain with an example. under all conditions, then all values including
the result will be within the range of D(2, 1, 14).
12. For a 16-bit processor, what does the following
Extend this result to an n-bit system.
mean?
21. Explain the multiplication algorithm.
D(2, 12, 0, 7), D(2, 12, −3, 3), D(2, 14, 0, 3)
and D(2, 13, 0, 3). 22. What are the advantages of iterative algo-
rithms?
13. What are the three primary arithmetic operators?
23. Discuss the computation of the reciprocal and
14. Discuss compatibility issues in the fixed-point
square roots using the closed-loop control
numeric format of the numbers for addition,
method.
subtraction and multiplication operations.
24. Write short notes on: (a) PI controller imple-
15. If x1 is in D(B, i1, f1) and x2 is in D(B, i2, f2),
mentation, (b) sine and cosine implementation.
then how to extract the product in D(B, i1, f1)?
25. Write short notes on: (a) PWM for two-level
16. If x1 is in D(B, i1, f1) and x2 is in D(B, i2, f2),
two-arm bridge inverter, (b) space vector
then how to extract the product in D(B, i2, f2)?
PWM.
17. What is normalization in fixed-point number
26. Write short notes on: (a) circular boundary
representation?
limit, (b) hexagonal boundary limit and (c)
18. What is scaling? extending switching period.
| PROBLEMS
1. Consider a 16-bit fixed-point system D(2, 0, d. x1 = (0.5)10 in D(2, 0, 16) and x2 = (100)10
15). What is its range? What is the relative in D(2, 8, 8)
error of the following equivalent decimal num- e. x1 = (+1.54)10 in D(2, 1, 14) and x2 =
bers: (a) 0.98536, (b) −0.234577, (c) 0.8379 (−113.54)10 in D(2, 7, 8)
and (d) 0.0011378.
4. Derive the generic iteration algorithm for any
2. Consider a 16-bit normalized floating-point given function F(x, y) = 0.
system D(2, 12, −3, 3). What is its range? What
5. For a 16-bit binary number system, compute
is the relative error of the following equivalent
the reciprocal for the following numbers:
decimal numbers: (a) 0.9853637, (b)
a. x = (5)10 in D(2, 3, 12)
−0.23457733, (c) 0.8379 and (d) 0.0011378.
b. x = (26)10 in D(2, 6, 10)
3. Perform multiplication for the following: c. x = (4.5)10 in D(2, 3, 12)
a. x1 = +913.74 in D(10, 3, 2) and x2 = d. x = (0.333)10 in D(2, 1, 14)
+908.14 in D(10, 3, 2) e. x = (10)10 in D(2, 4, 12)
b. x1 = +1.1354 in D(10, 1, 4) and x2 = +8.34
in D(10, 1, 2)
the square roots for the following numbers:
c. x1 = (+6.76846)10 in D(2, 3, 12) and x2 =
a. x = (5)10 in D(2, 3, 12)
(+134.78)10 in D(2, 8, 8)
Chapter 12.indd 733 3/17/2009 8:23:13 PM

b. x = (26)10 in D(2, 6, 10) 10. For a 16-bit binary number system, compute
c. x = (4.5)10 in D(2, 3, 12) the hypotenuse by the open-loop method for
d. x = (0.333)10 in D(2, 1, 14) the following values for the sides of the right-
e. x = (10)10 in D(2, 4, 12) angled triangle:
a. a = (5)10 in D(2, 4, 12); b = (7)10 in D(2, 4,
12)
the reciprocal of the square roots for the fol-
b. a = (26)10 in D(2, 6, 10); b = (13)10 in D(2,
lowing numbers:
6, 10)
a. x = (5)10 in D(2, 3, 12)
c. a = (4.5)10 in D(2, 4, 12); b = (7.5)10 in
b. x = (26)10 in D(2, 6, 10)
D(2, 4, 12)
c. x = (4.5)10 in D(2, 3, 12)
d. a = (0.333)10 in D(2, 1, 14); b = (0.895)10 in
d. x = (0.333)10 in D(2, 1, 14)
D(2, 1, 14)
e. x = (10)10 in D(2, 4, 12)
e. a = (10)10 in D(2, 4, 12); b = (5)10 in D(2,
8. For a 16-bit binary number system, compute 4, 12)
the square root using the reciprocal of the
square root iteration algorithm for the follow-
e x for the following numbers:
ing numbers:
a. x = (5)10 in D(2, 3, 12)
a. x = (5)10 in D(2, 3, 12)
b. x = (26)10 in D(2, 6, 10)
b. x = (26)10 in D(2, 6, 10)
c. x = (4.5)10 in D(2, 3, 12)
c. x = (4.5)10in D(2, 3, 12)
d. x = (0.333)10 in D(2, 1, 14)
d. x = (0.333)10 in D(2, 1, 14)
e. x = (10)10 in D(2, 4, 12)
e. x = (10)10 in D(2, 4, 12)
ln(x) for the following numbers:
e x for the following numbers:
a. x = (5)10 in D(2, 3, 12)
a. x = (5)10 in D(2, 3, 12)
b. x = (26)10 in D(2, 6, 10)
b. x = (26)10 in D(2, 6, 10)
c. x = (4.5)10 in D(2, 3, 12)
c. x = (4.5)10 in D(2, 3, 12)
d. x = (0.333)10 in D(2, 1, 14)
d. x = (0.333)10 in D(2, 1, 14)
e. x = (10)10 in D(2, 4, 12)
e. x = (10)10 in D(2, 4, 12)
| ANSWERS
Fill in the Blanks
1. range; density 8. lower
2. overflow 9. D(B, i, f )
3. resolution 10. not possible
4. uniform 11. D(B, i, f )
5. base 12. not possible
6. zero 13. different
7. hole; small 14. D(B, i1 + i2, f1 + f2)
Chapter 12.indd 734 3/17/2009 8:23:14 PM

T hermal Aspects
Learning Objectives
CHAPTER
13
understand the parameters used for thermal analysis.
design heat sinks for a given application.
T he power semiconductor devices dissipate power by virtue of the finite rise and fall times associated
with the device switching. In Chapter 1 the method of calculating the power dissipation in a power
semiconductor device is discussed. As a consequence, the junction temperature increases causing thermal
stresses in the device. It is necessary to ensure that under no condition the junction temperature should
exceed the maximum-rated junction temperature which is 150°C for most devices. This chapter discusses
the issues involved in removing the heat generated within the semiconductor devices and the selection of
heat sinks.
To obtain an insight into the thermal analysis of the power electronic systems, the heat flow mechanism
must be understood. In general, there are four types of heat transfer mechanisms that are operative at any
given time on a hot body. The degree to which a particular heat transfer mechanism dominates depends
upon the specific application and the physical system. There are four types of heat transfer mechanisms
which are as follows:
1. Conduction: Conduction is a process of heat transfer wherein the heat flows through solids from a
higher temperature zone to a lower temperature zone.
2. Convection: Convection is the process of heat transfer by means of fluids.
3. Radiation: Radiation is a process of heat transfer, where the heat flows out from the hot body without
the need for a heat transport medium.
4. Mass Transport: Mass transport is also a process of heat transfer wherein the heat is transferred by
means of fluids. In a way it is similar to convection. However, convection is an uncontrolled process
whereas mass transport is a controlled process where the discharge rate of the fluid can be controlled
resulting in controlled heat transfer from one region to another.
It should be noted that all the heat transfer processes mentioned above are dependent on the properties of
materials, fluids and the immediate environment.
Chapter 13.indd 735 3/18/2009 6:02:38 PM

13.1 Modeling Heat Transfer
T o analyze the thermal aspects, one must first model the system from the thermal point of view. A system
from the thermal point of view consists of a heat source, the heat transfer medium and the heat sink. In
the case of the power electronics systems, the power semiconductor devices are not ideal switches. As dis-
cussed in Chapter 1, they dissipate power both during the ON-state as conduction losses and during the
switching transition as switching losses. This causes the junction temperature of the power semiconductor
devices to rise. The semiconductor junction of the power devices is considered as the heat source in most
power electronic systems. The heat is transferred through the semiconductor material and the case of the
device to the external ambient. The external ambient is the heat sink.
A generic thermal system is shown in Figure 13.1. It consists of a hot body, H at temperature T1 at the
center of a spherical enclosure. The spherical space within the enclosure is filled with a fluid such as air. The
temperature at the inner surface of the enclosure is Tsi and the temperature at the outer surface of the enclo-
sure is Tso. At the far neighborhood of the spherical enclosure, the ambient temperature is TA. The heat from
the hot body flows outwards towards the spherical enclosure by radiation and convection. The heat flows
through the solid enclosure wall by the conduction mechanism. External to the enclosure, the heat further
flows into the atmosphere again by convection and radiation. It should be noted that as heat always flows
from the higher temperature zone to the lower temperature zone, T1 > Tsi > Tso > TA. The heat flow can be
modeled as shown in Figure 13.2.
In the steady-state heat transfer schematic diagram shown in Figure 13.2 wherein the system is in ther-
mal equilibrium, T1 is the temperature of the hot body, Tsi is the temperature at the inner surface of the
spherical enclosure, Tso is the temperature at the outer surface of the enclosure and TA is the ambient tem-
perature. Also Rq v1 is the thermal resistance to heat flow between the hot body and the inner surface of the
enclosure by the convection mechanism, Rqr1 is the thermal resistance to heat flow between the hot body and
T1
TA Tsi Tso TA
Figure 13.1 Heat flow from hot body to ambient.
Chapter 13.indd 736 3/18/2009 6:02:40 PM

Thermal Aspects 737
Rq v1 Rq c1 Rq v2
T1 Tsi Tso TA
Convection Conduction Convection
Rq r1 Rq r2
Radiation Radiation
Figure 13.2 Heat transfer model.
the inner surface of the enclosure by the radiation mechanism, Rqc1 is the thermal resistance to the heat flow
between the inner and the outer surfaces of the enclosure by the conduction mechanism, Rq v2 is the thermal
resistance to heat flow between the outer surface of the enclosure and the ambient by the convection mecha-
nism and Rqr2 is the thermal resistance to heat flow between the outer surface of the enclosure and the ambi-
ent by the radiation mechanism. As the convection and radiation occur in parallel within and outside the
enclosure, the corresponding thermal resistances are shown in parallel.
With reference to the heat flow indicated in Figure 13.2, the temperature difference is considered equiv-
alent to the potential difference of electrical circuit and the rate of heat flow or power flow is considered
equivalent to the current flow. It should be noted that the thermal domain is different from the other
domains in that there is no concept of dissipation in this domain. The thermal resistance is not a dissipative
element but only a notional element that indicates the conductivity of heat through the medium. There is
no heat loss in the thermal resistance, but only a temperature difference. Higher the thermal resistance,
lesser is the rate of heat flow in the medium and vice-versa. Thus, one can write
ΔT = PRθ (13.1)
where DT is the temperature difference in °K; P the heat power flow in W; Rq the thermal resistance in °K/W.
Equation (13.1) is sometimes referred to as the Ohm’s Law for thermal domain. As an example, referring to
Figure 13.2, the heat power flow through the convection mechanism from the hot body to the inside of the
enclosure is given as
T −T
Pθ v1 = 1 si (13.2)
Rθ v1
Referring to Figure 13.3, A is the cross-sectional area of the medium that is orthogonal to the heat flow. It is
usual to represent the rate of heat flow in terms of normalized power flow and is represented as
P
=q (13.3)
A
T1 T2
Figure 13.3 Power flow and cross-section area.
Chapter 13.indd 737 3/18/2009 6:02:40 PM

where q is the rate of heat flow per unit area or specific heat flow rate and expressed in W/m2. Applying
Eq. (13.3) to the system schematic of Figure 13.2, one obtains
Pθv1 ΔT ΔT
= = = q v1
A ARθv1 rθv1
where rθ v1 = Rθv1 A is called the thermal resistivity and is expressed in °Km2/W. In general, one can write
ΔT
q= = h ΔT (13.4)
rθ
where q is the specific heat flow rate in W/m2; rq the thermal resistivity in °Km2/W; h = 1/ rθ is called the
thermal coefficient and expressed in W/oK/m2. It should be observed that the material property of the
medium is represented as Rq , rq or h. In the literature, thermal resistance, thermal resistivity and thermal
coefficient have been used in various circumstances; however, in general they represent the characteristic of
the medium and can be used in any of the above forms and converted from one form to another using the
above relationships. Like finding the power flow through the thermal resistance, Rθ v1, the power flow
through all the other branches can also be estimated in a manner similar to Eq. (13.2).
13.2 Heat Transfer by Conduction
C onduction is a heat transfer mechanism wherein heat energy flows from a hotter zone to a cooler zone
in a solid. Heat transfer processes by conduction can be quantified in terms of the rate equation known
as the Fourier’s law. Consider a thin rectangular solid slab as shown in Figure 13.4.
Referring to Figure 13.4, q is the specific heat flow rate in the direction, T1 is the temperature of the
hotter side of the solid slab and T2 is the temperature of the colder side of the slab. From Eq. (13.4),
q = h(T1 − T2 ) = h ΔT (13.5)
where h is the thermal coefficient in W/°K/m2. Let k be another constant called thermal conductivity which
is defined in terms of the thermal coefficient and the thickness of the slab Δx. Thus,
k ≡ h Δx [W/°K/m]
k
h= (13.6)
Δx
T2
T1
Δx
Figure 13.4 Heat transfer through solids.
Chapter 13.indd 738 3/18/2009 6:02:40 PM

Thermal Aspects 739

ΔT
q =k (13.7)
Δx
Equation (13.7) is the heat flow rate equation. It implies a temperature gradient along a length of the solid
slab in the direction of the heat flow. Thermal conductivity, k, is a transport property and is the characteristic
of the solid slab material. If A is the cross-section area of the slab then the heat flow rate by conduction Pc is
the product of the specific heat flow rate and the area as per Eq. (13.3). Thus,
ΔT
Pc = kA (13.8)
Δx
The various thermal constants are summarized as follows:
1. Thermal resistance Rq expressed in °K/W.
2. Thermal resistivity rq expressed in °Km2/W.
3. Thermal coefficient h expressed in W/°K/m2.
4. Thermal conductivity k expressed in W/°K/m.
The relationship of the various thermal constants in terms of thermal conductivity is as follows:
1. Thermal coefficient h = k / Δx from Eq. (13.6).
2. Thermal resistivity
1 Δx
rθ = = (13.9)
h k
3. Thermal resistance
rθ Δ x
Rθ = = (13.10)
A kA
Thermal conductivity is a material property. Table 13.1 gives the values of the thermal conductivity for some
materials.
Table 13.1 Thermal conductivities for some materials
Material k value in W/°K/m
Silver 410
Copper 385
Aluminum 211
Steel 47.6
Glass 1.05
Heat-sink compound 0.4
Epoxy 0.2
Polyurethane 0.025
Polystyrene 0.035
Still air 0.026
Chapter 13.indd 739 3/18/2009 6:02:41 PM

Problem 13.1
Calculate thermal resistance to the flow of power through a solid slab wherein
(a) the slab is made up of aluminum having an area of 1 m2 and thickness of 10 mm.
(b) the slab is made up of glass having an area of 1 m2 and thickness of 5 mm.
Solution
The expression for the thermal resistance in terms of thermal conductivity is given as
Δx
Rθ =
kA
(a) Δ x = 10 mm = 10 × 10−3 m; k = 211; A = 1 m2
10 × 10−3
Rθ = = 0.000004739 °K / W
211 × 1
(b) Δ x = 5 mm = 5 × 10−3 m; k = 1.05; A = 1 m2
5 × 10−3
Rθ = = 0.004762 °K / W
1.05 × 1
A typical example of heat transfer by conduction is the case of heat dissipation in semiconductor devices
such as junction transistors (BJT and MOSFET). Any device that has a voltage across it and current through
it will dissipate power. This power is equal to the product of the voltage and the current. This power is dis-
sipated as heat. In particular, a power semiconductor device that is switching will dissipate heat due to the
ON-state loss and the switching loss as discussed in Chapter 1.
The maximum operating temperature of a semiconductor junction is usually 150°C. Exceeding this
junction temperature limit will destroy the device. An important part of the design process is to ensure that
the device junction temperature is within a certain specified limit. To safeguard this, heat sinks are used with
semiconductor devices such that the thermal resistance is lowered to increase the rate at which the dissipated
heat energy is drawn away from the junction to the external ambient. From Eq. (13.10), the thermal resis-
tance Rq is seen to depend on the thickness of the heat-sink material, thermal conductivity of the material
and the area of the heat-sink material.
Once the heat-sink material is selected, the thermal conductivity being a property of the material is
therefore determined for the material. The thickness and the surface area of the heat sink can be varied by
the designer to achieve the thermal resistance for a particular application. Commercially, heat sinks with
varied cross-sections are available as aluminum extrusions. Once the heat-sink cross-section for an applica-
tion is decided, then the designer needs to only vary the length of the heat sink to achieve the required
thermal resistance.
The heat generated in the junction should be removed in such a manner that at thermal equilibrium the
junction temperature is well below the rated junction temperature. The electrical equivalent for the flow of
heat from the junction to the ambient is depicted in Figure 13.5. In the steady-state equivalent circuit
shown in Figure 13.5, Rqjc is the effective thermal resistance between the junction and the outer case of the
device. This value is available in the datasheet of the specific device that is being used in the circuit. Rqcs is
the thermal resistance between the case and the heat sink and is dependent on the manner in which the
Chapter 13.indd 740 3/18/2009 6:02:41 PM

Thermal Aspects 741
Tj Tc Ts
j c s
Rq jc Rq cs
Pd
P Rq sa
a (Ambient) TA
Figure 13.5 Thermal equivalent circuit of a power semiconductor device mounted on a heat sink.
device is mounted on the heat sink. It varies between 0.1°K/W to 1°K/W depending on the type of mount-
ing (such as non-isolated mounting, isolated mica washer mounting and skill of mounting). Rq sa is the
thermal resistance between the sink and the ambient. Rqsa is to be calculated and a heat sink having a ther-
mal resistance that is lower than the calculated Rqsa should be selected.
A power semiconductor device mounted on a heat sink is shown in Figure 13.6. From Eq. (13.1), one
obtains
Tj − TA = Pd ( Rθjc + Rθcs + Rθsa ) (13.11)
The values Tj, Rqjc are known from datasheets. The heat sink should be selected such that even for the maxi-
mum ambient temperature, the power semiconductor switch’s junction temperature is within safe limits.
TA should be taken at 50°C (worst case) as the ambient in the immediate neighborhood of the device’s heat
sink. Pd is the estimated dissipation in the power semiconductor switch. Rq cs is between 0.1°K/W and 1°K/W.
Thus Rq sa, that is, the thermal resistance between sink and ambient can be computed from Eq. (13.11) and an
appropriate heat sink whose thermal resistance is less than the calculated value can be chosen. The concept of
transient thermal impedance is important when dealing with pulsed overload currents. Mounting of the device
on the heat sink and orientation of heat sink are to be considered carefully to increase the thermal efficiency
and thereby improving the reliability of the device.
Heat sink fins
Case
Leads
Figure 13.6 A power semiconductor device mounted on a heat sink.
Chapter 13.indd 741 3/18/2009 6:02:41 PM

Tjmax − TA
Rθjc + Rθcs + Rθsa =
Pd-max
Tjmax − TA
Rθsa = − Rθjc − Rθcs (13.12)
Pd-max
where Tjmax is the maximum allowable junction temperature which is usually assumed less than three-fourth
of the maximum-rated junction temperature. Equation (13.12) is the design equation for selecting a heat
sink considering only the conduction mechanism of heat transfer.
13.3 Heat Transfer by Convection

C onvection is a mode of heat transfer between a solid and a fluid when there is a temperature difference.
The convection heat transfer mode is composed of the following two mechanisms; (a) heat transfer due
to random molecular motion (diffusion) and (b) heat transfer by the bulk motion of the fluid. Consider a
heated surface as shown in Figure 13.7. The fluid above the heated surface is air in this case.
The fluid region close to the surface is at a temperature Ts and the temperature of the fluid at a farther dis-
tance from the surface is TF . The higher temperature of the fluid closer to the surface makes the fluid molecules
more agitated and less dense. As a result they rise up and the colder, denser, fluid portion reaches the surface
which in turn will get heated up through interaction between the fluid and the heated surface. This results in the
development of a region in the fluid through which the fluid velocity varies in the neighborhood of the surface.
The convection heat transfer mode is sustained both by diffusion and by the bulk motion of the fluid
within the boundary layer. The contribution due to diffusion dominates near the surface where the fluid
velocity is low. The contribution of bulk fluid motion is due to the fact that the boundary layer grows as the
flow progresses in a direction orthogonal to the surface or in a direction away from the heated surface. In
effect, the heat that is conducted into this layer in the immediate neighborhood of the surface is swept away
and is eventually transferred to the fluid outside the immediate neighborhood.
Convection heat transfer may be classified into the following types:
1. Free Convection: Here the flow is induced by the density gradient which arises from density differ-
ences caused by temperature variations in the fluid.
2. Forced Convection: Here the flow is caused by external means such as fan, pump or atmospheric winds.
From Eq. (13.8), the rate of heat flow for the system of Figure 13.7 is
ΔT
P = kA (13.13)
Δx
TF
Dx Fluid
TS
Heated surface
Figure 13.7 Illustration of the convection mechanism.
Chapter 13.indd 742 3/18/2009 6:02:41 PM

Thermal Aspects 743
where k is the thermal conductivity; A the cross-sectional area perpendicular to the heat flow; DT = TS − TF
the temperature difference between the surface and outer fluid layer; Dx the thickness of the fluid in the
immediate neighborhood of the heated surface. In Eq. (13.13), all the parameters are measurable except the
fluid thickness Δx which is a quantity that has a measure of uncertainty in it. It varies with surroundings and
external influences. An additional parameter X is included which is measurable and the fluid thickness is
considered as a function of the measurable parameter X. Thus, Eq. (13.13) can be re-written as:
⎛ X ⎞ ΔT
P = kA ⎜ ⎟ (13.14)
⎝ Δx ⎠ X
where X is called the characteristic dimension of the heated solid surface. The ratio of the characteristic
dimension to the fluid thickness is called the Nusselt number, a dimensionless quantity represented by N.
Eq. (13.14) is re-written as
ΔT
P = kAN (13.15)
X
Rearranging the terms of Eq. (13.15), one obtains the expression for heat flow rate/unit area and convective
thermal resistance as
P ΔT
= q = kN (13.16)
A X
ΔT X
Rθ v = = (13.17)
P kN A
From Eq. (13.9), the thermal resistivity is given as
X
rθ v = Rθv A = (13.18)
kN
The convective thermal coefficient is the reciprocal of the thermal resistivity and therefore,
1 kN
hθv = = (13.19)
rθv X
Substituting for convective thermal coefficient in the expression for heat flow rate of Eq. (13.16), one
obtains
q = hθ v ΔT
P = hθv AΔT (13.20)
Determination of Nusselt Number

In the determination of the convective thermal resistance or the thermal coefficient, the factor with a measure
of uncertainty is the Nusselt number. The Nusselt number is a dimensionless quantity which is a function of
the velocity of the fluid flow, the properties of the fluid medium and the geometry of the hot solid body
which transfers the heat to the fluid. Further, the Nusselt number also depends on the type of convection,
whether free convection or forced convection. In the case of free convection, one uses another dimensionless
quantity called the Rayleigh number to compute the Nusselt number. In the case of forced convection, one
uses the dimensionless quantity called the Reynolds number to compute the Nusselt number.
Chapter 13.indd 743 3/18/2009 6:02:42 PM

Rayleigh Number
This quantity is required for free convection. It is represented by A and is given by the following expression:
g β X 3 ΔT
A= (13.21)
δ ⋅ν
where g is the acceleration due to gravity = 9.81 m/s2; b the coefficient of thermal expansion of the fluid
medium; X the characteristic dimension of the hot solid surface; d the thermal diffusivity; v the kinematic
viscosity of the fluid medium.
Reynolds Number
This quantity is required for forced convection. It is represented by R and is given by the following expression:
uX
R = (13.22)
ν
where u is the mean velocity of forced fluid flow; X the characteristic dimension of the hot solid surface;
v the kinematic viscosity of the fluid medium.
The Nusselt number is a function of both the Rayleigh number and Reynolds number and is given as
N = f ( A, R ) (13.23)
When the fluid motion is highly ordered and has a streamlined flow, then such fluid motion is called lami-
nar flow. If the fluid motion is highly irregular and is characterized by velocity fluctuations, then such a fluid
motion is called turbulent flow. The Rayleigh number and the Reynolds number have been used for classify-
ing whether a fluid flow is laminar or turbulent. The laminar–turbulent distinction is dependent on the
solid–fluid interface geometries and may vary for various geometries and positions.
Free Convection
The Nusselt number is best found by experimentation for an arbitrary shape of the hot solid. However, for
standard shapes of the solids, the following curve-fit relationships can be used for preliminary designs. These
should, however, be verified and fine tuned after experimentation as there is always a certain degree of uncer-
tainty in estimating the flows and the nature of the flows, whether laminar or turbulent. For any convective-
based thermal system, both the characteristic dimension X of the solid and the Nusselt number for the fluid
are needed to evaluate the thermal coefficient as per Eq. (13.19).
Horizontal Flat Plate

Figure 13.8 illustrates horizontal flat plates of two different geometries, viz., a rectangular geometry with
dimensions a and b as shown and another having a circular geometry with diameter d as indicated.
The thickness of the plate is considered small compared to the other indicated dimensions. The plate is
placed horizontal and is the heat source. The fluid (air) heats up near the hot surface, becomes less dense and
moves up as shown. The Nusselt numbers for the plates of geometry given in Figure 13.8 are computed as
follows: The characteristic dimension X for the rectangular plate is given as
a +b
X =
2
and for the circular plate
X =d
Chapter 13.indd 744 3/18/2009 6:02:42 PM

Thermal Aspects 745
Fluid flow
Fluid flow
Hot plate Hot plate
a d
(a) (b)
Figure 13.8 Horizontal flat plate: (a) Rectangular plate; (b) circular plate.
For a laminar flow where 102 < A < 105, the Nusselt number is determined as
N = 0.54 A 0.25 (13.24)
For a turbulent flow where A > 105, the Nusselt number is determined as
N = 0.14 A 0.33 (13.25)
Vertical Flat Plate

Figure 13.9 illustrates vertical flat plates of two different geometries, viz., a rectangular plate with dimen-
sions a and b as shown and another having a circular geometry with diameter d as indicated.
The thickness of the plate is considered small compared to the other indicated dimensions. The plate is
placed vertical and is the heat source. The fluid (air) heats up near the hot surface, becomes less dense and
moves up as shown. The Nusselt numbers for the plates are computed as follows: The characteristic dimen-
sion X for a rectangular plate is given as
X =b
and for a circular plate
X =d
Hot
Fluid
flow
Fluid
flow
d
b
a
Hot
(a) (b)
Figure 13.9 Vertical flat plate: (a) Rectangular plate; (b) circular plate.
Chapter 13.indd 745 3/18/2009 6:02:42 PM

N = 0.56 A 0.25 (13.26)
For a turbulent flow where 109 < A < 1012, the Nusselt number is determined as
N = 0.20 A 0.40 (13.27)
From Eqs. (13.24) to (13.27), it can be observed that a flat plate placed vertically results in a higher Nusselt
number and consequently the thermal resistance is lower. This implies that the vertical plate will convect the
heat away from the solid surface better than from a horizontal surface.
Horizontal Cylindrical Solid

Figure 13.10 illustrates a horizontally placed cylindrical solid with diameter d and length l as shown.
The cylinder is the hot body and the heat source. The fluid (air) heats up near the hot surface, becomes
less dense and moves up as shown. The Nusselt number for the horizontal cylindrical geometry given in
Figure 13.10 is computed as follows: The characteristic dimension X is given as
X =d
N = 0.47 A 0.25 (13.28)
For a turbulent flow where A > 109, the Nusselt number is determined as
N = 0.1 A 0.33 (13.29)
Vertical Cylindrical Solid

Figure 13.11 illustrates a vertically placed cylindrical solid with diameter d and length l as shown. The
cylinder is the hot body and the heat source. The fluid (air) heats up near the hot surface, becomes less dense
and moves up as shown. The Nusselt number for the vertical cylindrical geometry given in Figure 13.11 is
computed as follows: The characteristic dimension X is given as
X =l
N = 0.56 A 0.25 (13.30)
Fluid flow
Hot body d
Figure 13.10 Horizontally placed solid cylinder.
Chapter 13.indd 746 3/18/2009 6:02:42 PM

Thermal Aspects 747
Fluid
flow
l
Hot
body
Figure 13.11 Vertically placed solid cylinder.
For a turbulent flow where 109 < A < 1012, the Nusselt number is determined as
N = 0.20 A 0.40 (13.31)
From Eqs. (13.28) to (13.31), it can be observed that a solid cylinder placed vertically results in a higher
Nusselt number and consequently lower thermal resistance as compared to a horizontally placed cylinder.
Parallel Flat Plates

Consider two plates placed parallel with respect to each other as shown in Figure 13.12(a). In Figure
13.12(b), the parallel plates are at an angle of q with respect to the horizontal. The two parallel flat plates are
separated by a distance d.
The plates have a difference in temperature ΔT with respect to each other as indicated. The characteristic
dimension for the parallel plate system is
X =d
As the fluid flow within the parallel plates is not laminar, only turbulent flows are considered for this config-
uration. For slope angles, q < 50° the Nusselt number for turbulent fluid flow ( A > 105) is determined as
N = 0.062 A 0.33 (13.32)
Forced Convection
In forced convection, a controlled fluid flow is applied along the surface of the solid hot body to improve the
convective heat transfer. A fan is usually employed to create the controlled fluid flow. The fluid flow is forced
T DT
d T+
T + ΔT q
(a) (b)
Figure 13.12 Parallel plates: (a) Horizontal; (b) at a slope with respect to the horizontal.
Chapter 13.indd 747 3/18/2009 6:02:43 PM

along the surface of the solid hot body irrespective of the orientation of the hot surface. Therefore, orienta-
tion is not as significant in forced convection as is the case in free convection. In forced convection, the
Reynolds number is used to compute the Nusselt number.
Flat Plate
Consider a flat plate of any geometry as shown in Figure 13.13. A fan is used to force a fluid flow over the
hot surface which has a length a along the flow as shown.
The characteristic dimension of the flat plat is the length of the flat plate along the direction of the flow.
This is given as
X =a
For a laminar flow where R < 5 × 105, the Nusselt number is determined as
N = 0.664 × R 0.5 × (v/d )0.33 (13.33)
For a turbulent flow where R > 5 × 105, the Nusselt number is determined as
N = 0.37 × R 0.8 × (v/d )0.33 (13.34)
Solid Cylinder
Consider a cylindrical solid as shown in Figure 13.14. A fan is used to force a fluid flow over the hot surface
of the cylinder as shown.
Forced fluid flow
Hot surface
a
Fan
Figure 13.13 Forced convection over flat plate.
Hot solid
Fan
Fluid flow d
Figure 13.14 Forced convection over cylinder.
Chapter 13.indd 748 3/18/2009 6:02:43 PM

Thermal Aspects 749
The characteristic dimension of the solid cylinder is given as

X =d
For a laminar flow where 0.1 < R < 1000, the Nusselt number is determined as
N = (0.35 + 0.56 R 0.52) × (v/d )0.3 (13.35)
For a turbulent flow where 1000 < R < 5 × 105, the Nusselt number is determined as
N = 0.26 R 0.6 × (v/d )0.3 (13.36)
Problem 13.2
Consider a power BJT mounted on a flat square aluminum plate measuring 100 mm on each side. The flat
plate is part of the enclosure of the product as shown in Figure 13.15. Let the temperature within the enclosure
be 45 °C and the external ambient at a distance from the flat plate is also at 45 °C. The power BJT is operating
in the linear region and dissipates power. As a consequence, the temperature of the square aluminum plate is
70 °C. Calculate the heat flow in watts from the plate to the external ambient by convection through the air.
45°C
70°C
BJT
45°C
Figure 13.15 Problem 13.2.
Solution
From Eqs. (13.19) and (13.20), the power flow P is given by
kN
P=A ΔT
X
where A is the area of the surface perpendicular to the heat flow = 0.01 m2; N the Nusselt number; k the
thermal conductivity of the fluid (air) = 0.028 W/°K/m; X = (a + b)/2 = 0.1 m for the horizontal flat
plate; ΔT the temperature difference between the plate and the external ambient = 70 − 45 = 25°K.
Considering free convection, one needs to first compute the Rayleigh number ( A) in order to evaluate
the Nusselt number. A is given by Eq. (13.21) as
g β X 3 ΔT
A=
δ ⋅υ
(air) = 1/330°K; X the characteristic dimension for flat plat = 0.1 m; ΔT = (273 + 70)°K − (273 + 45)°K =
25°K; d the thermal diffusivity of the fluid (air) = 2.6 × 10−5 m2/s (from scientific tables); v the kinematic
viscosity of the fluid (air) = 1.8 × 10−5 m2/s (from scientific tables). Using these values,
Chapter 13.indd 749 3/18/2009 6:02:43 PM

(9.81 m/s 2 )(1 / 330o K)(0.1 m)3 25°K

A=
( 2.6 × 10−5 m 2 /s)(1.8 × 10−5 m 2 /s)
A = 1.588 × 106
Since A > 105, the flow is turbulent. The Nusselt number is given by Eq. (13.25) as
N = 0.14 A 0.33
N = 0.14 × (1.588 × 106)0.33 = 15.57

Now, substituting the value of Nusselt number in the expression for power flow (P) one obtains the rate of
heat flow or the power flow as
⎛ 15.57 ⎞
P = (0.01 m 2 )(0.028 W/m o K) ⎜ ⎟ ( 25 K ) = 1.09 W
o
⎝ 0 . 1 m ⎠
Problem 13.3
Consider a MOSFET mounted underneath a solid aluminum cylindrical heat sink as shown in Figure 13.16.
The cylinder has a diameter of 22 cm and a height of 11 cm. The ambient temperature is 20 °C and the space
within the enclosure is also assumed to be at the ambient temperature. The cylinder is at a temperature of
100 °C due to the dissipation from the heat source, that is, the MOSFET mounted underneath the cylindrical
heat sink. Calculate the energy required to maintain heat-sink temperature at 100 °C under the following
conditions:
(a) Free convection with almost still air environment for the system as shown in Figure 13.16(a).
(b) Forced convection with a fan mounted that provides an air flow of 3 m/s for the system as shown in
Figure 13.16(b).
100°C 100°C
TA = 20°C
Cylindrical
heat sink
MOSFET MOSFET
20°C 20°C
(a) (b)
Solution
Case (a): This is a free convection problem. Therefore, the Rayleigh number needs to be evaluated to cal-
culate Nusselt number. The energy from the MOSFET is transferred to the ambient due to convection
in two directions: (a) through the top portion of the cylinder and (b) through the sides of the cylinder.
Chapter 13.indd 750 3/18/2009 6:02:44 PM

Thermal Aspects 751
It is assumed that the body of the solid cylinder has negligible thermal resistance to conduction from
MOSFET to the cylinder mass as compared to the convection thermal resistance from the cylinder to the
surrounding air. The cylindrical solid may be resolved into standard surfaces to solve the problem. The top
portion of the cylindrical heat sink is considered as a circular horizontal flat plate. The sides are considered
as a cylindrical surface.
Top portion of the heat sink: The top portion is considered as a circular flat plate. The Rayleigh number is
calculated from
Atop = g β X ΔT
3
δ ⋅υ
(air) = 1/330°K; X the characteristic dimension of the circular plate which is the diameter of the plate =
0.22 m; ΔT = (273 + 100)°K − (273 + 20)°K = 80°K; d the thermal diffusivity of the fluid (air) = 2.6 ×
10−5 m2/s (from scientific tables); v the kinematic viscosity of the fluid (air) = 1.8 × 10−5 m2/s (from scien-
tific tables). Using these values,
(9.81 m/s 2 )(1/330o K)(0.22 m)3 (80o K)
Atop =
( 2.6 × 10−5 m 2 /s)(1.8 × 10−5 m 2 /s)
Atop = 5.41087 × 107

Since Atop > 105, the Nusselt number is calculated from Eq. (13.25) as
N = 0.14 A 0.33 = 0.14 (5.41087 × 107)0.33 = 49.901
The expression for the rate of heat flow is given by
N
P=A k ΔT
X
where A is the surface area of the circular plate given by A = (p/4)d 2, d being the diameter of the plate.
Substituting the appropriate values into the above power flow equation, one obtains
⎛π ⎞ ⎛ 49.901 ⎞
Ptop = ⎜ ⎟ (0.22 m)2 ⎜ ⎟ (0.028 W/m°K)(80°K) = 19.3139 W
⎝4⎠ ⎝ 0.22 m ⎠
Side portion of the heat sink: The side portion of the heat sink is considered as a vertical cylinder where the
characteristic dimension X is the height of the cylinder. Substituting the appropriate values into the equa-
tion for the Rayleigh number, one obtains
(9.81 m/s 2 )(1 / 330o K)(0.11 m)3 (80o K)
Aside = −5 −5
= 6.7636 × 106
( 2.6 × 10 m /s)(1.8 × 10 m /s)
2 2
Since 104 < A < 109, the Nusselt number is given by Eq. (13.30) as
N = 0.56 A 0.25 = 0.56 × (6.7636 × 106)0.25 = 28.5583
The surface area of the side of the cylinder is given by A = πdh, where d is the diameter of the cylinder and
h is the height. Substituting the appropriate values into the power flow equation, one obtains
⎛ 28.5583 ⎞
Pside = (π × 0.22 m × 0.11 m) ⎜ ⎟ (0.0228 W/m°K)(80°K) = 44.2133 W
⎝ 0.11 m ⎠
Chapter 13.indd 751 3/18/2009 6:02:44 PM

The total heat flow is the sum of the above two. That is
Ptotal = Ptop + Pside = 19.3139 W + 44.2133 W = 63.5272 W
Hence, the energy required to maintain the heat sink at 100°C for 1 h is
63.5272 W × 1 h = 63.5272 Wh
Case (b): This is a forced convection problem where the air flows at the velocity of 3 m/s over the cylindri-
cal heat sink. Reynolds number needs to be evaluated to find the Nusselt number. The expression for
Reynolds number is given by
uX
R=
υ
where u is the average velocity in m/s = 3 m/s; X the characteristic dimension in m and v the kinematic
viscosity of the fluid in m2/s. As in the case of free convection, the loss of energy is from the top and sides.
The loss of energy in each of these directions can be found as follows:
Top portion of the heat sink: The top portion is considered as a circular flat plate with air flow over the plate.
The characteristic dimension X of the circular plate is its diameter. The Reynolds number for this plate is
(3 m/s)(0.22 m)
R= = 0.3667 × 105
(1.8 × 10−5 m 2 /s)
Since R < 5 × 105, the flow is laminar. The Nusselt number is found from the following expression using
Eq. (13.33),
N = 0.664 × R 0.5 × (n/d )0.33
where d is the thermal diffusivity of the fluid.
Substituting the appropriate values, the Nusselt number is
0.33
⎛ 1.8 × 10−5 m 2 /s ⎞
N = (0.664)(0.3667 × 10 ) 5 0.5
⎜ −5 2 ⎟
= 112.622
⎝ 2.6 × 10 m /s ⎠
The Nusselt number is used to calculate the heat flow rate from the following expression:
N
Ptop = A k ΔT
X
where A is the area of the circular plate = π d 2 / 4 m2; k is the thermal conductivity of the fluid and ΔT is
the temperature difference. Substituting the appropriate values, one obtains
⎛π ⎞ ⎛ 112.622 ⎞
Ptop = ⎜ ⎟ (0.22 m)2 ⎜ ⎟ (0.028 W/m°K)(80 °K ) = 43.5897 W
⎝4⎠ ⎝ 0.22 m ⎠
Side portion of the heat sink: The side portion constitutes a vertical cylinder with a forced air flow over it.
The characteristic dimension X for this case is the diameter of the cylinder. Hence the Reynolds number in
this case also would be same as that calculated for the top portion.
R = 0.3667 × 105 or 3.667 × 104
Chapter 13.indd 752 3/18/2009 6:02:44 PM

Thermal Aspects 753
Since 1000 < R < 5 × 104, the flow is considered turbulent. The Nusselt number is evaluated from the fol-
lowing expression as given by Eq. (13.36):
0.3
⎛ 1.8 × 10−5 m 2 /s ⎞
N = 0.26 R 0.6
(υ / δ ) = (0.26)(3.667 × 10 )
0.3 4 0.6
⎜ −5 2 ⎟
= 127.541
⎝ 2.6 × 10 m /s ⎠
and the heat flow rate is given as
⎛ 127.541 ⎞
Pside = (π × 0.22 m × 0.11 m) ⎜ ⎟ (0.028 W/m°K)(80°K) = 98.728 W
⎝ 0.22 m ⎠
The total heat flow is the sum of the heat flow through the top and through the sides.
Ptotal = (43.5897 + 98.728) W = 142.3177 W
Hence, the energy required to maintain the heat sink at 100°C for 1 h is
142.3177 W × 1 h = 142.3177 Wh
It can be observed that the forced convection handles greater power dissipation than the free convection
for the same heat-sink dimensions and heat-sink temperature. It should be noted that in real situations
even in the presence of the fan, both the free and forced convection are active simultaneously and the
resultant total power flow is not simply an algebraic addition as indicated above. Further, it should be
observed that in the calculation of the Reynolds number, the flow over the flat plate on the top is laminar
whereas the flow on the sides of the cylinder is turbulent. The demarcation between the laminar and the
turbulent flow along the cylinder height is not very distinct and therefore the calculated results obtained as
above should be considered with a measure of uncertainty. The actual values of the Nusselt number or the
power flow should be obtained only by experiment for greater accuracy. However, the above method will
give a preliminary design that can be used to size the components of the products reasonably.
13.4 Heat Transfer by Radiation

A substance at a finite temperature will emit energy called thermal radiation. The heat energy is trans-
ported by electromagnetic waves. The transfer of heat energy by radiation does not require the presence
of a material medium. In fact, radiation transfer occurs most efficiently in vacuum.
Consider a hot body at temperature T1 as shown in Figure 13.17. Let T2 be the temperature at some
imaginary horizontal boundary parallel to the radiating surface such that T2 < T1. From Eq. (13.4), the rate
of heat energy flow is given as
T2
Radiation
T1
Figure 13.17 Heat transfer by radiation.
Chapter 13.indd 753 3/18/2009 6:02:44 PM

Table 13.2 Emittances for common radiation surfaces
Material type Emittance
Anodized aluminum (black) 0.800
Polished aluminum 0.095
Rough surface aluminum 0.180
Rough surface iron 0.170
Tungsten at 1500°C 0.330
Enamel paint 0.85
Polished copper 0.07
Glass 0.940
Wood 0.900
P = hr AΔT
where hr is the radiation thermal coefficient; A the surface area perpendicular to the radiation direction; ΔT
the temperature difference between the radiating body and the reference region.
The radiating thermal coefficient is given by the following expression:
3
⎛ T +T ⎞
hr = 4σε eff (1 − φ ) ⎜ 1 2 ⎟ (13.37)
⎝ 2 ⎠
where s is the Boltzmann’s constant = 5.67 × 10−8 W/m2/°K4; eeff the effective emittance; f the shielding
factor. The shielding factor f gives a measure of the shading of the radiative heat transfer from a hot body
due to the presence of another body in the vicinity. For a single plate or two parallel plates, the shielding
factor is zero.
The emittance is the energy emitted relative to that of a black body. Its value is less than unity. The emittance
depends on the properties such as surface type and the color of the material. Table 13.2 gives the emittance of
some of the commonly used materials. A material having an emittance of 1 is called a black body and is a
perfect radiator. It also means that the material cannot retain any heat and the heat would be lost completely
over a period. Such materials would be useless in applications where heat retention is required.
Problem 13.4
Consider two parallel plates each having area of 1 m2. Let the first plate be made up of glass and kept at a
temperature of 350 °K. Let the second plate be made up of rough surfaced aluminum. Let the temperature
of the second plate be at 300 °K. Find the radiation heat flow from the first plate to the second plate.
Chapter 13.indd 754 3/18/2009 6:02:45 PM

Thermal Aspects 755
Figure 13.18 shows the arrangement of the two plates.

1m
T2 = 300°K
e2 1m
e1 T1 = 350°K
Solution
The bottom plate is a glass plate which is at 350°K and from Table 13.2 the emittance of glass is 0.940.
The top plate is made up of rough surfaced aluminum having emittance of 0.18. The temperature of that
plate is 300°K. The radiation heat flow is from plate 1 to plate 2. The radiation thermal coefficient of the
path between plate 1 and plate 2 can be calculated from the following expression:
3
⎛ T +T ⎞ 2
hr = 4σε eff (1 − φ ) ⎜ 1 2 ⎟ W/m /°K
⎝ 2 ⎠
where s is the Boltzmann’s constant = 5.67 × 10–8 W/m2/°K4;
f the shielding factor = 0 for parallel plates;
ε1ε 2
ε eff =
ε1 + ε 2 − ε1ε 2
for parallel plates. Using the above values, the thermal coefficient is given as
3
⎛ (0.94)(0.18) ⎞ ⎛ 350°K + 300°K ⎞
hr = 4(5.67 × 10−8 W/m 2 /°K 4 ) ⎜ ⎟ (1 − 0) ⎜ ⎟
⎝ 0.94 + 0.18 − (0.94)(0.18) ⎠ ⎝ 2 ⎠
hr = 1.3855 W/m2/°K
The radiation heat flow rate can now be calculated from the following expression:
P = hr AΔT = (1.3855 W/m 2 /°K)(1 m 2 )(350°K − 300°K) = 69.275 W
This implies that 69.275 W of power is lost due to radiation from plate 1 to plate 2.
Reflectance, Absorptance and Transmittance

Whenever a wave of energy hits a surface, part of the energy gets reflected, part of the energy is absorbed by
the surface and part of the energy gets transmitted depending on the type of the surface. The amount of
energy that gets reflected depends on a factor called the reflectance of the surface denoted by r. The amount
of energy that is absorbed depends on a factor called the absorptance of the surface denoted by a. The
amount of energy that gets transmitted depends on a factor called transmittance of the surface denoted by t.
For any surface, the following relationship holds good:
Chapter 13.indd 755 3/18/2009 6:02:45 PM

Reflectance (r )
Radiation from
hot body Absorptance (a)
Surface
Transmittance (t)
Figure 13.19 Reflectance, absorptance and transmittance.
ρ +α +τ =1 (13.38)
The relationship in Eq. (13.38) is deduced from the conservation of energy principle and these constants
depend on the type of surface. These three effects are illustrated in Figure 13.19 where radiated energy from
a hot body strikes the surface.
As an example, for a glass surface, a (absorptance) is almost equal to 0; the transmittance t is high and
equal to 0.92; the reflectance r is low and equal to 0.08. The sum of these factors is equal to 1, satisfying the
above relationship.
13.5 Heat Transfer by Mass Transport
H eat transfer by mass transport also involves the solid–fluid interaction. The solid is the heat source and
the heat from it is transferred to the fluid which takes away the heat energy. The flow rate of the fluid
can be used for controlling the heat energy transfer. This method of heat transfer is used in water-cooled
heat sinks to achieve improved heat power removal.
Consider a fluid flow through a heated pipe as shown in Figure 13.20. The pipe is heated by means of a heat
source which could be the power dissipated by a power semiconductor device or any dissipative device. Let Tp
be the temperature of the pipe. As the pipe is at a higher temperature compared to the ambient, there will exist
convection and radiation from the pipe surface to the external ambient. In addition to the power transferred to
Pipe
Exit
temperature
T2
Entry Fluid flow

temperature
T1
Heat source (e.g. power switching device)
Figure 13.20 Heat transfer by mass transport.
Chapter 13.indd 756 3/18/2009 6:02:45 PM

Thermal Aspects 757
the external ambient through convection and radiation, power is transferred from the pipe to the flowing fluid
also. This section discusses the amount of heat energy that is transferred from the pipe to the flowing fluid.
Let T1 be the temperature of the fluid at the entry region of the pipe. Let T2 be the temperature at the
exit region such that T1 < T2. Let m be the mass of the fluid flowing through the pipe and dm/dt the mass
flow rate in kg/s. The heat flow rate due to the transfer of a mass m of the fluid from one place to another is
called the heat transfer by mass transport. Let Pm be the power transferred, which is given as
dm
Pm = c p (T2 − T1 ) (13.39)
dt
where cp is called the specific heat of the fluid and is expressed in J/kg/°K which is a constant for a given
fluid and dm/dt is the mass flow rate. The thermal resistance of the mass transport process can be computed
as follows: Let Rm be the equivalent thermal resistance to heat transfer by mass transport. Then from
Eq. (13.1),
T2 − T1 = Pm Rm (13.40)
Therefore
T2 − T1
Rm = (13.41)
Pm
Substituting the value of Pm from Eq. (13.39) into Eq. (13.41), one obtains
T2 − T1 1
Rm = = (13.42)
(dm /dt )c p (T2 − T1 ) (dm /dt )c p
It can be observed here that temperature is not the driving function for the heat transfer unlike conduction,
free convection and radiation. The heat flow rate is determined by external factors controlling the rate of
mass flow (dm/dt).
The most effective means of heat transfer by mass transport is through the latent heat of vaporization of the
fluid. Latent heat of vaporization is the amount of energy required to vaporize 1 kg of the fluid that is already at its
boiling point. It is denoted by L. To vaporize 1 kg of water, 2.4 MJ of heat is required whereas to heat water
through 100°C, only 0.42 MJ is required. A schematic of a water-cooled system is shown in Figure 13.21. The
heat source is a power semiconductor device that dissipates heat. This heats up the water in the chamber
connected to the heat source to a temperature T1. When the water temperature reaches boiling point, then T1
Heat
sink T2
T1
Heat source
Product enclosure
Figure 13.21 Water-cooled system.
Chapter 13.indd 757 3/18/2009 6:02:45 PM

is at 100°C which is maintained as constant. Water takes the heat from the source; evaporates and carries the
heat energy corresponding to the latent heat of vaporization with it; and transfers to the sink which is at tem-
perature T2. The sink is maintained at temperature T2 by means of cold water or ice or peltier junctions. After
giving the heat to the sink, the water vapor again converts back to liquid form and flows into the collection
chamber which then links to the evaporation chamber as shown. Thus, it can be observed that heat taken from
the heat source at T1 is carried to wherever the vapor condenses at T2. The associated heat flow rate is given by
dm
Pm = Λ (13.43)
dt
where dm/dt is the rate at which fluid is being evaporated and L is the latent heat of vaporization of the
fluid. The associated thermal resistance is given as
T1 − T2 T −T
Rm = = 1 2 (13.44)
Pm (dm / dt )Λ
13.6 General Notes on Heat Transfer

Mixed Heat Transfer Mechanisms
It should be noted that even though the various heat transfer mechanisms have been discussed in an isolated
manner wherein one mode of heat transfer is decoupled from the other mechanism, in reality some or all of
the heat transfer mechanisms may be active simultaneously in a thermal system. For example, in the case of
the mass transport mechanism discussed in the previous section, the pipe through which the fluid flows is at
a higher temperature with respect to the external ambient. This implies that there will exist convective and
radiative heat energy-transfer mechanisms from the hot pipe to the ambient. Therefore, one will have to
consider all the effects together. The general rate of heat energy flow is given by Eq. (13.1). Let Pc, Pv, Pr and
Pm be the heat power flow through the conductive, convective, radiative and mass transport mechanisms,
respectively. If Rθ c , R θ v , Rθr and Rθm are the thermal resistance for the conductive, convective, radiative
and mass transport mechanisms, respectively, then
Ptot = Pc + Pv + Pr + Pm (13.45)
ΔTc ΔTv ΔTr ΔTm

Ptot = + + + (13.46)
R θc R θ v R θr R θm
Equation (13.45) follows from the conservation of energy principle. In Eq. (13.46), ΔTc , ΔTv , ΔTr and ΔTm
are the temperature differences between the hot body and the lower temperature references for the conduc-
tive, convective, radiative and mass transport mechanisms. If the temperature differences are same for all the
mechanisms, then Eq. (13.46) can be written as
⎛ 1 1 1 1 ⎞
Ptot = ⎜ + + + ⎟ ΔT (13.47)
R
⎝ θc R θv R θr R θm ⎠
or if the area perpendicular to the heat power flow is same for all mechanisms, then the power flow in terms
of the thermal coefficients for the various heat transfer mechanisms can be written as
Ptot = ( hθ c + hθ v + hθr + hθm ) AΔT (13.48)
Chapter 13.indd 758 3/18/2009 6:02:46 PM

Thermal Aspects 759
⎛ k kN ⎛ T + T ⎞ (dm / dt )s ⎞
3
Ptot = ⎜ + + 4σε eff (1 − φ ) ⎜ 1 2 ⎟ + ⎟⎟ AΔT (13.49)
⎜ Δx X ⎝ 2 ⎠ A
⎝ ⎠
Thus, the total power flow including all the heat transfer effects is given as
Ptot = heq AΔT (13.50)
where
⎛ k kN ⎛ T + T ⎞ (dm / dt )s ⎞
3
heq = ⎜ + + 4σε eff (1 − φ ) ⎜ 1 2 ⎟ + ⎟⎟
⎜ Δx X ⎝ 2 ⎠ A
⎝ ⎠
It should be noted that Eq. (13.50) is an algebraic summation and sometimes some of the power flow
mechanism may have a canceling effect on the power flow by other mechanism. However, Eq. (13.49)
would in general give a conservative value of the equivalent thermal coefficient that can be used for design
purposes.
Thermal Resistance of Varying Cross-Sections

Consider the heat transfer by conduction mechanism. The cross-section area A that is orthogonal to the heat
power flow direction is considered constant till now. However, there are cases where A varies with distance.
The power flow equation is given as
ΔT dT
P = kA = kA
Δx dx
P
dT = dx (13.51)
kA
Integrating Eq. (13.51) between temperature limits T2 and T1 and the length from 0 to L, one obtains
T1 L
P
∫ dT = ∫ kA
dx
T2 0
L
dx
ΔT = P ∫ (13.52)
0
kA
L
dx
Rθ = ∫
0
kA
Equation (13.52) is a general equation where A can vary along the path having changing cross-sections.
Further, if the solid material is a eutectic combination of different materials, then k also varies along the
path.
Consider a system as shown in Figure 13.22. The heat source is mounted at the center of a solid block
that is either rectangular or circular as shown. The heat power flow radiates in all directions from the center
as shown. At a greater distance from the heat source, the cross-section area is larger as the radius is larger.
The cross-section area at a radius r is given as 2prb. Thus,
Chapter 13.indd 759 3/18/2009 6:02:46 PM

dr
d
P
b
D
(a) (b)
Figure 13.22 Non-uniform cross-section area: (a) Rectangular block; (b) increasing radius in the
direction of power flow.
D /2
dr
Rθ = ∫ k ( 2πrb )
d /2
1 D /2 1 ⎛D⎞
= ln r d / 2 = ln ⎜ ⎟ (13.53)
2π kb 2π kb ⎝ d ⎠
It should also be noted that in the case of a flat plate, if both the surfaces are exposed to the ambient, then
the heat power flows to the ambient on both sides of the flat plate. In such cases, area A is taken as twice the
area of one side of the flat plate to account for the power flow from both surfaces for any heat transfer
mechanism.
Mounting of Power Devices on Heat Sink

Special care should be taken during mounting of the power semiconductor devices on the heat sink to mini-
mize the thermal resistance from the case of the device to the heat sink, that is, Rqcs. The surface of the power
device case and the surface of the heat sink have micro-undulations which reduce the contact between the
two surfaces as shown in Figure 13.23(a). As a consequence, the contact thermal resistance will increase
thereby increasing the case-to-sink thermal resistance. To improve the contact between the device case sur-
face and the heat-sink surface, a thin layer of heat-sink paste or heat-sink compound is applied over the
heat-sink surface to fill up the micro-undulations as shown in Figure 13.23(b). This improves the contact
between the two surfaces. Care must be taken to ensure that the heat-sink paste should be applied in a very
thin layer just enough to fill up the micro-undulations. Thicker layer of heat-sink paste will also increase the
overall thermal resistance.
In some applications, more than one device needs to be mounted on the same heat sink. The case of the
power devices are generally connected to one of the device leads, that is, the collector is also the case in the
case of the BJT, the drain is also the case in the case of MOSFETs, etc. In such cases, to prevent shorting of
the terminals of the devices through the heat sink, each power device is mounted on the heat sink with a
thin sheet of insulating mica layer interposed between the device and the heat-sink surface as shown in
Figure 13.23(c).
Chapter 13.indd 760 3/18/2009 6:02:46 PM

Thermal Aspects 761
Power device
Power device
Micro
air pockets
Heat sink
Heat sink
Heat sink
compound
(a) (b)
Power device
Mica layer
(insulation)
Heat sink
Heat sink
compound
(c)
Figure 13.23 Power device mounted on heat sink: (a) Without heat-sink compound;
(b) after applying heat-sink compound; (c) after applying heat-sink
compound and inserting a layer of mica for insulation.
Table 13.3 Case-to-sink thermal resistance for TO-220 and TO-3 device package
Rq cs in °K/W
Direct mount Mount with mica insulator
Without heat-sink With heat-sink Without heat-sink With heat-sink

Device package compound compound compound compound
TO-220 1.4 0.5 3 1.4
TO-3 0.6 0.1 1.25 0.5
Table 13.3 gives typical value of the case-to-sink thermal resistance for two of the popular device pack-
ages with a torque of 0.4–0.6 Nm applied on the device mounting screws.
Transient Model for Heat Transfer

In most of the applications, the steady-state model is sufficient to design or select the heat sink to achieve
proper heat transfer from the device to the external ambient. However, in some applications, the load is
pulsed and the dissipated power is also of a pulse nature. In such cases, the heat-sink size and volume may
Chapter 13.indd 761 3/18/2009 6:02:46 PM

be further optimized if the dynamics of the temperature rise and fall are taken into account. For the thermal
domain, the mass of the solid or the fluid acts as the capacitance which stores the thermal energy by virtue
of the raised temperature. The heat energy from the heat source is transferred into two sinks: (a) the external
ambient which is the infinite sink and (b) the mass of the medium through which the heat transfers. The
energy transferred into the medium is called the stored thermal energy. Thus,
E input = E ext-ambient + E stored
and the power flow is given as

dE input dE ext-ambient dE stored
P= = + (13.54)
dt dt dt
dT dT
P= kA + mc p (13.55)
dx
t
d
Steady-state part Transient part
where m is the mass of the medium in which heat is stored; cp the specific heat of the medium in which heat
is stored. The term mcp is equivalent to the capacitance in the electrical circuit. Equation (13.55) gives the
dynamic equation that defines the heat energy flow rate.
Best Method of Heat Transfer

There is no single best method of heat transfer for removing heat from the power device to the external
ambient. Each of the heat transfer mechanism has its inherent advantages and disadvantages. However, in
general, the lower the thermal resistance required, more expensive will be the heat transfer system. The fol-
lowing illustration in Figure 13.24 gives a general guideline on the heat transfer method that can be used for
various heat source power levels.
Natural
Rq convection
+ radiation
Forced air cooling
Forced liquid cooling
Liquid evaporation
0.01 0.1 1 10 100 Heat specific

power, W/cm2
Figure 13.24 Relative effectiveness of different heat transfer mechanisms.
Chapter 13.indd 762 3/18/2009 6:02:47 PM

Thermal Aspects 763
Referring to Figure 13.24, it can be observed that for low specific heat power flows, the thermal resistance
obtainable from natural or free convection and radiation is sufficient to maintain the junction temperature of
the semiconductor devices within the safe temperature limits. For very high specific heat power, the mass
transport with liquid evaporation (similar to Figure 13.21) is the most effective in reducing the thermal
resistance.
In the chapters discussed till now, the focus has been limits, the heat sinks are used to conduct away the
essentially on addressing electrical parameters and heat from the junction to the ambient. Therefore,
stresses. However, this is only half the job. Unless the heat-sink design would mean the estimation of
the thermal stress aspect is addressed, the design of the required sink to ambient thermal resistance that
power electronics systems is never complete. This is needed to transfer the heat from the device case to
chapter addresses this important issue of keeping the the ambient such that the junction temperature is
thermal stresses in the power semiconductor devices within safe limits.
within safe limits. Heat sinks are available commercially in the
The power semiconductor devices generate heat form of aluminum extrusions. There are numerous
due to the conduction loss and the switching loss as aluminum extrusion profiles that are available that
discussed in Chapter 1. Therefore the power semi- enables a designer to meet both the thermal aspects
conductor devices behave as heat sources. If the heat and the product form requirements. The next sec-
generated by the devices are not conducted away to tion gives few tutorial tasks that should necessarily
the ambient, the temperature of the device junction be performed by prospective designers in order to
will increase and damage the device. In order to achieve a certain measure of skill and confidence in
ensure that the junction temperature of the semi- handling the thermal stress aspects.
conductor devices are maintained within permissible
1. Heat source circuit setup: In order to study the the operating point of the BJT is as shown in
thermal aspects, one needs a controllable heat Figure 13.25b. The power dissipation in the
source to experiment with the heat transfer BJT is given by the product of IC and VCE. By
mechanisms. The circuit shown in Figure 13.25a adjusting Ib, any desired power dissipation
consists of a BJT that is operated in the linear can be achieved. Therefore, the BJT operated
region. The base source Vb and the collector in the linear region becomes a controllable heat
supply Vc are dc sources. The base drive resistor source.
Rb is a variable resistor. Adjust either Vb or Rb to
allow a pre-determined base current, Ib. This will
can be studied by
result in a collector current depending on the
a. Hardware bread-boarding
hFE of the BJT. The base current is set such that
Chapter 13.indd 763 3/18/2009 6:02:47 PM

Vc
ic
Rc
Vc
Ic Rc
Rb Ib Operating
Q Vce point
Ic
Vb
Pd = Vce·Ic
VCE Vc Vce
(a) (b)
Figure 13.25 BJT test circuit.
Tasks for study: Tasks for study:

(a) Set Vb to be a particular dc value. Measure (a) Set the heat source power to a particular
the collector current and the collector– value. Allow the system to reach a steady state
emitter voltage across transistor for diffe- in terms of the BJT case temperature for hor-
rent values of Rb. Tabulate the BJT collector izontal position of the aluminum plate. From
current, base current, emitter current and the datasheet of the BJT, obtain the junction-
collector–emitter voltage values. Calculate to-case thermal resistance value.
the product of IC and VCE which is the (b) Measure and tabulate the heat source
power dissipation of the BJT. power, the BJT case temperature, the temp-
(b) For every setting of Ib, allow the case erature of the aluminum plate near the
temperature of the BJT to reach the equili- heat source and also at the edges. Measure
brium state. (Allow the circuit to operate and tabulate the room temperature also.
for around 15 min to reach steady state.) (c) From the above data, calculate the case-to-
Also measure the room temperature or the sink thermal resistance and sink-to-ambient
ambient temperature at every power- thermal resistance.
dissipation setting. Tabulate the above (d) Estimate the junction temperature of the
readings. BJT heat source.
(The above two tasks are for understanding the (e) Repeat tasks (b) to (d) by mounting the
heat power source.) BJT heat source with and without heat-
sink compound. Tabulate and compare the
2. Thermal resistance measurements: Mount
thermal resistances.
the heat source BJT, as described in Tutorial 1,
(f ) Theoretically calculate the aluminum sink
at the center of a flat rectangular aluminum
to ambient thermal resistance due to con-
plate of known dimensions.
vection and radiation heat transfer mecha-
Mode of implementation: nisms. Compare the theoretical estimation
a. Hardware bread-boarding with the experimental results.
Chapter 13.indd 764 3/18/2009 6:02:47 PM

Thermal Aspects 765
(g) Now repeat tasks (b) to (f ) for vertical ori- (h) Repeat tasks (b) to (g) after including a fan
entation of the rectangular plate. The rect- for providing forced convection. Perform
angular plate can be placed vertical in two the experiment for different fan speeds (air
ways, (i) smaller side vertical and (ii) longer velocities), tabulate and plot the sink-to-
side vertical. Conduct the experiment for ambient thermal resistance versus fan
both orientations. speeds or air velocities.

1. The junction temperature should not exceed 11. If there is a heat flow through a solid, then it
the junction temperature which is implies a temperature along the length
for most devices. of the solid slab in the direction of the .
2. Conduction is a process of heat transfer 12. Thermal conductivity is a property and
wherein the heat flows through from is the characteristic of the solid slab material.
a higher temperature zone to a lower tempe-
13. The thermal resistance of silver is than
rature zone.
that of steel.
3. Convection is the process of heat transfer from
14. The convection heat transfer mode is composed
solids by means of .
of following two mechanisms: (a) and
4. Radiation is a process of heat transfer, wherein (b) of the fluid.
the heat flows out from the hot body without
15. The flow induced by diffusion which arises
the need for a heat transport .
from density differences caused by temperature
5. Convection is an process where as variations in the fluid is called convec-
mass transport is a controlled process where tion.
the discharge rate of the fluid can be controlled
16. The flow caused by external means such as fan,
resulting in heat transfer from one
pump or winds is called convection.
region to another.
17. The Nusselt number is a quantity.
6. With reference to the heat flow, the
difference is considered equivalent to the 18. Nusselt number is a function of the of
potential difference of electrical circuit. fluid flow, of the hot solid and the
properties of the fluid medium.
7. The rate of heat flow or power flow is considered
equivalent to the of electrical circuit. 19. Nusselt number also depends on the type of
convection, whether or .
8. There is no concept of in the thermal
domain. 20. Nusselt number in the case if free convection is
estimated using the number.
9. The thermal resistance is not a
element, it is only a notional element that 21. Nusselt number in the case of forced convec-
indicates the conductivity of heat through the tion is estimated using the number.
medium.
22. In forced convection, a fluid flow is
10. is sometimes referred to as the ohm’s applied along the surface of the solid hot body
law for thermal domain. to improve the convective heat transfer.
Chapter 13.indd 765 3/18/2009 6:02:47 PM

23. The convective thermal resistance is dependent 30. The value of emittance is less than .
on the of the hot solid surface in the
31. Heat transfer by mass transport involves the
case of free convection.
interaction.
24. The convective thermal resistance is of
32. The is the heat source and the heat
the orientation of the hot solid surface in the
from the source is transferred to the
case of the forced convection.
which takes away the heat energy.
25. A substance at a finite temperature will emit
33. The flow rate of the fluid can be used for
energy called radiation that is trans-
the heat energy transfer.
ported by waves.
34. The heat energy from the heat source is trans-
26. The transfer of heat energy by radiation occurs
ferred into two sinks, (a) the external ambient
most efficiently in .
which is the sink and (b) the mass of
27. The shielding factor gives a measure of the the medium through which the heat transfers.
of the radiative heat transfer from a
35. The energy transferred into the mass of the
hot body due to the presence of another body
medium is called the thermal energy.
in the vicinity.
36. The term is equivalent to the capaci-
28. For a single plate or two parallel plates, the
tance in the electrical circuit.
shielding factor is .
37. Mass transport by evaporation gives the lowest
29. The emittance is the energy emitted relative to
to the heat flow among the four heat
that of a .
transfer mechanisms.
1. What are the four types of heat transfer 7. Explain why there is a measure of uncertainty
mechanisms? in the estimation of the convective thermal
resistance.
2. Write short notes on: (a) conduction, (b) con-
vection, (c) radiation and (d) mass transport. 8. How is Rayleigh number evaluated for a fluid?
3. Write short notes on: (a) specific heat flow rate, 9. How is Reynolds number evaluated for a fluid?
(b) thermal resistance, (c) thermal resistivity,
10. A flat plate that is oriented vertically has a lower
(d) thermal coefficient and (e) thermal conduc-
thermal resistance as compared to the same flat
tivity.
plate that is oriented horizontally.
4. What is the heat flow rate equation? Discuss.
11. Discuss the determination of Nusselt number
5. Write short notes on: (a) Nusselt number, for few standard geometries of the hot solid for
(b) Rayleigh number and (c) Reynolds free convection.
number.
12. Discuss the determination of Nusselt number
6. Write short notes on: (a) free convection and for few standard geometries of the hot solid for
(b) forced convection. forced convection.
Chapter 13.indd 766 3/18/2009 6:02:47 PM

Thermal Aspects 767
13. Write short notes on: (a) reflectance, 15. Discuss the mounting considerations for the
(b) absorptance and (c) transmittance. power devices on the heat sink.
14. How does heat transfer by mass transport differ 16. Discuss the choice of the heat transfer mecha-
from heat transfer by convection? nism for any given application.
| PROBLEMS
1. Calculate thermal resistance to the flow of square aluminum plate is 70οC. Heat flows to
power through a solid square slab of 100 cm the ambient from both faces of the aluminum
length wherein plate. Calculate the heat flow rate in watts from
a. the slab is made up of aluminum with the plate to the external ambient by convection
thickness of 10 mm. through the air.
b. the slab is made up of glass with thickness
5. For Problem 4, if a fan is placed at the bottom
of 5 mm.
of the vertical plate such that air flows up along
c. the slab is made up of steel with thickness
both sides of the vertical plate at a velocity of
of 2 mm.
2 m/s, then estimate the convective thermal
2. A MOSFET that is used in a dc–dc converter is resistance.
dissipating 50 W. The thermal resistance to
6. Consider the problem given in Problem 13.3
conduction from the junction to the case is
which is illustrated in Figure 13.16. The speci-
0.5οK/W and the thermal resistance to conduc-
fications are modified as follows. The solid
tion from the case to the heat sink is 1.5οK/W.
cylindrical heat sink has a diameter of 22 cm
If the ambient temperature in the neighbor-
and a height of 11 cm. The ambient temper-
hood of the heat sink is 50οC, then calculate
ature is 20οC and the space within the enclo-
the thermal resistance requirement for the heat
sure is at 45οC. The cylinder is at a temperature
sink if the junction temperature does not exceed
of 100oC due to the dissipation from the heat
(a) 100οC, (b) 120οC.
source, that is, the MOSFET mounted under-
3. A MOSFET that is used in an inverter is dissi- neath the cylindrical heat sink. The dimen-
pating 20 W. The thermal resistance to conduc- sions of the cuboidal steel enclosure are
tion from the junction to the case is 0.5οK/W 100 cm × 100 cm × 10 cm. There is conduc-
and the thermal resistance to conduction from tive heat flow from the MOSFET source to
the case to the heat sink is 1.5οK/W. The heat the heat sink; and from the heat sink to the
sink used ensures that the device junction ambient there is both convective and radiative
temperature does not exceed 100οC. If the heat flow. There is conductive, convective and
ambient temperature in the neighborhood of radiative heat flow from the enclosure to the
the heat sink is 50οC, then estimate the device ambient. Considering all these heat flows, cal-
case temperature. culate the power flow from the heat source
(MOSFET) that is required to maintain the
4. Consider a hot flat square aluminum plate
heat-sink temperature at 100οC under the fol-
measuring 100 cm on each side that is placed
lowing conditions:
vertically. The temperature at a distance from
the flat plate is 45οC. The temperature of the
Chapter 13.indd 767 3/18/2009 6:02:48 PM

a. Almost still air environment for the 8. Derive the thermal resistance due to mass
system. transport mechanism of heat transfer by evapo-
b. Forced convection with a fan mounted that ration of water.
provides an air flow of 3 m/s for the heat
9. Consider a circular heat source of diameter
sink.
10 mm mounted at the center of a circular
7. Consider two parallel plates each having an aluminum plate of diameter 100 cm and thick-
area of 1 m2. Let the first plate be made up of ness 15 mm. What is the temperature gradient
anodized aluminum and kept at a temperature between the circumference of the circular plate
of 373οK. Let the second plate be made up of and the heat source?
rough surfaced aluminum and the temperature
of the plate be 300οK. Find the radiation heat
flow from first plate to the second plate.
| ANSWERS
Fill in the Blanks
1. maximum-rated; 150οC 14. diffusion; bulk motion 27. shading
2. solids 15. free 28. zero
3. fluids 16. forced 29. black body
4. medium 17. dimensionless 30. unity
5. uncontrolled; controlled 18. velocity; geometry 31. solid–fluid
6. temperature 19. free; forced 32. solid; fluid
7. current flow 20. Rayleigh 33. controlling
8. dissipation 21. Reynolds 34. infinite
9. dissipative 22. controlled 35. stored
10. ΔT = P ⋅ Rθ 23. orientation 36. (m . cp)
11. gradient; heat flow 24. independent 37. thermal resistance
12. transport 25. thermal; electromagnetic
13. lower 26. vacuum
Chapter 13.indd 768 3/18/2009 6:02:48 PM

Reliability Modeling
Learning Objectives
CHAPTER
14
understand the need for reliability.
understand the concepts of reliability and reliability modeling.
apportion the reliability and specify reliability for various modules.
design for reliability.
T here are many constraints that need to be considered in the overall design of a system. One such
constraint is a set of performance requirements that needs to be addressed. Apart from performance,
cost is another important constraint that needs to be addressed in the design of the system. The third aspect
is the life constraint that considers aspects related to the life of the product. In most cases, the three constraints
may be contradictory to each other. For example, a high-accuracy performance and a low-cost product may be
contradictory to each other. Likewise, long life and low cost are also contradictory by nature. In general, any
product may be said to be a compromise among the three primary constraints: (a) performance, (b) cost and
(c) life (Figure 14.1).
Performance Cost
Product
Life
Figure 14.1 The three primary constraints in designing a product.
Chapter 14.indd 769 3/20/2009 7:26:28 PM

A concept of the product that achieves an optimal compromise among the three primary constraints of
performance, cost and life is said to be a good design. The previous chapters discuss about aspects related to
the performance of the power electronic systems. However, a power electronic system design or analysis is
not only about performance. This chapter focuses on the life aspect and discusses issues related to reliability
of the products and the process of integrating reliability requirements into the design of the system.
One can appreciate the significance of designing for reliability when one visualizes the power electronic
system or any product development process. A typical product development process is illustrated in Figure 14.2.
The first task is to define the system in terms of the specifications. The specifications should incorporate the
requirements of all the three primary constraints, that is, performance, life and cost. Specification for
reliability is in terms of failure rate or mean time to failure (MTTF). The specification process will result in
a set of target specifications that needs to be achieved by the design and development process. The system is
designed to meet the performance specifications followed by validation of the design through simulation
and bread-boarding. This is followed by the reliability modeling of the system and reliability prediction to
estimate the life of the product. This is an important process as the guarantees, warranties and service infra-
structural requirements are decided by achieving this constraint. This is followed by cost analysis and cost
optimization. There is a feedback loop on completing the process related to each of the primary constraints.
The system design has to be modified or fine-tuned to realize all the specification requirements.
The previous chapters focused on the performance design process whereas this chapter will focus on the
reliability modeling issues. The next chapter will discuss reliability prediction in detail and the method of
integrating the reliability requirements into the design process to achieve better design cycle time.
Specifications
(performance, cost, life)
Performance
design
Validate
Reliability modeling
Reliability prediction
Cost optimization
Pre-production
Figure 14.2 Design process for a product.
Chapter 14.indd 770 3/20/2009 7:26:29 PM

Reliability Modeling 771
14.1 Reliability Concepts
T he problem of assuring reliability has many facets including equipment design, control of quality during
production, acceptance inspection, field trials, life testing and design modifications. The various elec-
tronic components are subjected to two types of stresses: (a) electrical stress and (b) temperature stress. The
electrical stress may be in terms of voltage-withstanding capability of the device or the current-handling
capability of the device. The temperature stress are due to the internal dissipation which will raise the junc-
tion temperature of the semiconductor devices. The component fails due to the application of these two
types of stresses during operation. However, the failure of a component is a random process as the exact
cause and reason for the failure is uncertain. Therefore, reliability is defined as the probability that an item will
perform its intended function for a specified period of time under the stated conditions. Thus, it can be seen from
this definition that reliability encompasses both performance (intended function) and life (specified period
of time) requirements.
Since reliability has been defined as a probability, the quantitative analysis is based on probability-based
tools. For the purpose of reliability discussion in this chapter, let T be a continuous random variable that
denotes the time at which failure of an item occurs. Therefore, according to the definition of the continuous
random variable, there must exist a function f (t ), called the failure-time probability density function. There-
fore, one has
f (t ) ≥ 0, 0 < t < ∞
∞ ∞
∫ f (t )dt = ∫ f (t )dt = 1
−∞ 0
b
P( a < T < b ) = ∫ f (t )dt
a
t
P(T < t ) = ∫ f (t )dt = F (t )
0
dF (t )
= f (t )
dt
One should note that F (t ) = P(T < t ) denotes the probability that the random variable T takes on values
less than t, that is, it denotes that failure occurs in the time interval (0, t). Evidently, 1 – F(t) denotes that the
failure does not occur in the time interval (0, t). This is called reliability R(t). Thus the reliability of an item
can be mathematically defined as
R(t) = 1 – F(t) (14.1)
Failure Rate
Reliability has been described as quality in the time dimension. It is classically defined as the probability that
an item will perform satisfactorily for a specified period of time under a stated set of use conditions. From a
functional point of view, for an item to be reliable, it must do more than meet an initial factory performance
or quality specification and must also operate satisfactorily for an acceptable period of time in field applica-
tion for which it is intended. Determining reliability, therefore, involves understanding of several concepts
which relate to the four elements in the classical definition, that is, probability, performance, time and stated
Chapter 14.indd 771 3/20/2009 7:26:29 PM

use conditions. Among such concepts is that of the failure rate which can vary as a function of time as
discussed earlier. A failure rate is a measurement of the number of malfunctions occurring per unit time.
Referring to Figure 14.3, let E1 be an event that indicates the occurrence of failure in the time interval
(0, t). Then E1 is the event that failure does not occur in the time interval (0, t). Let E2 denote the event
that the failure occurs in the interval (t, t + Δt).
Clearly, E1 and E2 are mutually exclusive events which are indicated in Figure 14.3(b). It is also evident
from the figure that E1 and E2 are not mutually exclusive events. One is interested in the probability of
occurrence of a failure in the interval (t, t + Δt). Thus, one has to evaluate the probability of occurrence of
event E2 on the condition that E1 has not occurred or in other words, the probability of occurrence of event
E2 on the condition that E1 has occurred. Thus,
P( E 2 ∩ E1 )
P(E 2 / E1 ) = (14.2)
P( E1 )
From Figure 14.3(b), it is evident that event E2 is wholly contained in E1 . Thus,
P( E 2 ∩ E1 ) = P(E 2 ) (14.3)
P( E 2 )
P(E 2 / E1 ) = (14.4)
P( E1 )
P(E 2 ) = P(t < T < t + Δt ) = F (t + Δt ) − F (t ) (14.5)
P( E1 ) = 1 − P( E1 ) = 1 − P(T < t ) = 1 − F (t ) = R (t ) (14.6)
E1 E2
Time
t t + Δt
(a)
E1
EE11 E2
(b)
Figure 14.3 Events to describe the failure-time distribution.
Chapter 14.indd 772 3/20/2009 7:26:29 PM

Substituting Eqs. (14.5) and (14.6) in Eq. (14.4), the probability that an item will fail in the interval (t, t + Δt)
on the condition that the component has survived till time t is given by
F (t + Δt ) − F (t )
P( E 2 / E1 ) = (14.7)
R (t )
The probability in the case of continuous sample spaces is the area under the probability density function
within the given interval. Referring to Eq. (14.7), dividing it by the base Δt, the average rate of failure in the
interval t to t + Δt on the condition that the item has survived till time t is given by
P( E 2 / E1 ) F (t + Δt ) − F (t ) ⎛ 1 ⎞
= ⎜ R (t ) ⎟ (14.8)
Δt Δ(t ) ⎝ ⎠
Taking the limit as Δt → 0, the instantaneous failure rate or simply the failure rate denoted by Z(t) is given
by
⎡ F (t + Δt ) − F (t ) ⎛ 1 ⎞⎤
Z (t ) = lim ⎢ ⎜ R (t ) ⎟⎥ (14.9)
Δt →0 ⎣ R (t ) ⎝ ⎠⎦
F ′(t )
Z (t ) = (14.10)
R (t )
where
dF (t )
F ′(t ) = = f (t )
dt
f (t ) f (t )
Z (t ) = = (14.11)
R (t ) 1 − F (t )
One should note that in several literature, the failure rate Z(t) is also referred to as the hazard rate. Here, the
terms failure rate and the hazard rate will be used interchangeably.
Failure-Time Probability Density Function

From the definition of reliability as indicated by Eq. (14.1), the following relationship can be obtained by
differentiating Eq. (14.1):
R ′(t ) = − F ′(t ) (14.12)

R ′(t ) d[ln R (t )]
Z (t ) = − =− (14.13)
R (t ) dt
Thus integrating Eq. (14.13), one obtains
t
∫ Z (t )dt = − ln R(t ) (14.14)

0
Equation (14.14) can be re-written as
t
R (t ) = e ∫0
− Z (t )dt
(14.15)
Chapter 14.indd 773 3/20/2009 7:26:30 PM

From Eq. (14.10) it is evident that f (t ) = Z(t)R(t). Thus, from Eq. (14.15), the failure-time probability den-
sity function is given by t
f (t ) = Z (t ) e ∫0
− Z (t )dt
(14.16)
Equation (14.16) is an interesting relationship which shows that the failure-time probability density func-
tion is completely determined by the failure rate or the hazard rate Z(t) for a particular process. Consider, for
example, that the failure rate for some system is a constant l. The failure-time probability density function
can be uniquely determined by using Eq. (14.16). Thus,
t
f (t ) = λe ∫0 = λe − λt
− λdt
(14.17)
Clearly, it is observed that when the failure rate is constant, the resulting failure-time probability density
function has the characteristics of an exponential distribution. In some cases, one may encounter a distribu-
tion where the failure rates may not be a constant but can increase with time or decrease with time. Therefore,
one can use the following general model for the failure rate,
Z (t ) = λβ t β −1 (t > 0) (14.18)
It is evident from Eq. (14.18) that if b = 1, then the failure rate is a constant. If b < 1, then the failure rate
decreases with time. On the other hand, if b > 1, the failure rate increases with time. Using the failure rate
model of Eq. (14.18) in Eq. (14.16), the failure-time probability density function is given as
β
f (t ) = λβt β −1e − λt (14.19)
The distribution given in Eq. (14.19) is called the Weibull distribution where the exponential distribution is
a special case at b = 1.
To show the variation in the failure rate, separate consideration is given to three discrete periods when
viewing the failure characteristics of a product or item over its life span. These periods, which are indicated
in Figure 14.4, are
z (t )
Quality
failures
Wearout
Equivalent life period
failure rate curve
Useful life
Wearout
failures
Infant
Stress-related failures
mortality
Figure 14.4 Failure rate curve showing the failure rate during the life of an electronic equipment.
Chapter 14.indd 774 3/20/2009 7:26:30 PM

1. infant mortality period;

2. useful life period;
3. wear out period.
Infant Mortality Period

Initially, the item population exhibits a high failure rate. This failure rate decreases rapidly during this first
period, often called the infant mortality, burn-in or debugging period. The failure rate then stabilizes when
the weak units have died out. The failures in this period are generally related to quality. It may be caused by
a number of things: gross built-in flaws due to faulty workmanship (manufacturing deviations from the
design intent), transportation damage or installation errors. This initial failure rate is unusually pronounced
in new equipment. Many manufacturers provide a burn-in period for their product prior to delivery as this
helps to eliminate a high portion of the initial failures and assists in establishing a high level of operational
reliability. Examples of early failures are:
1. poor welds or seals;
2. poor solder joints;
3. poor connections;
4. dirt or contamination on surfaces or in materials;
5. chemical impurities in metal or insulation;
6. voids, cracks, thin spots in insulation or protective coatings;
7. incorrect positioning of parts.
Many of these early failures can be prevented by improving the control over the manufacturing process.
Sometimes, improvements in design or materials are required to increase the tolerance for these manufactur-
ing deviations, but fundamentally these failures reflect the manufacturability of the component or product
and the control of the manufacturing process. Consequently, these early failures would show up during
1. in-process and final tests;
2. process audits;
3. life tests;
4. environmental test.
Useful Life Period

The item population, after having been burned-in, reaches its lowest failure rate level which is normally
characterized by a relatively constant failure rate. This period is characterized by the occurrence of stress-
related failures. As indicated in Figure 14.4, the stress-related failures have a constant failure rate. Thus the
exponential failure-time probability density function is widely used as a mathematical model to approximate
this time period. This period is the interval which is given the most weightage in reliability design and is the
most significant period for reliability prediction and assessment activities.
Wear Out Period

The third and the final life period occurs when the item population reaches the point where the failure rate
starts to increase noticeably. This point is identified as the end of the useful life and the start of wear out or
ageing. Beyond this point on the time axis, the failure rate increases rapidly. When the hardware failure rate
due to wear out becomes unacceptably high, replacement or repair of the item should be made. Replacement
schedules are based on the recognition of this failure rate.
Chapter 14.indd 775 3/20/2009 7:26:30 PM

Wear out failures as shown in Figure 14.4 are primarily due to deterioration of the strength of the device
as a consequence of operation and exposure to environmental fluctuations. Deterioration results from a
number of familiar chemical and physical phenomena:
1. corrosion or oxidation;
2. insulation breakdown or leakage;
3. frictional wear or fatigue;
4. shrinkage and cracking in plastics.
Mean Time Between Failure (MTBF), Mean Time to Failure (MTTF)

The period of interest during the life cycle of a product or a system is the useful life period. During this
period, the hazard rate or the failure rate is constant. Therefore, the failure-time probability density function
of the product or the system can be modeled as an exponential distribution. If the failure rate is l, then
f (t ) = λe − λt (14.20)
where l is the failure rate which is expressed as the number of failures that occur per unit time. It implies
that the expected value of the random variable, that is, the expected time to failure of the item is 1/l.
Clearly, the MTTF of an item is the reciprocal of the failure rate, that is
1 1
MTTF = = (14.21)
λ Failure rate
Figure 14.5 gives a visualization of the MTTF. Referring to Figure 14.5, consider that a new item starts
functioning from time t = 0. The failure-time probability density function starts decreasing as shown.
At time t = 1/l, the item is expected to fail. If at that instant the item is replaced by a new item, then
its failure-time probability density function again starts to decrease from l and so on.
MTBF is used for repairable systems and for the case of non-repairable items the MTTF is used. Thus,
1
MTBF = MTTF =
λ
and l is the failure rate in number of failures per unit time.
f (t )
0 MTBF = 1/l t
T = 1/l T = 2/l
Figure 14.5 A visualization of the MTBF.
Chapter 14.indd 776 3/20/2009 7:26:30 PM

It should be understood that as reliability is a probabilistic value. If the MTTF (MTBF) for a product is
10 years, say, then it means that the reliability is R (t ) = e − λt = e −t / MTTF = e − 0.1t. This means that at the end
of 10 years, the reliability would have become e −1 = 0.366. This implies that at the end of 10 years, 63.6%
of the products released in the market would have come back for servicing. It may not be a good idea to give
a 10-year warranty for such a product. On the other hand, after 1 year, the reliability is e − 0.1 = 0.904 imply-
ing that only 10% of the products originally released in the market would be back for servicing. Statistically,
it may be better to give a one-year warranty for the product that is designed to have a MTTF of 10 years.
14.2 Modeling for Reliability
P rior to the use of reliability analysis techniques, it is essential to model any given system in a manner that
is conducive for quantitative analysis. Therefore, a method is needed to reflect the reliability connectivity
of the many part types having different failure rates that would normally make up a complex system. This is
accomplished by establishing a relationship between equipment reliability and individual item failure rates.
In practice, a system is frequently represented as a network in which the system components are connected
together either in series, parallel, meshed or a combination of these. It is vital that the relationship between the
system and its network model be thoroughly understood before considering the analytical techniques that can
be used to evaluate the reliability of these networks. It must be appreciated that the actual system and the reli-
ability network used to model the system may not necessarily have the same topological structure. This consid-
eration involves the essential point that the analyst must be fully familiar with the requirements of the system
and be able to phrase these requirements in a form which can be quantitatively assessed.
Two fundamental reliability network models that are frequently encountered are the series systems and
the parallel systems. In a series system, from the reliability point of view, the system will fail even if one of
the items of the series system fails. On the other hand, in a parallel system, the system will fail only if all the
items of the parallel system fail.
Series System
Consider a system consisting of n independent components as shown in Figure 14.6. The n components are
connected in series from the reliability point of view. This arrangement implies that all components must
work to ensure system success.
Let R1(t), R2(t), …, Rn(t) be the respective probabilities that the components do not fail upto time t. Let
R(t) be the reliability of the entire system. Let T1 be the random variable that indicates the time to failure of
component 1; T2 be the random variable that indicates the time to failure of component 2 and so on. Let
T be the random variable that indicates the time to failure of the overall series-connected system. Then
P(T > t ) = P(T1 > t )& P(T2 > t )& … & P(Tn > t ) (14.22)
As the operations of the n components are independent, Eq. (14.22) can be re-written as
R(t) = R1(t)R2(t)…Rn(t) (14.23)
1 2 n
R1(t ) R2(t ) Rn(t )
Figure 14.6 Series configuration.
Chapter 14.indd 777 3/20/2009 7:26:30 PM

Thus the reliability of the series-connected system, where R1(t), R2(t), … Rn(t) are the respective individual
component reliabilities, is given by
n
R (t ) = ∏ Ri (t ) (14.24)
i =1
In the case of components with constant failure rate, the reliability of the series system is given by
…+ λn )t
R (t ) = e − λ1t ⋅ e − λ2t ⋅⋅⋅⋅⋅ e − λnt = e −(λ1 +λ2 + (14.25)
where l1, l2, …, ln are the respective individual component failure rates. It is evident from Eq. (14.25) that
if the individual components have failure-time probability densities that are exponential, the resulting series
system also has an exponential failure-time probability distribution with an equivalent failure rate of l = l1 +
l2 + … + ln.
As li > 0 for all i, it is evident that the equivalent failure rate of the series-connected system is higher
than the component having the largest failure rate.
MTBF of the equivalent series system is given by
1 1
MTBF = = (14.26)
λ λ1 + λ2 + … + λn
Problem 14.1
A system design requires 200 identical components in series. If the overall reliability must not be less
than 0.99, what is the minimum reliability of each component?
Solution
Here, 0.99 = R200
Thus, R = 0.991/200 = 0.99995
Parallel System
Consider a system consisting of n independent components connected in parallel, from the reliability
point of view, as shown in Figure 14.7. In this case, the system fails only if all the components fail. Let Ri(t)
be the reliability of the ith component and Fi(t) be the unreliability or the probability that the component
fails in time t of the ith component. Let R(t) and F(t) be the reliability and unreliability of the entire paral-
lel system respectively. Then, the system fails only if (component 1 fails) & (component 2 fails) & … so
on. Thus,
F(t) = F1(t)F2(t) … Fn(t) (14.27)
Equation (14.27) can be re-written as
1 − R (t ) = [1 − R1 (t )][1 − R2 (t )] ⋅ … ⋅ [1 − Rn (t )] (14.28)
Chapter 14.indd 778 3/20/2009 7:26:30 PM

1
R1(t )
2
R2(t )
n
Rn(t )
Figure 14.7 Parallel configuration.
The reliability of the parallel system can be deduced from Eq. (14.28) and is given by
n n
R (t ) = 1 − ∏ [1 − Ri (t )] = 1 − ∏ Fi (t ) (14.29)
i =1 i =1
It is evident from Eq. (14.29) that even if the individual components have failure-time distributions that are
exponential in nature, the failure-time distribution of the equivalent parallel system is not exponential in
nature. Therefore, if one has to evaluate the equivalent failure rate or the MTBF for the parallel system, one
has to use the following relations:
F ′(t ) f (t )
Z (t ) = = (14.30)
R (t ) R (t )
∞
MTBF = E (T ) = ∫ t ⋅ f (t )dt (14.31)
0
One should note that the parallel system is nothing but a redundant system. As all the components are
working, it is called full-on redundant system. Further, as the whole system works successfully (survives)
even with just one component working, the parallel system described till now is called a full-on, single-
survivor parallel redundant system. In later sections, different types of redundant configurations will be
discussed.
In redundant systems, it is common to encounter identical individual components in the parallel system.
Thus the reliabilities of each of the system would be identical and equal to e–lt where l is the failure rate of
each component of the parallel system. In such a case, for n components in parallel, using Eq. (14.29) one
obtains
R (t ) = 1 − (1 − e − λt )n (14.32)
Using the binomial expansion, Eq. (14.32) can be expanded in series form as
R (t ) = nC1e − λt − nC 2e −2λt + … + ( −1)n −1 e − nλt (14.33)
Chapter 14.indd 779 3/20/2009 7:26:30 PM

Using the relationship that the failure-time probability density function f (t) is related to the reliability as
f (t) = –R ′(t), the failure-time probability density function f (t) can be obtained from Eq. (14.33). Thus,
f (t ) = nC1λe − λt − nC 2 (2λ )e −2λt + … + (−1)n −1 (nλ )e − nλt (14.34)
Thus, Eq. (14.34) describes the failure-time probability density function for the parallel redundant system
with identical individual components. To obtain the MTBF for a parallel system with identical components,
one can start from the definition of the mean of the random variable which is
∞
MTBF = E (T ) = μ = ∫ t ⋅ f (t )dt (14.35)
0
∞ ∞ ∞
μ = nC1 ∫ t ⋅ λe − λt dt − nC 2 ∫ t ⋅ 2λe −2λt dt + … + (−1)n −1 ∫ t ⋅ nλe − nλt dt (14.36)
0 0 0
Equation (14.36) reduces to
⎛1⎞ ⎛ 1 ⎞ ⎛ 1 ⎞
μ = nC1 ⎜ ⎟ − nC 2 ⎜ ⎟ + … + (−1)n −1 ⎜ ⎟ (14.37)
⎝λ ⎠ ⎝ 2λ ⎠ ⎝ nλ ⎠
Expanding the combination terms and substituting n = 1, 2, 3, … in Eq. (14.37), the MTBF for a parallel
system with identical individual components is given by
1⎛ 1 1⎞
MTBF = μ = ⎜1 + + + ⎟ (14.38)
λ⎝ 2 n⎠
Problem 14.2
Consider the reliability block diagram shown in Figure 14.8. Evaluate the reliability of the system.
C F
0.7 0.75
A B D H
0.95 0.99 0.7 0.9
E G
0.7 0.75
Solution
The reliability of the system is given by
R(t) = (0.95)(0.99)[1 – (1 – 0.7)3][1 – (1 – 0.75)2](0.9) = 0.772
Chapter 14.indd 780 3/20/2009 7:26:30 PM

Problem 14.3
Consider the full-bridge rectifier–capacitor filter circuit as shown in Figure 14.9. The input is a sinusoidal
waveform given by 100sin(wt). The capacitor has been so designed such that the output voltage ripple is
10 V for full-wave rectification operation and 20 V for half-wave rectification operation. Let qd be the prob-
ability of failure of the diodes and qc be the probability of failure of the capacitor. Draw the reliability
block diagram and find the reliability of the system if the minimum output voltage required is (a) 85 V,
(b) 75 V.
D1 D2
100sin(wt ) C RL Vo
D3 D4
Solution
Let EDi denote the event indicating the occurrence of failure of the ith diode and EC denote the event
indicating the occurrence of failure of the capacitor. Let R(t) and F(t) be the reliability and the unreli-
ability of the system. Then,
Case (a)
Here it is required that the output voltage should be at least 85 V for proper operation. If the
intended function of at least 85 V output is not maintained, then it is considered as system failure.
Thus,
System fails IF (ED1 and/or ED2 and/or ED3 and/or ED4) occurs and/or (EC) occurs
The reliability block diagram is shown in Figure 14.10(a). From the reliability block diagram, the
reliability is given by
R (t ) = (1 − qd )4 (1 − q c )
Chapter 14.indd 781 3/20/2009 7:26:31 PM

D1 D2 D3 D4 C
(1 − qd)4 1 − qc
(a)
2
(1 − qd)
D1 D4
D2 D3
(1 − qd)2 1 − qc
(b)
Figure 14.10 (a) For Case I; (b) for Case II.
Case (b)
Here it is required that the output voltage should be at least 75 V for proper operation. In this case, for
example if D1 open circuits, then half-wave rectification operation occurs through D2 and D3. As the
output voltage ripple is 20 V under this mode of operation, it is within the limits of proper operation.
Therefore, any failures or combination of diode failures that leads to half-wave rectification operation is
also proper operation. Thus,
System works IF E D1 & E D4 occurs and/or E D 2 & E D3 occurs and E C occurs
The reliability block diagram is as shown in Figure 14.10(b). From the reliability block diagram, the reli-
ability is given by
R (t ) = {1 − [1 − (1 − qd )2 ]2 }(1 − q c )
Problem 14.4
Consider the half-wave rectifiers shown in Figure 14.11. All diodes are identical with an open-circuit failure
probability of qo and a short-circuit failure probability of qs. If the intended function of the circuit is in obtain-
ing a half-wave rectified waveform at the output, then evaluate the reliabilities of the three circuits through
reliability block diagrams. If qo = 0.05 and qs = 0.3, then evaluate the reliabilities for the three circuits.
Chapter 14.indd 782 3/20/2009 7:26:31 PM

A B
Vin RL Vo
(a)
D1 D2
A B
Vin D3 D4 RL Vo
(b)
D1 D2
A B
Vin D3 D4 RL Vo
(c)
Solution
If the system has to fail then there should either exist a short or an open between the points A and
B in the case of all three circuits. In all three cases let EDi represent an event where the diode Di
fails.
Case I
Here the system fails if the diode fails either in the short-circuit mode or in the open-circuit mode. Thus,
the reliability is given by
R (t ) = 1 − (q s + qo )
as the failure of the diode in any one of the modes (short or open) are mutually exclusive. Substituting
the values of qs and qo, one obtains
R(t) = 0.65
Chapter 14.indd 783 3/20/2009 7:26:31 PM

(1 − qo)2
D1 D3 D1 D2
D2 D4 D3 D4
(1 − q 2s) (1 − q 2s) (1 − qo)2
(a) (b)
Figure 14.12 Case II: (a) short-circuit failure mode; (b) open-circuit failure mode.
Case II
Here one has to consider the two failure modes between the points A and B, that is, the short-circuit
failure mode and the open-circuit failure mode as shown in Figure 14.12.
Short-circuit failure mode: The system fails by a short between points A and B
IF (ED1 & ED2) occur and/or (ED3 & ED4) occur
This results in the short-circuit failure mode reliability block diagram as shown in Figure 14.12(a). From
the reliability block diagram the unreliability is given by
Fs (t ) = 1 − (1 − q s2 )2
Open-circuit failure mode: The system fails by a open between points A and B
IF (ED1 and/or ED2) occur & (ED3 and/or ED4) occur
The open-circuit reliability block diagram for this failure mode is indicated in Figure 14.12(b). The unre-
liability is given by
Fo (t ) = [1 − (1 − qo )2 ]2
The short-circuit failure mode and the open-circuit failure mode are mutually exclusive; thus the unreli-
ability and reliability of the total system are given as
F (t ) = Fs (t ) + Fo (t ) = 1 − (1 − q s2 )2 + [1 − (1 − qo )2 ]2
R (t ) = (1 − q s2 )2 − [1 − (1 − qo )2 ]2
Substituting the values of qo and qs in the above reliability equation, one obtains
R(t) = 0.8186
Chapter 14.indd 784 3/20/2009 7:26:31 PM

Case III
Here again, one has to consider the two failure modes between the points A and B, that is, the short-
circuit failure mode and the open-circuit failure mode.
(1 − qs)2
D1 D3 D1 D2
D2 D4 D3 D4
(1 − qs)2 (1 − q 2o) (1 − q 2o)
(a) (b)
Figure 14.13 Case III: (a) short-circuit failure mode; (b) open-circuit failure mode.
Short-circuit failure mode: The system fails by a short between points A and B
IF (ED1 and/or ED3) occur & (ED2 and/or ED4) occur
From the short-circuit failure mode reliability block diagram [Figure 14.13(a)] the unreliability is given by
Fs (t ) = [1 − (1 − q s )2 ]2
Open-circuit failure mode: The system fails by a open between points A and B
IF (ED1 & ED3) occur and/or (ED2 & ED4) occur
From the open-circuit failure mode reliability block diagram [Figure 14.13(b)] the unreliability is given by
Fo (t ) = 1 − (1 − qo2 )2
The short-circuit failure mode and the open-circuit failure mode are mutually exclusive, thus the unreli-
ability and reliability of the total system are given as
F (t ) = Fo (t ) + Fs (t ) = 1 − (1 − qo2 )2 + [1 − (1 − q s )2 ]2
R (t ) = (1 − qo2 )2 − [1 − (1 − q s )2 ]2
Substituting the values of qo and qs in the above reliability equation, one obtains
R(t) = 0.735
Referring to Problem 14.4, it is seen that both circuits of Figure14.11(b) and (c) are more reliable than the
circuit of Figure 14.11(a). Between Case II and Case III, it is seen that Case II is more immune to short-
circuit failures, that is, this configuration is preferred where the short-circuit failure probability of the
diodes is higher than the open-circuit failure probability. Case III is more immune to open-circuit failures,
that is, this configuration is preferred where the open-circuit failure probability of the diodes are higher
than the short-circuit failure probability.
Chapter 14.indd 785 3/20/2009 7:26:31 PM

Mesh System
Many systems do not have the simple structure as expressed by systems exhibiting series and parallel struc-
tures. They may have complex operational logic. Additional modeling and evaluation techniques are neces-
sary to determine the reliability of such systems. A typical system not having a series/parallel structure is the
bridge-type network as shown in Figure 14.14. A visual inspection of the network shown in Figure 14.14
indicates that the components are not connected in a simple series/parallel arrangement. There are a number
of techniques available for solving this type of network such as the conditional probability approach, cut and
tie set analysis, tree diagrams, logic diagrams and connection matrix techniques. It should be noted that
most of the methods are very similar in concept. The essential difference between them lies in the formal
presentation or logic of the method and not the underlying concept.
Consider the system shown in Figure 14.14 in which success requires that at least one path exists from
point X to point Y. Here component E is a bi-directional component. If each component has a failure prob-
ability of q, then it is required to evolve a method to evaluate the reliability of the system.
Let EA, EB, EC, ED, EE be the events that represent the failures of components A, B, C, D and E, respec-
tively. One can assume that the failures of each component are independent of the failures of the other
components.
On visual inspection, it is evident that there are four possible paths from X to Y. They are AC, BD, AED
and BEC. Thus the reliability block diagram for the system is as shown in Figure 14.15.
This is now a parallel system where the system works if any one of the paths exist. The probability that
the system works or the reliability is given by
R (t ) = P( E A & E C ) and/or P( E B & E D ) and/or P( E A & E D & E E ) and/or P( E B & E C & E E )
Using the third axiom of probability and substituting “q” as the failure probability for each component, one
obtains
R(t) = 2p2 + 2p3 – 5p4 + 2p5
where p = 1 – q. However, the above method is not strictly a parallel system as discussed in section “Parallel
System”. In a true parallel configuration, the failure of each of the path is independent of the failures of the
other paths. In the case of the parallel configuration of Figure 14.15, the parallel paths are not independent
as the failure of one component influences the failure or success of other paths. Therefore, the results
obtained by this method can at times be dubious.
A C
E
X Y
B D
Figure 14.14 Mesh configuration.
Chapter 14.indd 786 3/20/2009 7:26:31 PM

A C
B D
X Y
A E D
B E C
Figure 14.15 Multiple path equivalent of the mesh configuration.
Conditional Probability Approach

In this approach, complex systems are reduced sequentially into sub-system structures that are connected in
series/parallel and then to recombine these sub-systems using the conditional probability method. The basic
concept of this approach is as follows:
P(system success or failure) = P(system success or failure if component C is good) × P(C is good) +
P(system success or failure if component C is bad) × P(C is bad)
To apply the conditional probability to the problem of Figure 14.14, it is first necessary to choose the ref-
erence component which will be considered “good” and considered “bad”. Any of the components A to
E may be chosen but some choices will result in smaller sub-branches. Here component E is chosen as the
reference component. When E is good, the system reduces to a simple parallel–series system as shown in
Figure 14.15(b). When E is bad, the system reduces to a simple series–parallel system as shown in Figure
14.16(c). These two sub-systems are mutually exclusive because E E and E E are mutually exclusive and
further they are collectively exhaustive. Therefore, the system reliability is given by
Rs(t) = P(system works if E works) × P(E works) + P(system works if E fails) × P(E fails)
Rs (t ) = P( E S / E E ) ⋅ P( E E ) + P( E S / E E ) ⋅ P( E E ) (14.39)
where ES is the event that the system fails. Referring to Figure 14.16,
P( E S / E E ) = (1 − FA FB )(1 − FC FD ) (14.40)
P( E S / E E ) = 1 − [(1 − RA RC )(1 − RB RD )] (14.41)

Substituting Eqs. (14.40) and (14.41) in Eq. (14.39), one obtains
Rs (t ) = (1 − FA FB )(1 − FC FD )RE + {1 − [(1 − RA RC )(1 − RB RD )]}FE (14.42)
Equation (14.42) can be simplified to obtain
R(t) = 2p2 + 2p3 – 5p4 + 2p5
Chapter 14.indd 787 3/20/2009 7:26:32 PM

A C
E
X Y
B D
(a)
E is good E is bad
A C A C
X Y X Y
B D B D
(b) (c)
Figure 14.16 Conditional probability approach.
where p = 1 – q. One should note that in the above example, only one level of sub-division was performed.
But in very complex systems, one or more sub-systems may need further sub-division before a series/parallel
structure is obtained. This is only an extension of the technique that was discussed above since each time a
sub-division is made, the two sub-divisions must be recombined using the conditional probability approach
starting at the lowest hierarchical level.
14.3 Redundant Systems
T o cope with technological developments, systems have been compelled to expand both in size and com-
plexity at a rather rapid rate. Of equivalent importance to the need for this growth has been the simul-
taneous need for greater system reliability. As industrial systems become more complex, the philosophy of
simply increasing the reliability by making parts more reliable became unrealistic due to technological limits.
The only answer to such situations has been the design of redundant systems, sometimes termed as fault
tolerant systems.
Redundancy exists when one or more components of a system fail and the system can continue to
perform its intended function satisfactorily. The system may adapt to component failures by performing
the required function with the remaining components or it may switch-in an available spare component to
take the place of the failed one. The components used to take the place of the failed ones might be func-
tionally redundant, that is, not identical in form to the failed unit but capable of compensating for the
failed function; for example, one processor taking over the role of a different type of processor, a VHF
transceiver taking over the function of a UHF transceiver, etc., or it could be an exact duplicate of the
failed component.
There are many different redundancy strategies. Some work in conjunction with maintenance and/or
corrective action performed periodically. Some are applied when no maintenance or corrective action
Chapter 14.indd 788 3/20/2009 7:26:32 PM

can be performed. Each has its advantages (like reliability gain) and disadvantages (like system weight, cost,
volume, number of duplicative elements, etc.). In the following discussion on redundant systems, systems
without maintenance are considered.
Full-On Redundancy: Multiple-Survivor Sub-System

This type of redundancy is a more general case of the simple parallel system discussed earlier. The system
consists of n fully energized parallel components but requires a minimum of m working components for the
system to work. The m-survivor system is depicted in Figure 14.17. For this m-survivor system,
1. Rc(t) is the probability of a component of the system functioning for a period of time t.
2. n is the number of components in the parallel system.
3. Functioning of each of the n components is independent.
4. There are only two possible states of the components either working (success) or failed (failure) state.
The minimum number of components required to work for the system to function satisfactorily is given by
m. It is evident from points (1) to (4) that the choice of m working components out of n components is a
Bernoulli process. Thus, binomial distribution is applicable to the choice of m working components. One
should note that at the start, that is, at t = 0, all n units are assumed to be functional.
The reliability R(t) of the system is the probability that the random variable T takes on values greater
than t. This happens only if at least m components of the n components are working. Thus,
R(t) = P(T > t) = Probability that at least m units must function
Let X be a random variable denoting the number of functional units. Now it is of interest to find the proba-
bility that the random variable X takes on values greater than or equal to m for satisfactory system
functioning.
R (t ) = P(T > t ) = P( X ≥ m ) (14.43)
1
R1(t ) = Rc(t )
2
R2(t ) = Rc(t )
n
Rn(t ) = Rc(t )
Figure 14.17 The m-survivor full-on redundant system.
Chapter 14.indd 789 3/20/2009 7:26:32 PM

From the Binomial distribution,

n
P( X ≥ m ) = ∑ nC k [ Rc ]k [1 − Rc ]n−k (14.44)
k =m
Alternatively consider Y to be a random variable denoting the number of non-functional units. Then for the
system to be functioning, there should be less than or equal to n – m failed units. Thus,
P( X ≥ m ) = P(Y ≤ n − m ) (14.45)
Again from the Binomial distribution,
n −m
P( X ≥ m ) = P(Y ≤ n − m ) = ∑ nC k [1 − Rc ]k [ Rc ]n−k (14.46)
k =0
where Rc = e − λ t is the reliability of each of the components of the redundant system.
MTTF or MTBF for Full-On m-Survivor Redundancy

From the basic definition of the mean or expectation of the random variable, the MTBF is given by,
∞
MTBF = ∫ t ⋅ f (t )dt (14.47)
0
It is known that
f (t) = F ′(t) = –R ′(t) (14.48)
∞
MTBF = − ∫ t ⋅ R ′(t )dt (14.49)
0
Integrating by parts and making use of the fact that R(∞) = 0 (because the probability that the system will
work for infinite time is zero), one obtains
∞
MTBF = ∫ R (t )dt (14.50)
0
One should note that Eq. (14.50) is applicable for any distribution. For the m-survivor case where all
the components have exponential distribution with identical failure rates, the MTBF is given by
1⎛ 1 1 1⎞
MTBF = ⎜ + + + ⎟ (14.51)
λ ⎝ m m +1 n⎠
One should note at this point that for m = 1, the system becomes a full-on, single-survivor redundant
system which is the simple parallel network discussed earlier.
Problem 14.5
An array of signal processors in a parallel computing system performs a specific function. A total of 100 processors
make up the array. The failure rate of each processor is 0.0005 failures/h. The system can perform satisfactorily
as long as at least 90 processors are operating.
(a) What is the probability that the system will operate satisfactorily over a 100-h period, given that all
components were operable at the start?
(b) What is the MTTF of the system?
(c) What would the above reliability characteristics be if a minimum of 95 processors must operate?
Chapter 14.indd 790 3/20/2009 7:26:32 PM

Solution
Rc = e − λt = e −0.0005×100 = 0.95
Here n = 100, m = 90, 1 – Rc = 0.05. For n > 20 and p < 0.05, one can also use the Poisson approxi-
mation to the Binomial distribution. Thus,
10 10
α k e −α
R (t ) = ∑ 100C k (0.05)k (0.95)100 −k = ∑
k =0 k =0 k !
where a = np = (100)(0.05) = 5.
10
5k e −5
(a) R (t ) = ∑ = 0.0067 + 0.0337 + 0.0842 + 0.1403 + 0.1754 + 0.17546 + 0.1462 + 0.1044
k =0 k !
+ 0.06527 + 0.03626 + 0.01813
= 0.986
1⎛ 1 1 1 1 ⎞
(b) MTTF = ⎜ + + + + ⎟ = 232 h
λ ⎝ 90 91 92 100 ⎠
5k e −5
5
(c) R (t ) = ∑ = 0.6157
k =0 k !
MTTF = 123.11 h
Switching in Redundant Systems

In redundant systems, it is necessary in many cases to isolate the failed components. This function of detec-
tion of failed components and isolation of these components from the network is performed by a switching
network. In the analysis of the full-on, multiple-survivor redundant systems, the switching was considered to
be perfect, that is, reliability of unity. But the switching network itself is an electronic circuit which has a
reliability of less than unity. It is therefore of interest to analyze the reliability of redundant systems consider-
ing the effects of imperfect switching, that is, switch reliability that is less than unity.
In modeling the switching network, the question that arises is whether the switching network be consi-
dered as a component in series with the redundant system or as a component in parallel. From the definition
of the series system it is evident that if any component fails, then the system fails. However, in this case, the
switching network comes into play only at the time when one of the component fails, at which time the
fault is detected (sensed) and isolated by the switching network.
It is clear that the switching network performs two distinct functions: (a) sensing or detecting the failed
component; (b) isolating the failed component. Consider a situation where the sensing mechanism of the
switching network has failed. The system as a whole will continue to operate successfully till one of the
redundant units fail. On failure of a redundant unit, the switching network will be unable to detect the failed
unit and therefore will not be able to isolate the failed unit. As a consequence, the system will cease to func-
tion satisfactorily. One should note that even though the sensing mechanism of the switching network had
failed, the system did not immediately stop functioning. The system continued to operate satisfactorily till
Chapter 14.indd 791 3/20/2009 7:26:32 PM

the occurrence of the failure of one of the redundant units. Thus, one cannot say that the switching
network is in series with the redundant system from the reliability point of view.
On the other hand, if one considers a situation wherein the switching network has failed in a manner
that does not allow signal or power to flow through its switch, then the system would immediately fail. In
such a case, the switching network can be considered as a component in series with the redundant system.
Thus, the switching network is governed by two modes of failure that leads to two failure probabilities. They
are as follows:
1. The dormant mode failure of the switching network wherein the sensing mechanism fails and the system
continues to work till the failure of one of the components of the parallel redundant system. This failure
probability of the switching network, denoted by qsd and psd = 1 – qsd, is the probability that the switch-
ing network does not fail in the dormant mode, that is, probability that the switching network performs
sensing and isolation successfully.
2. The immediate failure mode of the switching network wherein this type of failure in the switching
network leads to instantaneous failure of the system. This failure probability of the switching network,
denoted by qsi and psi = 1 – qsi, is the probability that the switching network does not cause immediate
breakdown of the system.
Thus, the redundant system with imperfect switching can be represented as indicated in Figure 14.18. The
switching network is represented as two parts: One which causes immediate failure of the system (shown in
series with the redundant system) and other that causes the dormant mode failure of the system.
The switching network behavior can be illustrated with the following example. Consider the circuit
shown in Figure 14.19. It is required that two power supplies PS1 and PS2 be paralleled to deliver some
specific power to the load. The power supplies PS1 and PS2 are individually capable of handling the full
load power. But on paralleling, the electrical stresses on each of the power supplies are reduced, thereby
enhancing the reliability. The system is said to function satisfactorily when at least one power supply is
working satisfactorily. When any one of the power supply fails, the output voltage is required to be sensed
and the failed power supply should be isolated from the circuit. These functions of sensing the fault and
isolating the failed unit are done by a switching network as shown. The switching network consists of two
1
Rc(t )
2
Rc(t ) Psd Psi
n
Rc(t )
Figure 14.18 Full-on, multiple-survivor redundant system with imperfect switching.
Chapter 14.indd 792 3/20/2009 7:26:32 PM

PS1 RL1
Mains
To load
RL2
PS2
Relay
drive
Sense
circuit
Switching network
Figure 14.19 Power supplies with relay network.
relays RL1 and RL2 through which power supplies PS1 and PS2, respectively, are connected to the load.
The relays are controlled by a relay drive circuit. The control of the relay is based on the sense circuitry
which senses the output voltage of the power supplies for fault condition.
The sensing circuitry of the switching network can fail. Then, in such a case depending on the type of
relays RL1 and RL2 (normally closed or normally open), the system could continue to operate satisfactorily
or not. The relay drive circuit may fail. Here again, depending on the relay types (i.e., normally closed or
normally open), the system will continue to work or not.
The relay contacts may open circuit. In this case the system will fail immediately if both the relay con-
tacts have failed. If only one relay contact has failed, then the system continues to work till the power supply
corresponding to the working relay contact fails.
Thus it is evident that one type of failure of the switching network is the dormant mode failure and the
other type of failure of the switching network causes immediate failure of the system. It is of interest to find
the reliability of the full-on, multiple-survivor redundant system considering imperfect switching. Consider a
three-component full-on, single-survivor redundant system as shown in Figure 14.20. To find the reliability
of this system, one can first evaluate the reliability of the redundant system together with the dormant failure
mechanism component of the switching network and then combine the resulting system with the immediate
failure mechanism component of the switching network in a simple series network configuration.
Considering the system with only dormant failure mechanism part, the system works successfully
IF (no component fails) or
(1 component fails & the failure is successfully sensed and isolated) or
(2 components fail & both failures are successfully sensed and isolated)
Chapter 14.indd 793 3/20/2009 7:26:32 PM

1
Rc(t )
2
Rc(t ) Psd Psi
3
Rc(t )
Figure 14.20 Three-component, full-on, single survivor with imperfect sensing and switching.
Let X be a random variable that denotes the number of failures of components, then
Rs(t) = P(X = 0) + P(X = 1) psd + P(X = 2)psd psd (14.52)
Using Eq. (14.46), Eq. (14.52) can be re-written as
2
Rs (t ) = ∑ 3C k (1 − Rc )k ( Rc )3−k ( psd )k (14.53)
k =0
For full-on, m-survivor, n-component redundant system with imperfect switching, the reliability Rs(t), as
given by Eq. (14.53), can be generalized as
n −m
Rs (t ) = ∑ nC k (1 − Rc )k ( Rc )n−k ( psd )k (14.54)
k =0
Considering the reliability of the immediate failure mechanism component of the switching network,
psi, the reliability R(t) is given by
⎡n − m ⎤
R (t ) = ⎢ ∑ nC k (1 − Rc )k ( Rc )n −k ( psd )k ⎥ psii (14.55)
⎣ k =0 ⎦
Equation (14.54) defines the reliability of the full-on redundant system with dormant mode failure mecha-
nism. One should note that if there are k units that have failed (k < n – m), then there must be k successful
fault detection and isolation. Here psd provides practical boundaries for the reliability improvement due to
the application of redundancy. A reasonable state-of-the-art value of psd = 0.95 is assumed to enable some
comparisons to be made.
Referring to Figure 14.21, the reliabilities for full-on, single-survivor systems with and without perfect
switching network are compared for various values of n. The light lines indicate the reliabilities for the case
of perfect switching and the solid lines indicate the reliabilities for the case of imperfect switching. As can be
seen from Figure 14.21, even for a psd = 0.95, a significant difference results between the reliability gained
with perfect switching and that gained with imperfect switching. Further, one can note that as the number
of redundant components increases, the difference is more pronounced.
Chapter 14.indd 794 3/20/2009 7:26:32 PM

1
R(t )
0.95
n =4
0.9
0.85 n =3
0.8
n =4
n =3
0.75
0.7 n=2
n =2
0.65
0.6 Psd = 1
0.55 Psd = 0.95
0.5
0 0.2 0.4 0.6 0.8 1
t
Figure 14.21 Full-on, m-survivor comparison with and without perfect sensing and switching
network.
Mean Time to Failure

Considering the redundant system with only dormant mode failures of the switching network, the MTBF
is obtained by using
∞
MTBF = ∫ R (t )dt
0
Substituting Eq. (14.54), one obtains
1 ⎛ ( psd )n −m ( psd )n −m −11⎞
MTBF = ⎜ + + ⎟⎟+ (14.56)
λ ⎜⎝ m m +1 n⎠
One should note at this point that for psd = 1 (i.e., perfect switching), the MTBF of the system reduces to
that of a full-on, multiple-survivor redundant system with perfect switching.
Problem 14.6
A system comprises two components in full-on, single-survivor redundant configuration. The failure rate of each
component is 0.001 failures/hour. For a period of operation of 2 h, what is the
(a) reliability of a single component?
(b) reliability of the system with perfect switching?
(c) reliability of the system with psd = 0.2 and psd = 0.5?
Chapter 14.indd 795 3/20/2009 7:26:33 PM

Solution
(a) Let Rc(t) be the reliability of each of the component. Then,
Rc (t ) = e − 0.001× 2 = 0.998
(b) Let R(t) be the reliability of the system. Then,

R(t) = 1 – (1 – Rc)2 = 0.99999
(c) With psd = 0.2,

1
R (t ) = ∑ 2C k (1 − Rc )k ( Rc )2 −k (0.2)k = 0.984
k =0
With psd = 0.5,

1
R (t ) = ∑ 2C k (1 − Rc )k ( Rc )2 −k (0.5)k = 0.998
k =0
Therefore, it is evident that with psd > 0.5 the redundant system is more reliable than a single component.
As a consequence it would make sense to use redundancy in this case only if the switching network has a
reliability of greater than 0.5.
Problem 14.7
A system comprises two identical units which are connected in full-on, single-survivor redundant configuration.
If the failure rate of one of the units is 0.001 failures/h, what should be the minimum reliability requirement
for the switching network, with regard to dormant failures, for a period of 5 h of continuous operation, so that
the redundant system is meaningful? (Assume that the switching network does not give rise to immediate failure
of the system).
Solution
Let Rc(t) be the reliability of each unit and psd the reliability of the switching network with regard to the
dormant failures. For the redundant system to be meaningful, the reliability of the redundant system
should be more than the reliability of the single system.
1
∑ 2C k (1 − Rc )k ( Rc )2−k ( psd )k > Rc
k =0
where Rc = e–0.001 × 5 = 0.995.

Solving the above equation for psd, one obtains the condition that psd > 0.5 should be satisfied if the
redundant system is to have a reliability greater than the reliability of the individual components.
Chapter 14.indd 796 3/20/2009 7:26:33 PM

Standby Redundancy
Another frequently used redundancy configuration is the standby redundancy which is shown in
Figure 14.22. Let there be n components in a system. Out of these n components, m components are
required to function for satisfactory operation of the system. The remaining n – m components are kept in
standby. These n – m components are not fully energized as in the case of the full-on redundancy. Thus, in
the standby mode of operation, the failure rate of the standby units can be assumed to be zero. One should
note that Figure 14.22 is not a reliability block diagram but the functional topology of the system. But
from the reliability point of view, the m-block unit in operation is a series network which is indicated in
Figure 14.23.
When one of the m operating units fails, the failure of the unit is detected and the failed component is
immediately replaced by one of the standby units. The system continues to perform its function until the
store of standby units is exhausted. Then on the immediate next failure of one of the components, the
system would fail. Thus the system can withstand upto n – m failures and on the occurrence of n – m + 1
failures, the system fails.
In the analysis of the standby redundancy with multiple failure, the following points must be noted.
1. The occurrence of failure among the m operational units is a Poisson process, that is, at any given
instant no two units will fail simultaneously.
2. There are m units functioning at any given time. If any one of the m functioning units fails, then the
block of m-unit sub-system experiences a failure. Therefore, the m-unit sub-system forms a series net-
work with an equivalent failure rate of ml, where l is the failure rate of the individual units.
If one of the m units fails, then the failed unit is replaced by a standby unit thus resurrecting the m-unit
block immediately. The reliability of such an m-unit sub-system is dependent on the number of resurrec-
tions possible. As there are n – m standby units, n – m resurrections of the m-unit block are possible. Thus
the system can withstand up to n – m failures. For any failure more than n – m, there will be less than
m functioning units and the system fails.
1
Rc(t )
m
Rc(t )
n
Rc(t )
Figure 14.22 The m-survivor standby redundancy topology.
Chapter 14.indd 797 3/20/2009 7:26:33 PM

m-unit operation block
1 2 m
m ++1 1
m
n−m
standby units
n
Figure 14.23 The m-survivor standby redundancy reliability block diagram.
The expected number of failures in time t is given by mlt. Let X be the random variable that indicates
the number of failures in a given time t. Then,
n −m
(mλt )k e − mλt
R (t ) = P( X ≤ n − m ) = ∑ k!
(14.57)
k =0
MTBF for Standby Redundancy Multiple-Survivor Configuration

The MTBF can be found by using
∞
MTBF = ∫ R (t )dt
0
Substituting Eq. (14.57), one obtains
n − m +1
MTBF = (14.58)
mλ
Alternatively the MTBF can also be obtained intuitively in the following manner.
1
Mean time to first failure =
mλ
1 1 2
Mean time to second failure = + =
mλ mλ mλ
1 1 1 3
Mean time to third failure = + + =
mλ mλ mλ mλ
Chapter 14.indd 798 3/20/2009 7:26:33 PM

n − m +1
Mean time to n − m + 1 failure =
mλ
This last equation is identical to Eq. (14.58).
Standby Redundancy with Imperfect Switching

Let psd be the probability that the switching network successfully senses a faulty unit and switches in a
standby unit. By using a similar argument used for the case of full-on, multiple-survivor redundancy with
imperfect switching, the reliability for the standby, multiple-survivor redundancy is given by
n −m ⎡
(mλt )k e − mλt ⎤
R (t ) = ∑⎢ k!
( psd )k ⎥ (14.59)
k =0 ⎣ ⎦
The MTBF for standby redundancy taking into account imperfect switching is given by
1
MTBF = (1 + psd + psd 2 + ... + psd n −m ) (14.60)
λ
It can be noted that for perfect switching when psd = 1, the MTBF reduces to Eq. (14.58).
14.4 Redundant Systems with Non-Identical Components
T he preceding section has assumed that all the components of the standby system are identical. In a prac-
tical system, this is not necessarily true in all cases. The failure rates of the operating and standby compo-
nents can be different owing to the nature of the components used. A common example is when a generator
is used as the normal operating component of a DC supply and a battery is used as the standby component.
In this case the failure rates could be very different. Further, in many cases, the failure rates of the standby
components will not be zero. For example, in the case of an off-line UPS system, the battery–inverter system
which is in standby till the mains power fails, will be generally in hot standby, that is, the battery–inverter
system is energized but it operates under no-load. Once the mains power fails, the battery–inverter system
will take on the full load. Thus, in such cases, the failure rate during standby is not zero but a finite value,
which would of course be much lower than the failure rate when delivering full load power.
The approach used to model these systems is a direct and intuitive one. This approach is perfectly gen-
eral and involves very little effort and computation time. In this method all events leading to system success
are divided into mutually exclusive events. The expression for system reliability can then be derived by
adding the reliability associated with each of these individual mutually exclusive events. This method is
explained with the help of a few examples.
EXAMPLE 14.1 Consider the case of two non-identical components 1 and 2 forming the standby
system in which component 2 is the standby component and cannot fail in the
standby mode. It is of interest to find the reliability of the system till time t.
Let l1 be the failure rate of component 1 in the energized mode. Let l2 be
the failure rate of component 2 in the energized mode. The failure rate of compo-
nent 2 in the standby mode is zero.
The system operates successfully
IF (1 is good till t)
or (1 fails at t1 < t) & (2 is good for t – t1 period)
Chapter 14.indd 799 3/20/2009 7:26:33 PM

Let
R1(t) = P(1 is good till t)
R2(t) = P[(1 fails at t1 < t) & (2 is good for t – t1 period)]
Let R(t) be the reliability of the system. As R1(t) and R2(t) are mutually exclusive,
R(t) = R1(t) + R2(t)

− λ1t
R1(t ) = e
f (t ) f (t )
l1e−l 1t
l 1e−l 1t
l 2e−l 2(t−t1)
dt1
0 t 0 t1 t
(a) (b)
Figure 14.24 Graphical representation of reliability contributions:

(a) Contribution of R1(t); (b) contribution of R2(t).
Referring to Figure 14.24, it is evident that the probability that component 2

is good for the period of time (t – t 1) is the area shown shaded below the
failure-time probability density function of component 2 after time t. The
probability that component 1 fails at time t1 is given by the area under the
infinitesimal interval dt1. However, one should note that the failure of compo-
nent 1 at time t1 can occur at any time between 0 and t. Therefore, the area
below the failure-time probability density function of component 2 for every
value of t1 between 0 and t should be integrated. Thus,
t
R2 (t ) = ∫ λ1e − λ1t ⋅ e − λ2 (t −t1 ) ⋅ dt1
t1 = 0
λ1
= [e − λ1t − e − λ2t ]
λ2 − λ1
λ1
R (t ) = e − λ1t + [e − λ1t − e − λ2t ]
λ2 − λ1
Chapter 14.indd 800 3/20/2009 7:26:33 PM

Problem 14.8
Consider the situation as discussed in Example 14.1, but in this case the failure rate of the component 2 when
in standby is l2s. Using similar logic, what is the reliability of the system?
Solution
IF (1 is good till t)
or (1 fails at t1 < t) & (2 is good in standby till t1) & (2 is good in energized mode for t – t1 period)
R1(t) = P(1 is good till t)
R2(t) = P[(1 fails at t1 < t) & (2 is good in standby till t1) & (2 is good in energized mode for t – t1
period)]
If R(t) is the reliability of the system, then
R (t ) = R1(t ) + R2 (t )
R1(t ) = e − λ1t
t
R2 (t ) = ∫ λ1e − λ1t ⋅ e − λ2 st e − λ2 (t −t1 ) ⋅ dt1
t1 = 0
λ1
= [e −(λ1 +λ2 s )t − e − λ2t ]
λ2 − λ1 − λ2 s
λ1
R (t ) = e − λ1t + [e −(λ1 +λ2 s )t − e − λ2t ]
λ2 − λ1 − λ2 s
Problem 14.9
Consider the system shown in Figure 14.25. Here the components 1 and 2 operate as a full-on, single-survivor
redundant system and component 3 is used when both 1 and 2 have failed. For this system the following data
is available:
(a) failure rate of component 1 when energized = l1e
(b) failure rate of component 2 when energized = l2e
(c) failure rate of component 3 when energized = l3e
(d) failure rate of component 3 when in standby = l3s
(e) failure rate of sensing device = ls
(f) failure rate of changeover device in position a = lea
(g ) failure rate of changeover device in position b = lcb
(h) probability of successful changeover = ps
What is the reliability of the system?
Chapter 14.indd 801 3/20/2009 7:26:33 PM

Solution
As components 1 and 2 are operating in full-on, single-survivor redundant configuration, component 3
is switched into operation only after both components 1 and 2 have failed. To evaluate the reliability of
the system, it is required to find the probability that the random variable T takes on values greater than t,
that is, the system should not fail in the interval (0, t).
1
Figure 14.25 System schematic for Problem 14.9.

IF (1 is good till t) & (2 is good till t) & (changeover device is good in position a till t)
or (1 is good till t) & (2 has failed before t) & (changeover device is good in position a till t)
or (1 has failed before t) & (2 is good till t) & (changeover device is good in position a till t)
or (1 fails at t1 < t) & (2 fails at t2 > t1) & (3 is good in standby till t2) & (3 is good for t – t2 in energized
mode) & (sensing is good till t2) & (changeover device is good in position a till t2) & (changeover
device has successfully changed over) & (changeover device is good in position b for t – t2)
or (2 fails at t1 < t) & (1 fails at t2 > t1) & (3 is good in standby till t2) & (3 is good for t – t2 in energized
mode) & (sensing is good till t2) & (changeover device is good in position a till t2) & (changeover
device has successfully changed over) & (changeover device is good in position b for t – t2)
In the above description of events t1 < t2 < t. Let
R1(t ) = e − λ1et ⋅ e − λ2et ⋅ e − λcat

R2 (t ) = e − λ1et ⋅ (1 − e − λ2et ) ⋅ e − λcat
R3 (t ) = (1 − e − λ1et ) ⋅ e − λ2et ⋅ e − λcat
t ⎡ t2 ⎤
−λ t −λ t −λ t − λ (t −t ) − λ t −λ t −λ t
R4 (t ) = ∫ ⎢⎢ ∫ λ1ee 1e 1 dt1 ⎥⎥ ⋅ λ2ee 2e 2 ⋅ e 3s 2 ⋅ e 3e 2 ⋅ e s 2 ⋅ e ca 2 ⋅ ps ⋅ e cb 2 dt 2
t 2 = 0 ⎣ t1 = 0 ⎦
t ⎡ t2 ⎤
R5 (t ) = ∫ ⎢ ∫ λ2 e e − λ2 et1 dt1 ⎥ ⋅ λ1e e − λ1et 2 ⋅ e − λ3 st 2 ⋅ e − λ3e (t −t 2 ) ⋅ e − λst 2 ⋅ e − λcat 2 ⋅ ps ⋅ e − λcbt 2 dt 2
t2 = 0 ⎢⎣ t1 = 0 ⎥⎦
Chapter 14.indd 802 3/20/2009 7:26:33 PM

If R(t) is the reliability of the system, then

R(t) = R1(t) + R2(t) + R3(t) + R4(t) + R5(t)
Problem 14.10
Consider the parallel connection of the two MOSFETs as shown in Figure 14.26. The required load current is
50 A. What is the reliability of the system if each MOSFET is rated
(a) for 30A?
(b) for 60A?
50 A
Q1 Q2
Figure 14.26 Paralleling of MOSFETs.
Solution
When both the MOSFETs are operating, each of the MOSFET carries half the current (i.e., 25 A). Let l25
and l50 be the failure rates of Q1 and Q2 when carrying 25 A and 50 A currents, respectively. Let the portions
of the failure rates associated with open-circuit failures of the transistors be lo25 and lo50, respectively.
Case I (MOSFETs rated for 30 A each): Here both Q1 and Q2 should operate successfully for the system
to operate. Therefore, this is the series network from the reliability point of view. Thus,
R (t ) = e − λ25t ⋅ e − λ25t = e −2λ25t
Case II (MOSFETs rated for 60 A each): The system operates successfully
IF (Q1 is good till time t − l25) & (Q2 is good till time t − l25)
or (Q1 fails at t1 < t due to open-circuit failure − lo25) & (Q2 is good till t1 carrying 25 A − l25) &
(Q2 is good for t − t1 carrying 50 A − l50)
or (Q2 fails at t1 < t due to open-circuit failure − lo25) & (Q1 is good till t1 carrying 25 A − l25) &
(Q1 is good for t − t1 carrying 50 A − l50)
Chapter 14.indd 803 3/20/2009 7:26:34 PM

Let
R1(t) = P[(Q1 is good till time t – l25) & (Q2 is good till time t – l25)]
R2(t) = P[(Q1 fails at t1 < t due to open-circuit failure – lo25) & (Q2 is good till t1 carrying 25 A – l25) &
(Q2 is good for t – t1 carrying 50 A – l50)]
R3(t) = P[(Q2 fails at t1 < t due to open-circuit failure – lo25) & (Q1 is good till t1 carrying 25 A – l25) &
(Q1 is good for t – t1 carrying 50 A –l50)]
As the events described above are mutually exclusive,
R (t ) = R1(t ) + R2 (t ) + R3 (t )
R1(t ) = e −2λ25t (using th
he third axiom of probability)
t
R2 (t ) = ∫ (λo 25e − λo 25t1 ) ⋅ e − λ25t1 e − λ50 (t −t1 ) ⋅ dt1
t1 = 0
λo25
= [e −(λ25 +λo 25 )t − e − λ50t ]
λ50 − λ25 − λo 25
Similarly,
λo25
R3 (t ) = [e −(λ25 +λo 25 )t − e − λ50t ]
λ50 − λ25 − λo 25
Thus,
2λo25
R (t ) = e −2λ25t + [e −(λ25 +λo 25 )t − e − λ50t ]
λ50 − λ25 − λo 25
14.5 Design for Reliability

A primary attribute of any equipment is its ability to provide the required service under a specified range
of constraints for a specified period of time, that is, the reliability of the equipment. The reliability of an
equipment might be expected to be inversely proportional to its complexity, but this is not generally the case
for equipments in large-scale production. A piece of equipment must reach a certain standard of reliability
to gain acceptance. Hence with increasing complexity of equipment, the reliability of individual component
must be correspondingly increased. In the field of electronics, new systems and new technologies are intro-
duced before the engineering practices for assuring that their reliability has been fully evolved. In such cases,
the contribution of electronic components to system reliability is more difficult to predict and where high
reliability is required, the designer uses his best judgment in selecting components of assured quality and
operating them under less stringent conditions.
Reliability, which is a measure of quality, is an essential element at each stage of the equipment manu-
facturing procedure, right from the initial specification of the equipment through the design and production
stage, to the final delivery to the user. In designing a piece of equipment for improved reliability, the factors
Chapter 14.indd 804 3/20/2009 7:26:34 PM

to be considered are manifold and interacting. Some of these factors that affect the reliability of the equip-
ment are the following:
1. Design Process: The methodology of designing equipment at the system level has a significant impact
on the reliability of the equipment. In this section, a systematic approach to design of a given equip-
ment will be illustrated using a case study.
2. Inter-connection/Wiring: One of the major factors that affect the reliability of the equipment is the
manner in which the various modules of the system are inter-connected during the design process. One
should give lot of consideration while selecting the type of connector or wires (conductors) for a given
application.
3. Effect of Circuit Design on Reliability: The circuit must be designed taking into consideration the
various component tolerances at all extremes of the operating conditions. The function of every compo-
nent must be related to its capabilities as given in the published datasheet.
4. Component Reliability: The selection of appropriate component types is essential in equipments that
are designed for reliability. As the reliability of components are reduced by operating them at ratings
above those recommended, the components must be chosen such that they are operated well below the
recommended ratings.
5. Environmental Effects on Reliability: Components must be provided with suitable layout, adequate
heat sinking and protection against voltage transients and surges. The physical handling of the equip-
ment and any special requirements of the equipment must be borne in mind, as well as the ambient
conditions under which the equipment is to operate.
The main objective of the design process is to combine the various demands and constraints in an appro-
priate manner to result in the conception of the product. The various demands and constraints that are to
be taken into account while designing a product/equipment will be the input to the design process. This
input is called the design specifications or the target specifications that the final product should meet. The
target specifications reflect the needs of the equipment user. These specifications are generally obtained by
intensive market survey, user survey and survey of contemporary products.
The design process sets in after the target specifications are finalized. The general attitude to design of an
electronic product is rather ad hoc. Once the specifications are obtained, most amateur designers would
immediately start rigging up the relevant electronic circuits, fabricating the pcbs and integrating them to
build the product. Such an approach would lead to solutions being added to rather than integrated in the
design. For example, one may have ignored the target specifications and forgotten to incorporate a moni-
toring circuit or a protection circuit for a specific signal. This discrepancy may be noticed rather late in the
development. As a consequence, the designer may add some extra circuits to correct the oversight. This
would result in a poorly designed product from the reliability and the cost point of view. One should
approach the design of the products in a more systematic manner wherein all the demands and constraints
of the specifications are taken into account and inherently integrated within the design.
Further, one should note that there is no unique solution to achieve an intended function in electronic
products. As there are many solutions, the question of which is the best solution arises. From the point of
view of the product, the circuit configuration or module configuration that achieves the highest overall reli-
ability to cost ratio would be the best choice. In view of the non-uniqueness in the approaches to design a
specific system, this section will illustrate with a case study a sample design process keeping a measure of
algorithmic style to the design process such that it can be extended to other applications.
Chapter 14.indd 805 3/20/2009 7:26:34 PM

Belt-Conveyor System: An Example Design Process

A belt-conveyor system is used to transfer material from one point to another. An induction motor is used
to operate the belt-conveyor system as shown in Figure 14.27. It is required that the material should be
transferred from one point to another at a user-defined fixed rate (kg/s). A product/equipment has to be
developed that controls the speed of the induction motor in such a manner as to keep the feed rate constant
at the user-defined value. The belt-conveyor system is in an environment that is not conducive for human
presence. Therefore, the setting of the feed rate by the user has to be done at the remote location. A load cell
is used as a transducer to measure the weight. A speed transducer that is mounted on the induction motor
shaft is used to measure the conveyor speed. The following specifications are obtained in the context of this
system.
500 m
Control unit
(user interaction
unit)
Belt-conveyor system
Gear
3 phase box
AC
Drive unit IM
Figure 14.27 Schematic of the belt-conveyor system.
1. Supply voltage: 3-phase, 400 V AC, 50 Hz, ±15%

2. Induction motor: 3-phase, 400 V, 4 pole, 5 HP machine
3. Speed range required: 10–90% of base speed
4. Distance between the control and drive: 500 m
5. Ambient temperature: 50°C
6. MTBF: 1 year
Protection and indication should be provided for over-voltage, over-heating, under-voltage and
over-current.
Stimuli and Responses: To evolve a systematic methodology to the design of electronic equipments, the
outer-to-inner approach is used. In this approach, the equipment is viewed first as a black box and gradually
the various modules and sub-modules are segregated. The actual electronics design of the various sub-mod-
ules is the innermost stage of the outer-to-inner approach.
In this case, there are two sub-systems: (a) the control unit or the user interaction unit and (b) the drive
unit. The control unit is where the user will interact with the system in terms of setting the feed rate, moni-
toring the faults, etc. Therefore, this unit is also called the user-interaction unit. The drive unit performs the
actual speed control of the induction motor. There are two possible ways in which the two sub-systems can
be located.
Chapter 14.indd 806 3/20/2009 7:26:34 PM

1. The control unit and the drive unit can be located away from the induction motor driven belt-
conveyor system. In this case, the output of the drive unit which drives the induction motor
is situated about 500 m from the induction motor. Power will be transmitted through the 500 m
distance and as a consequence, heavy power cable will have to be used which will increase the cost
of the system. Further, as the drive output is pulse-width modulated (PWM), transmission line
effects will result in reflections and waveform distortions in addition to the power loss in the line.
2. The control unit can be placed in the control room which is 500 m away from the belt-conveyor
system. The drive unit is placed close to the induction motor. In this case, there are no transmission line
effects and further the line loss is negligible as only the DC control signals are transmitted over the 500
m distance. Hence this scheme is preferred.
The electronic portion of the system has been physically divided into two sub-systems: (a) the control
unit and (b) the drive unit. It is now required to design these two electronic sub-systems such that the
targeted specifications are met. The first stage of the design process aims at categorizing all the stimuli to
the system and all the responses from the system. This stage is very crucial because all the following pro-
cesses (interconnection, electronic design, etc.) will be affected by the outcome of this stage. Therefore,
one must provide sufficient time to contemplate on the product so that all possible stimuli that the
system may experience are considered. Likewise, one should account for all possible responses from the
system.
Figure 14.28 shows the stimuli and the responses for both the control unit and the drive unit. Those
stimuli and responses that are marked with a “*” are the critical stimuli and responses which correspond to
the primary function of the system. Consider first the control unit.
Referring to Figure 14.28, the stimuli for the control unit are as follows:
1. 230 V AC: This is the mains input which is essential for developing the control power for the entire
electronics of the control unit.
2. Feed Rate set: This is the user-defined feed-rate setting.
3. Weight Feedback: A load cell is used to measure the weight on the belt-conveyor at a given instant of
time.
4. Speed from Tacho: The speed of the induction motor, which is a direct measure of the speed with
which the belt-conveyor is moving, is obtained from a tachogenerator that is mounted on the induction
motor. The weight feedback and the speed feedback from the tacho are used to determine the feed rate
of the belt-conveyor system.
5. Power ON: This stimulus is used to switch ON the control unit to enable interaction and control of the
drive unit.
6. Fault Reset: When any fault occurs, the system is shut down. After maintenance and rectification of
the fault, this stimulus is used to restart the system.
7. Voltage Sense: This is a stimulus that is obtained from the voltage sensors that are mounted in the drive
unit. The DC-link voltage is measured and signal conditioned. This signal is used by the control unit to
detect over- and under-voltage conditions.
8. Current Sense: This is a stimulus that is obtained from the current sensors that are mounted in the
drive unit. The motor currents are measured and the conditioned signal is used as a stimulus for the
control unit for the purpose of current control and over-current detection.
9. Thermal Sense: The temperature of the power-switching devices in the drive unit is sensed and signal
conditioned. This signal is used by the control unit to detect over temperatures.
Chapter 14.indd 807 3/20/2009 7:26:34 PM

230 V AC ∗ Speed command
Feed rate Overvoltage

∗
SET indication
Control unit or user interaction unit

Weight Undervoltage
∗
feedback indication
Speed Overcurrent
∗
from tacho indication
Overheat
Power ON
indication
Fault reset Fault occurred
Current sense
from drive unit
Voltage sense
from drive unit
Thermal sense
from drive unit
(a)
3 phase input
∗ ∗ 3 phase PWM
from mains
output to IM
400 V AC
Voltage sense
Speed output
∗ Drive unit
command Current sense
output
Fault Thermal sense
occurred output
(b)
Figure 14.28 Stimuli and response of the sub-systems.
The responses for the control unit are as follows:

1. Speed Command: The control unit must generate the command signal for the speed so that the drive
unit can use this signal and appropriately control the speed of the induction motor.
2. Over-Voltage Indication: This is a response which is used to drive an LED that lights up when an over-
voltage occurs.
3. Under-Voltage Indication: This is a response which is used to drive an LED that lights up when an
under-voltage occurs.
4. Over-Current Indication: This is a response which is used to drive an LED that lights up when the
load current (motor currents) exceed a certain value.
Chapter 14.indd 808 3/20/2009 7:26:34 PM

5. Over-heat Indication: This is a response which is used to drive an LED that lights up when the power
devices in the drive unit exceed a certain temperature.
6. Fault Occurred: When any fault condition occurs (over- or under-voltage, over-current or over-
temperature), then this response is asserted. This signal is used by the drive unit to shut down the
system under fault conditions.
The stimuli for the drive unit are as follows:
1. Three-Phase Input from Mains: The power to drive the induction motor is drawn from the three-
phase mains.
2. Speed Command: This stimulus is the response (or output) of the control unit. This is used to appropri-
ately set the speed of the induction motor to achieve a feed rate equal to the SET feed rate.
3. Fault Occurred: This stimulus is also the response (or output) of the control unit. This is used to shut
down the drive system when any fault occurs.
The responses of the drive unit are as follows:
1. Three-Phase PWM Output: This is the response (output) that is used to drive the induction motor.
2. Current Sense: This response is obtained by sensing the motor line currents by using a suitable current
sensor and appropriately signal conditioning it.
3. Thermal Sense: This response is obtained by measuring the case temperature of the power-switching
devices (or the heat-sink temperature on which the devices are mounted) with proper signal condi-
tioning.
Functional Block Schematic: The functional block schematic of the system shows the essence of the
system to be designed. Many of the functions such as indications, protections, etc., though they are impor-
tant in a product, do not indicate the primary function. They are only subsidiary circuits that enhance the
main function of the product. Therefore, to obtain a better understanding of the product, one must recog-
nize those stimuli and responses that correspond to the primary function of the product. Thus referring to
Figure 14.28, the stimuli and responses marked with “*” (called the primary stimuli and responses of the
product) correspond to the primary function of the product. All other stimuli and responses are
subsidiary.
From the primary stimuli and responses, the functional block schematic of the system can be devel-
oped. For the current case, the functional block schematic is shown in Figure 14.29. The functional segre-
gation of the control unit and the drive unit is also shown in Figure 14.29. The weight from the load cell
and the speed wm of the induction motor which is fed back is compared with the mechanical speed com-
mand from the output of the PI controller to obtain the speed error. The speed error is passed through
another PI controller to obtain the slip speed wsl. The mechanical speed is added to the slip speed to obtain
the synchronous speed ws. This ws becomes the speed command value that is the primary response of the
control unit.
For the drive unit, the primary stimulus is the speed command value ws from the control unit. The
desired stator voltage Vs for the corresponding ws is obtained from a lookup table. The desired Vs and ws are
the inputs to the PWM modulator. The PWM modulator generates appropriate signals to switch the inverter
switches. The three-phase mains is rectified and filtered to obtain the DC-link voltage for the inverter. The
output of the inverter is the three-phase PWM output of the drive unit which is used to drive the induction
motor.
Chapter 14.indd 809 3/20/2009 7:26:34 PM

Drive unit
3 phase
400 V AC
50 HZ
3 phase
PWM output
to IM
PWM
modulator
ws Vs
Control unit
SET feed rate + + +

wsl ws Vs
Weight − − +
wm V/f curve
from load cell
Feed rate
wm
evaluation Feed rate
from tacho
Figure 14.29 Functional block schematic of the system.
Segregation into Modules: The system should now be broken up into smaller modules and sub-modules
with clear stimuli and responses for each module as shown in Figures 14.30 and 14.31. At this point, it is
also essential to indicate the number of wires or lines required for each stimulus or response. Note that if the
number of wires for a particular stimulus or response is not indicated, then it is taken to be a single wire/
line. It is also important to indicate the power supply requirements for each module at this stage. The segre-
gation of modules for the control unit is shown in Figure 14.30 and the segregation of the modules for the
drive unit is shown in Figure 14.31. Based on the overall target specification of the system, the designer
should generate the specifications for each module at this stage.
The specifications for the modules of the control unit are as follows:
Power Supply Module
Stimuli:
Input: 230 V AC, 50 Hz
Range: 140 V AC to 270 V AC
Power ON switch: 230 V AC, 0.5 A rating with LED indication in switch
Responses:
Outputs: 5 V/2 A ± 5%
: +12 V/1 A ± 10%
: –12 V/1 A ± 10%
Controller Module
Stimuli:
SET feed rate: 0 to 10 V
Weight: 0 to 10 V (equivalent voltage output from load cell)
Speed: 0 to 10 V (equivalent voltage output from tacho)
Chapter 14.indd 810 3/20/2009 7:26:34 PM

230Vac
Power 5V
2 2
supply +12 V
Power ON module 2
−12 V
2 2
+12 V −12 V 5 V
SET
feed rate 2
Weight Controller Speed
feedback 2 module 2 command
Speed
from tacho 2
gnd12 gnd−12 gnd5
+12 V −12 V
Fault reset Overvoltage

2 2
Voltage sense Undervoltage
2 Protection 2
and
Current sense Overcurrent
2 indication 2
Thermal sense module
Overheat
2 2
Fault occured
2
gnd12 gnd−12
Control unit
Figure 14.30 Segregation into sub-modules for the control unit.

Response:
Speed command: 0 to 10 V
Protection and Indication Module
Stimuli:
Voltage sense: 0 to 2 V
Current sense: 0 to 2 V
Thermal sense: 0 to 2 V
Fault reset: Press to ON pushbutton switch
Responses:
Over-voltage: > 660 V DC-link
Under-voltage: < 500 V DC-link
Chapter 14.indd 811 3/20/2009 7:26:34 PM

Power 5V
2
3 phase 400 V AC supply +12 V
3 module 2
−12 V
2
+12 V
AC-DC convertor Vdelink

(Rect. + filter + 2 560 V nominal
3 phase 400 V AC inrush current
3 Voltage sense
limiting ckt.
2 (isolated)
+ bleeder ckt.)
gnd12
+12 V
Vtlclink 3 phase PWM

2 3 output to IM
Vgs1
2
Vgs2 Current sense
2 (isolated)
2
Vgs3 Inverter module
2 Thermal sense
Vgs4 2
2
Vgs5
2
Vgs6
2
gnd12
+12 V 5V
Vgs1
Speed command 2
2 Vgs2
2
Fault occured V/f controller Vgs3
2 and 2
pwm module Vgs4
2
Vgs5
2
Vgs6
2
gnd12 gnd5
Drive unit
Figure 14.31 Segregation into modules for drive unit.
Chapter 14.indd 812 3/20/2009 7:26:34 PM

Over-current: > 10 A line current of IM

Over-heat: > 75oC of inverter heat sink
Fault occurred: 0 V for not asserted
: 12 V for asserted (i.e., fault condition)
The specifications for the modules of the drive unit are as follows:
Power Supply Module
Stimuli:
Input: 3-phase, 400 V AC, 50 Hz, ±20%
Responses:
Outputs: 5 V/2 A ± 5%
: +12 V/1 A ± 10%
: –12 V/1 A ± 10%
AC–DC Converter Module
Stimuli:
Input: 3-phase, 400 V AC, 50 Hz, ±20%
Responses:
Outputs: 560 V DC nominal, ±20%
Output power: 5000 W
Voltage sense: 0 to 400 V AC (transformer isolated)
Inverter Module
Stimuli:
Vdclink: 560 V DC nominal, ±20%
Vgs1…6: 0 to 12 V
Responses:
Output: 3-phase PWM
Current sense: 0 to 10 A rms
Thermal sense: 0 to 75oC
V/f Controller and PWM Modulator Module
Stimuli:
Speed command: 0 to 10 V (represents 0 to 50 Hz)
Fault occurred: 0 V for no fault
: 12 V for occurrence of fault
Response:
Vgs1…6: gate drive
Signals: 0 to 12 V
14.6 Reliability Apportioning/Allocation
A fter the system has been segregated into various modules, it is now essential to allocate the desired
reliabilities for each module or sub-module, such that the total specified system reliability is realized. It
is therefore required to translate the reliability of the entire system into reliability requirements for the mod-
ules. This process is called reliability apportionment or allocation.
Chapter 14.indd 813 3/20/2009 7:26:35 PM

The allocation of system reliability involves solving the following basic inequality
f ( R1 , R2 , …, Rn ) ≥ R (14.61)
where Ri is the allocated reliability for the ith module; R is the system reliability requirement; f is the func-
tional relationship between the module and the system reliabilities.
For the case of a simple n-component series network system, if the reliabilities are considered for t h, then
R1 ⋅ R2 ⋅⋅⋅⋅⋅ Rn ≥ R (14.62)
where Ri is the allocated reliability of the ith module; R is the system reliability requirement.
For the case of an n-component, full-on, single-survivor redundant system, the reliability allocation
inequality is given by
1 − (1 − R1 ) ⋅ (1 − R2 ) ⋅⋅⋅⋅⋅ (1 − Rn ) ≥ R (14.63)
One should note that there are infinite solutions to the inequalities of Eqs. (14.62) and (14.63). However,
it is only required that the allocation of the reliabilities be reasonable. The reliability allocation improves
with experience. The reliability allocation gives a starting reliability value for the module design, which helps
in choice of components, derating and design. After the design of the modules, the reliability prediction will
give more accurate values of the module reliabilities. The reliability allocation and reliability prediction of
the various modules and sub-modules are iterated till the desired system reliability is achieved.
If it is found that the reliability requirements of some individual modules are not met, the designer must
use one or more of the following strategies:
1. Use components with lower hazard rate.
2. Simplify the design by using fewer component parts if it is possible without degrading the functional
performance.
3. Apply component derating techniques to reduce the failure rates of the components (to be dealt later).
4. Use redundancy techniques.
Method 1: Equal Apportionment Technique

In the absence of definite information on the system other than the fact that the n modules are in series con-
figuration, equal apportionment to each module would seem reasonable. In this case, the nth root of the
system reliability requirement would be apportioned to each of the n modules. The main drawback in this
approach is that the reliabilities are not allocated in accordance with the degree of difficulty associated with
achieving these reliabilities for the various modules. For this method, the model is
R = R1 ⋅ R2 ⋅⋅⋅⋅⋅ Rn (14.64)
Here R1 = R2 = … = Rn. Thus,
Ri = R 1/n for i = 1, 2, …, n (14.65)
where R is the required system reliability obtained from the target specifications; Ri is the reliability require-
ment apportioned to module i.
Method 2: AGREE Apportionment Technique

AGREE (Advisory Group on Reliability of Electronic Equipment) has outlined a method of apportionment
for electronic equipments. This technique takes into consideration the complexity and importance of each
Chapter 14.indd 814 3/20/2009 7:26:35 PM

sub-system. It assumes a series of k sub-systems. The apportioned reliability goal is expressed in terms of
MTBF. The minimum acceptable mean life of the ith sub-system is defined as:
Nwi t i
MTBFi = (14.66)
ni [ − ln R (t )]
and the corresponding ith sub-system reliability requirement is given as
−ti / MTBFi
Ri (t i ) = e (14.67)
where i = 1, 2, …, k; t is the required time for which the system should function (mission time); ti the
required mission time for the ith sub-system; wi the importance factor that is expressed as the probability
that the failure of the ith sub-system will result in the system failure, that is, P(system fails/ith sub-system
fails); ni the number of modules in the ith sub-system; N the total number of modules in the system (mod-
ules in all sub-systems put together); R(t) the required reliability of the system for the system mission time;
Ri(ti) the reliability apportioned to the ith sub-system for its mission time; MTBFi the apportioned MTTF
of the ith sub-system.
Method 3: ARINC Apportionment Technique

This method assumes series modules such that any module failure causes system failure and the module mis-
sion time is the same as the system mission time. This apportionment technique requires expression of reli-
ability requirements in terms of failure rate.
The following steps are used in this technique of reliability apportionment:
1. The objective is to choose the failure rates of the modules such that
n
∑ λi ≤ λ
i =1
where li is the failure rate allocated to the ith module and l is the required system failure rate.
2. Determine the module failure rates (lie) from past experience or from modules designed previously that
has similar complexity.
3. Assign a weighting factor (wi ) to each module according to the failure rates determined in (2). Thus,
λie
wi =
∑ λie
all i
4. Allocate the module failure rate as

li = wi l
For the belt-conveyor system being discussed, there are two sub-systems: the control unit and the drive
unit. The control unit consists of three modules and the drive unit consists of four modules. It is now
required to allocate reliabilities for these modules based on the apportionment techniques discussed, so
that there is a starting reliability specification for the design of the electronic circuits for the various mod-
ules. For this case, method 3 can be used as there is some experience with products of similar nature and
complexity.
The required MTBF that is specified is 1 year. Thus, the system failure rate should be less than 1 failure/
year for 24 h a day operation. The various modules of the control unit and the drive unit are connected in
series from the reliability point of view.
Chapter 14.indd 815 3/20/2009 7:26:35 PM

The reliability allocation table (Table 14.1) is based on the reliability apportionment method 3. All
the symbols and relationships have the same meanings as discussed in method 3. It is found that the
total failure rates of all the modules put together is equal to 0.75 failure/year, which is less than the
required failure rate of 1 failure/year for the system. The allocated failure rates for each individual
module is given in the last column, which is the product of the weighting factor wi and the required
failure rate of the system, that is, 1 failure/year in this case. Thus, the starting specifications for the fail-
ure rates of the various modules are as given in the allocated failure rates given in the last column of
Table 14.1. Now the electronics for the various modules will be designed in a manner such that the
allocated failure rates are met.
Table 14.1 Failure rate allocation for the various modules of the belt-conveyor system
Failure rates from Allocated failure
past experience Weighting rates (li failures/
Sub-system Modules (l ie failures/year) factor (wi) year)
Control unit Power supply 0.1 0.133 0.133

Controller 0.1 0.133 0.133
Protection and 0.05 0.066 0.066
indication
Drive unit Power supply 0.1 0.133 0.133
AC–DC converter 0.15 0.2 0.2
Inverter 0.2 0.266 0.266
V/f controller and 0.05 0.066 0.066
PWM modulator
Total 0.75 1 1
14.7 Inter-Connection and Wiring

I n the previous sections, the modules were appropriately segregated and the reliability allocation for the
individual modules was performed. The next step in the design process is to clearly specify the nature of
the inter-connections between the various modules. In this context, it is first necessary to segregate the mod-
ules into printed circuit board (PCB) and non-PCB sub-modules. This exercise is necessary as it has a sig-
nificant bearing on the nature of the inter-connection and wiring. The segregation of the various modules
into PCB and non-PCB modules for the control unit is given in Table 14.2. This segregation into PCB or
non-PCB sub-modules requires some experience and also some insight into the product as a whole. Like in
the case of reliability allocation, this exercise of segregation of the modules into PCB and non-PCB sub-
modules will improve with experience.
After segregating the various modules into the PCB and the non-PCB sub-modules, the various connec-
tors and terminations should be labeled as indicated in Figures 14.32 and 14.33. Any consistent naming
convention can be followed for the inter-connections. Here, the first letter J is reserved for PCB connectors
and the first letter T is reserved for non-PCB terminations. This is followed by one or two letters to identify
the module or sub-module and then followed by a letter count to indicate the number of connectors for a
Chapter 14.indd 816 3/20/2009 7:26:35 PM

Table 14.2 Segregation of modules into PCB and non-PCB sub-modules
Module segregation Module/sub-module

Module into sub-modules type Remarks
Control Unit
Power supply – PCB –
Controller – PCB –
Protection and indication – PCB –
Drive Unit
Power supply – – –
AC–DC converter Rectifier Non-PCB Heat-sink assembly
Filter capacitors –
Inrush + Bleeder circuits –
Driver circuit for inrush PCB –
and bleeder switches
Inverter Inverter switches – Heat-sink assembly
Current and thermal PCB –
sense circuit
V/f controller and PWM PCB –
modulator
given sub-module. The connectors and the associated signals on every pin of the connectors have to be
decided. The connector symbolic names corresponding to the various modules are indicated in Figure 14.32
for the control unit and Figure 14.33 for the drive unit.
It can be noted from the Figures 14.32 and 14.33 that the first letter for the connector nomenclature is
“J” for PCB-mountable connectors and “T ” for non-PCB mountable terminations.
Connector nomenclature is followed by identification of every signal line on the connectors. It should
be understood that the identification of the various signals with the respective connector pins will in
general iterate with the PCB layout and module layouts. Referring to Figures 14.30 and 14.31, it can be
noted that many of the stimuli and responses have two lines (signal and ground). Therefore, a “#” is used
to indicate one of the signal lines and a “*” is used to indicate the other signal line. The last column of
Tables 14.3 and 14.4 indicates which of the signal lines on the connector are connected to specific control
power supply grounds.
Protection
Power JCA JPIA
Controller and
supply
JPA JPB module JCC Indication JPIC
module
JCB (PCB) module
(PCB) JPIB
(PCB)
Figure 14.32 Connector symbolic names for the modules of the control unit.
Chapter 14.indd 817 3/20/2009 7:26:35 PM

TRA Rectifier TRC

inrush
current
limit circuit
filter
Power JDD TRB − bleeder TRD JDA
supply circuit Drive ckts. JDC
JPDA JPDB for bleeder
module
(PCB) switches and
voltage sense
circuit (PCB).
JDB JDD
AC–DC converter module
TIA TIC
Switches
TIB TID JIA JVA V/f controller

and PNM
Current and JVC
module
thermal JIC JVB (PCB)
sense
(PCB)
JIB
Inverter module
Figure 14.33 Connector symbolic names for the modules of the drive unit.
The identification of the various signals of the various modules to the pins of the various connectors for
the control unit is given in Table 14.3.
The identification of the various signals of the various modules with the pins of the connectors for the
drive unit is given in Table 14.4.
Thus, the belt-conveyor system has been divided into sub-systems, that is, the control unit and the drive
unit. All possible stimuli and the responses for the system are identified based on the target specifications. This
is followed by segregation of the sub-systems into modules. The functional specifications for each module are
identified. The reliability apportionment for every module of the system is performed. This is followed by the
segregation of the various modules into PCB and non-PCB sub-modules. The identification of the various
Chapter 14.indd 818 3/20/2009 7:26:35 PM

Table 14.3 Signal–connector pin details for modules of control unit

Signals
Connector Control power
Module name Pin no. Stimuli Responses supply
Power supply JPA 1 230 V – L

2 230 V – N
3 Power on#
4 Power on*
JPB 1 5V
2 gnd5
3 12 V
4 gnd12
5 –12 V
6 gnd–12
Controller JCA 1 Set feed rate#
2 Set feed rate* gnd12
3 Weight#
4 Weight* gnd12
5 Speed#
6 Speed* gnd12
JCB 1 12 V
2 gnd12
3 –12 V
4 gnd–12
5 5V
6 gnd5
JCC 1 Speed com#
2 Speed com* gnd12
Protection JPIA 1 Fault reset#
and indication
2 Fault reset* gnd12
3 Voltage sense#
4 Voltage sense* gnd12
5 Current sense#
6 Current sense* gnd12
7 Thermal sense#
8 Thermal sense* gnd12
JPIB 1 12 V
2 gnd12
3 –12 V
4 gnd–12
(Continued )
Chapter 14.indd 819 3/20/2009 7:26:35 PM

Table 14.3 Continued

Signals
JPIC 1 Over-voltage
2 Under-voltage
3 Over-current
4 Over-heat#
5 Over-heat* gnd12
6 Fault occurred#
7 Fault occurred* gnd12
Note: “#” is used to indicate one of the signal lines and “*” is used to indicate the other signal
line.
Table 14.4 Signal–connector pins identification for the modules of drive unit
Signals
Power supply JPDA 1 400 V – R

2 400 V – Y
3 400 V – B
JPDB 1 5V
2 gnd5
3 12 V
4 gnd12
5 –12 V
6 gnd–12
AC–DC TRA 1 400 V – R
converter
2 400 V – Y
3 400 V – B
TRB 1 Bleeder – Vgs#
2 Bleeder – Vgs* gnd12
TRC 1 Vdclink#
2 Vdclink*
TRD 1 Vdclink – sense#

2 Vdclink – sense*
(Continued )
Chapter 14.indd 820 3/20/2009 7:26:35 PM


Signals
JDA 1 Vdclink – sense#

2 Vdclink – sense*
JDB 1 12 V
2 gnd12
JDC 1 Voltage sense#

2 Voltage sense* gnd12
JDD 1 Bleeder – Vgs#

2 Bleeder – Vgs* gnd12
Inverter TIA 1 Vdclink#

2 Vdclink*
TIB 1 Vgs1#
2 Vgs1*
11 Vgs6#
12 Vgs6*
TIC 1 3-phase PWM
–R
2 3-phase PWM
–Y
3 3-phase PWM
–B
TID 1 Line current#
2 Line current*
3 Heat sense#
4 Heat sense*
JIA 1 Line current #
2 Line current* gnd12
3 Heat sense#
4 Heat sense* gnd12
JIB 1 12 V
2 gnd12
JIC 1 Current sense#
2 Current sense* gnd12
3 Thermal sense#
4 Thermal sense* gnd12
(Continued )
Chapter 14.indd 821 3/20/2009 7:26:35 PM


Signals
V/f controller JVA 1 Speed com#

and PWM 2 Speed com# gnd12
modulator 3 Fault occurred#
4 Fault occurred* gnd12
JVB 1 12 V
2 Gnd12
3 5V
4 gnd5
JVC 1 Vgs1#
2 Vgs1* gnd12
11 Vgs6#
12 Vgs6* gnd12
Note: “#” is used to indicate one of the signal lines and “*” is used to indicate the other signal line.
signals with the various connector pins for all the modules is performed so that all the stimuli, responses and
the control power supply lines are identified and accounted. Now the electronics design for each of the mod-
ules can be performed to achieve the desired performance specification on the responses from the given stimuli.
While designing the electronics for the various modules, one should take care that the allocated reliabilities for
the individual modules are clearly met, such that the overall system reliability requirement is satisfied.
The design process that is discussed for the belt-conveyor system is a general procedure which can be
adopted for any product design. One should note that the number of modules and the segregation of the
modules are not unique for any product. However, the approach to design the product, irrespective of
the non-uniqueness of the segregation of the modules, will be the same. Thus, one will be able to account
for all the stimuli and responses indicated in the target specifications and as a consequence the designs will
be integrated right from the beginning leading to faster design turn-around time.
14.8 General Notes on Designing for Reliability
W hile designing electronic circuits there are some precautions that one must take with regard to some
key aspects in electronic design that will enhance the reliability of the electronic circuits. The follow-
ing are some issues that are generally not addressed in the traditional functional design of the electronic
circuits as most of these aspects are part of system integration.
Connectors
In any product, connections present a major problem from the point of view of reliability. Larger number
of connectors implies lesser reliability. However, with more connectors, it becomes easier to maintain the
product. Thus, connectors present a classic case of trade off between reliability and maintainability.
Chapter 14.indd 822 3/20/2009 7:26:35 PM

Table 14.5 Relative failure rates for various connections

Connection Index
Wrapped joint 1
Welded connection 3
Machine soldered joint (wave soldering) 7
Crimped joint 8
Hand soldered joint 10
Edge connectors (per pin) 30
Table 14.5 presents the various types of connections ranked in the order of reliability starting with the
most reliable. An approximate idea of the relative failure rate is indicated by the index. The most reliable
has an index of unity.
In this context, it is important to realize that the inter-connection stage of the design process must be
iterated to optimize the number of connections. Apart from directly affecting the reliability of the product
because of the inherent failure rates of the connectors, the connectors also affect the failure rate of
product indirectly by adversely affecting the functioning of the circuits (especially high-frequency circuits)
in the following ways:
1. Mutual inductance effect which causes crosstalk between signals.
2. Series inductance effect which slows down the signal propagation and creates electromagnetic interfer-
ence (EMI).
3. Parasitic capacitance effect which slows down signal propagation.
Mutual Inductance Effect: Crosstalk

The mutual inductance effect between signal lines is illustrated in Figure 14.34. The current leaving gate
A returns to its ground through the ground return path of the connector. This results in the loop area A.
The current leaving gate B returns to its ground through the ground return path of the connector resulting
in the loop area B. As the loop areas A and B overlap, magnetic fields from signal A induce electric voltages
in signal path B and vice-versa.
The amount of crosstalk between any signal pins can be estimated if the following quantities are
known:
1. The mutual inductance between the two loops.
2. The maximum rate of change of the source signal di/dt.
3. The impedance of the receiving network and whether it is source or end terminated.
The crosstalk between directly overlapping loops like the loop A and loop B is the worst-case crosstalk.
The contributions to total magnetic flux in loop B come from two places. The first contribution is from cur-
rents flowing out of gate A along its signal wire. The second contribution is from returning signal currents
flowing in the ground wire. The mutual inductance is given by
⎛c ⎞ ⎛ b ⎞
LA,B = 5.08H ln ⎜ ⎟ + 5.08H ln ⎜ ⎟ (14.68)
⎝ ⎠
a ⎝d /2⎠
Chapter 14.indd 823 3/20/2009 7:26:35 PM

Connector
Loop A
A Connector pin
B c
Loop B
Figure 14.34 Mutual inductive coupling effect in connectors.
where a is the distance of signal A to signal B (inches); b the distance of signal B to ground pin (inches);
c the distance of signal A to ground pin (inches); d the diameter of connector pin (inches); H the pin length
in the connector (inches); LA,B the mutual inductance between loops A and B (nH).
Equation (14.68) determines accurately enough whether or not connector crosstalk performance will be
a significant issue. To estimate the amount of crosstalk, the maximum di/dt for the signals in question is
required. This value can be measured using a high-bandwidth oscilloscope. The amount of crosstalk for
drivers that are close to the connector is given by
di
Crosstalk = LA,B (14.69)
dt
It is evident from Eq. (14.69) that slowing down the rise time of the driving signal reduces the crosstalk
levels. The driving signal rise time can be reduced with a capacitor on the source side of the connector as
shown in Figure 14.35. It does not flow through the connector. Therefore, the signal currents through the
connector pins will have lower rise times which will drastically reduce the crosstalk between signals. Use of
a resistor or an inductive bead as shown in Figure 14.35, improves the filtering and makes it more effective
in reducing the crosstalk.
With respect to the crosstalk problems, which are predominant in high-speed digital circuits, the follow-
ing points will aid in their solution:
1. Changing the pattern of the ground connections as indicated in Figure 14.36, one can increase or
decrease the mutual inductance between specific lines. If the ground pin is moved further away from
the signal lines A and B, the dimensions b and c will both increase. As a consequence, from Eq. (14.68)
it becomes evident that the mutual inductance LA,B will increase. Conversely, moving the ground closer
to the signals A and B will decrease their mutual inductance. The change in the inductance is propor-
tional to the logarithm of the distance.
Chapter 14.indd 824 3/20/2009 7:26:35 PM

Resistor or inductive bead

makes capaciter more
effective in reducing crosstalk
Correct capacitor
placement
Wrong capacitor
placement
Figure 14.35 Slowing the rise time of the signal by filtering.
2. Adding extra grounds will have a stronger effect. By adding more grounds, the return currents will
divide among the various ground connectors thereby reducing the mutual inductance drastically. If a
ground pin is placed above the signal line A besides the already existing ground pin below the signal
line B as shown in Figure 14.36, the mutual inductance would drop almost by a factor of two. This
is because the return currents will divide between the two grounds and reduce the mutual induc-
tance. Further, the currents returning through the ground pin above signal A (loop A1 and loop B1)
will be flowing in anti-clockwise direction and the currents returning through the ground pin below
signal B (loop A2 and loop B2) will be flowing in clockwise direction. This would result in some can-
cellation of the fluxes and further reduce the mutual inductance. However, adding more ground pins
above and below the A and B signal lines will not reduce the mutual inductance significantly any
further.
Interposing ground pins between signal lines A and B as shown in Figure 14.37 makes a much bigger differ-
ence than adding grounds outside the signal lines A and B. If n ground pins are added between A and B as
shown in Figure 14.37, then the coupling reduces drastically with n. This is given as follows:
The coupling is proportional to 1/(n2 + 1). Noise coupled onto any given wire is contributed by each of
the other wires in the connector. By reducing the number of signals in a connector, the aggregate crosstalk
can be reduced. Alternately, partitioning a connector into several weakly interacting signal groups by placing
intervening grounds between each group accomplishes the same thing. Adding extra grounds at the end of
a connector does almost nothing to reduce the crosstalk. Large ground lugs at the end of a connector also
will do nothing to reduce crosstalk.
Series Inductance Effect: EMI

Electromagnetic interference (EMI) emanates from signal currents flowing in large loops. This results in
malfunction of the system, thereby affecting reliability. Referring to Figure 14.38, card A sources a 64-bit bus
Chapter 14.indd 825 3/20/2009 7:26:35 PM

Loop A1
A
Loop B1
Loop A2
Loop B 2
Figure 14.36 Adding ground pins on either side of the signals.
n
grounds
Figure 14.37 Interposing ground pins between signal pins.
through connector 1 onto another card B. Signal return currents from these 64 lines flows from card B to
card A mostly through the ground pins in connector 1. However, small fractions of the returning signal cur-
rent flows back to card A through a different path. It may take a path through one of the ground pins of
connector 2 or through some other remote path which may flow through many other cards. It is this small
fraction of the return currents which takes these longer return paths that cause the major problem due to
Chapter 14.indd 826 3/20/2009 7:26:36 PM

Connector
Card A Card B
1
P1
P2
Connector
2
Remote path
Figure 14.38 EMI due to large current loops while using connectors.
EMI. High-frequency currents (especially from digital circuits) flowing in large loops will radiate lots of
electromagnetic energy and will in general not pass the standard radiated emission tests. Therefore, it is
highly important to contain all signal flow to loops having small cross-sections.
Referring to Figure 14.38, a portion of the signal currents flowing from card A to card B through con-
nector 1 will return through the ground pins of the connector 1. This results in the loop path P1 as indi-
cated in Figure 14.38. Likewise, a small fraction of the signal currents will return to card A through some
ground pins of connector 2, which will result in the loop path P2. A still smaller fraction of the signal cur-
rents will return to card A, taking a rather circuitous loop as shown in the figure.
Consider that there are only two connectors as shown in Figure 14.38. The portion of the high-speed
returning signal current that flows through connector 2 depends on the ratio of the inductance of loop path
P1 to the inductance of the loop path P2. At very low frequencies, the amount of the current returning
through the connector 2 depends on the ratio of the resistances. At higher frequencies, it is determined by
the ratio of the inductances. As EMI is a high-frequency problem (> 30 MHz), one need only consider the
inductance ratio of the two loops. Above 30 MHz, the standard limits on radiation are approximately
100 μV/m measured at a distance of 3 m from the equipment. A simple method to estimate the radiated
electric field is as follows:
⎛ AI f ⎞
E = 1.4 × 10−18 ⎜ P c ⎟ (14.70)
⎝ tr ⎠
where E the radiated electric field (in V/m) at 3 m distance; A the radiating loop area (inches2); Ip the peak
current in the loop (A); tr the signal rise time (s); fc the frequency of the loop current (Hz).
Some useful rules for reducing connector emissions are given as follows:
1. Use more grounds in connector 1. This will bring the grounds closer to each signal and lower the
effective radiating loop area in connector 1, that is, reduce the loop path area AP1.
2. Adding more grounds in connector 1 also lowers the inductance of the connector 1. This reduces the
current flow in the remote loops.
Chapter 14.indd 827 3/20/2009 7:26:36 PM

3. Increase series inductance of the remote paths. This is done by using common mode chokes. The
common mode chokes are connected to enclose the signal and return lines of connector 1. This will not
provide any inductance for return paths through connector 1. However, these common mode chokes
will act as differential chokes for all other remote return paths, thereby decreasing the currents through
the remote paths.
4. All connectors on the mother boards should be as close as possible, so that the loop area of the remote
return paths is reduced.
5. If possible, slow down the signal rise and fall times.
14.9 Part Derating

D erating can be defined as the operation of a part at less severe stresses than those for which it is rated.
In practice derating can be achieved by
1. reducing stresses (e.g., snubber circuits as discussed in Chapter 2);
2. increasing the strength of the part.
Selecting a part of greater strength is usually the most practical approach. Derating is very effective because the
failure rate of most parts tends to decrease as the applied stress levels are decreased below the rated value.
Conversely, the failure rate increases when a part is subjected to higher stresses and temperature. Derating is
one of the methods that is used in the iterative design process to improve the reliability. The initial design
should incorporate default derating for the various components. Later the actual derating for the component
should be adjusted based on the overall reliability requirement and the cost requirement. Component derating
is definitely an important aspect of the design process, but it will improve only with experience. The reliability
prediction process that will be discussed in the next chapter will provide the reliability values of the various
components and modules. This too will act as an indicator to derate the components satisfactorily.
Electronic parts are prone to premature failure due to thermal overstress. Decreases in failure rate can be
achieved in these cases by reduction of stress with adequate thermal design. Derating procedures vary with
different types of parts and their application. Resistors are derated by decreasing the ratio of operating power
to rated power; capacitors are derated by maintaining the applied voltage at a lower value than the voltage for
which the part is rated. Semiconductors are derated by keeping the power dissipation below the rated level.
As a general rule, derating should not be conservative to the point where costs rises excessively. Neither
should the derating criteria be so loose as to render reliable part application ineffective. Optimum derating
occurs at or below the point on the stress/temperature curves where a rapid increase in failure rate is noted
for a small increase in temperature or stress.
The term “reliability” has always been used in a sub- system. The functional aspects (intended function)
jective sense to indicate the trustworthiness of a is deterministic and the life aspect (perform the
product or system. However, in order to design a intended function for a specified period of time) is
system for a specified trustworthiness, one should in a sense uncertain. Therefore, reliability is a prob-
be able to quantify reliability. As reliability encom- abilistic measure of the trustworthiness of the
passes both the functional and life aspects of the system.
Chapter 14.indd 828 3/20/2009 7:26:36 PM

In designing a product for a specified reliability, This chapter discusses essentially the methods
the life aspect is specified in terms of the failure rate involved in expressing a given system in terms of
or the MTTF of the product. The issue that arises the reliability models such that they are amenable
after design is with regard to validation. Will the for mathematical analysis and reliability predic-
designed product meet the life requirement that is tion. At the circuit level, all the devices of the
specified? How does one ascertain that the life spec- circuit are in general essential and therefore, the
ifications are incorporated? Evidently one cannot series model is predominant at this level. At a still
wait till the system fails. Validation is performed by higher level, probably the board level, parallel
reliability prediction. However, before even per- model may prevail in some cases. Redundant
forming reliability prediction, one needs to model models are prevalent only at the sub-system and
the system from the reliability point of view in system levels.
order to apply the reliability prediction rules.
1. Consider a building that is powered from the 2. In some cases, especially for new components,
mains grid. Some of the equipments are essen- neither the base failure rate model nor the
tial items that should not face a power shut- nomo-graphs are available to estimate the fail-
down. Therefore, these equipments are provided ure rate of the component. In such cases, the
with alternative power sources. A battery- failure rate has to be estimated by conducting
inverter system is one such alternative power experiments on a population of the device by
source. A diesel generator set is another alterna- inducing failures through accelerated tests. After
tive power source. The essential equipments are the failure data for the population is tabulated,
normally powered by the mains grid. The diesel the device may be characterized either by non-
generator set is maintained in cold standby. The parametric analysis or parametric analysis. In
battery-inverter system is maintained in hot non-parametric analysis, the statistical parame-
standby. The moment the grid power fails, the ters like mean and standard deviations are used
sense/switch network isolates the grid and con- to characterize the device failure. In the para-
nects the battery-inverter system that is in hot metric analysis, a curve fit to exponential distri-
standby thereby continuing to supply power to bution is performed to extract the parameters
the essential load. The battery-inverter system is like failure rate. In this regard, conduct a study
rated to supply power for only 15 min. Before a on the following:
maximum time of 15 min, the diesel generator a. How are accelerated tests conducted?
system starts-up, becomes operative and initi- b. Where is Arrhenius model used in accele-
ates the sense/switch network to change over rated test experiments?
the load from the battery-inverter system to the c. Study the methods of performing non-
diesel generator system. Using variables for the parametric analysis.
failures rates of the various sub-systems, obtain d. Study the methods of performing para-
a relation for the reliability of the uninterrupt- metric analysis.
ible power system. e. What is goodness of the curve fit?
Chapter 14.indd 829 3/20/2009 7:26:36 PM


1. A product may be said to be a compromise 14. In a series reliability model,
among the three primary constraints , components must work to ensure system
and . success.
2. The various electronic components are sub- 15. In a parallel reliability model, compo-
jected to two types of stresses that are nent must work to ensure system success.
and .
16. The switching network and
3. Reliability is the that an item will per- the failed component.
form its intended function for a specified period
17. In dormant-mode failure of the sense/switch
of under the stated conditions.
network, the sensing mechanism fails and the
4. T is a continuous random variable that denotes system to work till the of
the time at which of an item occurs. one of the components of the parallel redun-
dant system.
5. The reliability of an item can be mathemati-
cally defined as . 18. In immediate failure mode of the sense/switch
network, failure in the switching network leads
6. Reliability has been described as quality in the
to failure of the system.
dimension.
19. In standby redundancy with multiple failure
7. Exponential distribution is a special case of
the occurrence of failure among the m opera-
Weibull distribution when the Weibull parame-
tional units is a process.
ter, b is .
20. The main objective of the design process is to
8. Infant mortality period is also called
combine the various and constraints
period.
in an appropriate manner to result in the
9. The failure rate with time during the of the product.
infant mortality period.
21. Larger number of connectors implies
10. The failure rate is with respect to reliability.
time during the useful life period.
22. Connectors present a classic case of trade-off
11. The failure rate with time during the between and .
wear out period.
23. Failure rate of most parts tends to as
12. Useful life period is characterized by the occur- the applied stress levels are decreased below the
rence of failures. value.
13. The failure-time probability density function 24. Th e failure rate of a part when a
of the system during the useful life period can part is subjected to higher stresses and
be modeled as an distribution. temperature.
Chapter 14.indd 830 3/20/2009 7:26:36 PM

1. Discuss the general design process of a product. 16. Write short notes on: (a) dormant mode failure
and (b) immediate mode failure.
2. Discuss the relationships between R(t), F(t)
and Z(t). 17. What is hot standby?
3. What is failure rate? 18. What is the effect of non-ideal sense/switch on
the MTBF of the system?
4. Failure rate or the hazard rate completely deter-
mines the failure-time probability density func- 19. Discuss the method used to evaluate the reli-
tion. Discuss. ability of non-identical components.
5. What is the Weibull model of the failure rate? 20. What are the factors affecting the reliability of
Explain under what conditions the failure rate a system?
becomes constant or increasing with respect to 21. What are target specifications? What are its
time or decreasing with respect to time? uses?
6. Plot and discuss the failure-rate curve during 22. Explain the stimuli and responses for a system.
the life of an electronic equipment.
23. Discuss segregation of modules into sub-
7. Write short notes on: (a) infant mortality, modules.
(b) useful life and (c) wear-out period.
24. Discuss the inter-connection and the naming
8. Discuss the relationship between failure rate, convention for PCB and non-PCB inter-
mean time to failure and mean time between connections.
failures.
25. What is reliability apportionment and what is
9. Write short notes on: (a) series system, (b) para- its importance?
llel system and (c) meshed system.
26. Discuss the various methods of reliability
10. Discuss the methods for evaluating the reliabi- apportionment.
lity of mesh systems.
27. List the connection types in the ascending
11. What is conditional probability approach towards order of failure rates.
evaluating the reliability of mesh systems?
28. What are the methods of reducing the crosstalk
12. What is the limitation of the multiple-path between high-speed signal pins?
equivalent method for evaluating mesh systems?
29. Explain the problems imposed by connectors
13. What is the motivation for using redundant in high-speed digital circuits.
systems?
30. What is parts derating? Why and where is it
14. What are the different redundant topologies? used?
15. What is the function of switching networks in 31. What are the methods of reducing the stresses
redundant systems? within a device?
Chapter 14.indd 831 3/20/2009 7:26:36 PM

| PROBLEMS
1. A system design requires 100 identical compo- 4. A system’s reliability is 0.95 at the end of
nents in series. If the overall reliability must 2 years. What is the mean time to failure of the
not be less than 0.98, what is the minimum system? If the system comprises of 10 identical
reliability of each component? sub-components, then what is the failure rate
of each sub-component?
2. A system’s reliability is 0.9 at the end of 5 years.
What is the mean time to failure of the system? 5. Consider the reliability block diagram shown
If the system comprises 10 identical and essen- in Figure 14.39. Evaluate the reliability of the
tial sub-components, then what is the failure system.
rate of each sub-component?
6. Consider the system shown in Figure 14.40
3. A system design requires 100 identical compo- wherein success requires that at least one path
nents in series. If the overall reliability must exists from point X to point Y. Here the com-
not be less than 0.98, what is the minimum ponent E is a unidirectional component. If
reliability of each component? each component has a failure probability of q,
evaluate the reliability of the system.
0.94
0.25
0.87 0.57
0.99
0.89 0.6
A C
E
X Y
B D
Chapter 14.indd 832 3/20/2009 7:26:36 PM

7. Derive the relationship between MTBF and 13. Derive the relationship between MTBF and
reliability. failure rate for an m-survivor standby redun-
dant configuration.
8. Derive the relationship between MTBF and
failure rate for a m-survivor full-on redundant 14. Derive the reliability equation for the m-survivor
configuration. standby redundant configuration with imper-
fect sense/switch network.
9. A system has three units having identical fail-
ures rates that are connected as 2-survivor full- 15. Consider a system wherein there are five sub-
on redundant configuration. If the reliability of systems. The guess estimate of the failure rates
the overall system is 0.985 at the end of 2 years, of these five sub-systems are 10 FITs, 100 FITs,
then what should be the minimum reliability 140 FITs, 45 FITs and 200 FITs respectively. If
of each redundant unit? What is the failure rate the reliability requirement from the target
of each unit? specification is 700 FITs, then using the
10. Derive the reliability equation for the m-survivor ARINC method of reliability apportionment,
full-on redundant configuration with imperfect obtain the failure rate requirements for the five
sense/switch network. sub-systems.
11. A system comprises two identical units which are 16. Two I/O pins of a microcontroller are buff-
connected in full-on, single-survivor redundant ered and connected to another printed circuit
configuration. If the failure rate of one of the board by means of a connector. The two sig-
units is 0.001 f/h, what should be the minimum nals are connected to adjacent pins of the
reliability requirement for the switching network, connector. On either side of the two signal
with regard to dormant failures, for a period of 5 h pins are two ground pins. The distance
of continuous operation, so that the redundant between any two pins of the connector is 2 mm
system has a reliability greater than the individual and the diameter of the pin is 0.8 mm. The
units? (Assume that the switching network does rise time of the digital switching currents
not give rise to immediate failure of the system). flowing through the buffered signal pins
is 0.1 A/ms. Estimate the mutual inductance
12. For Problem 11, if the probability of immedi- between the two signal pins. Estimate also the
ate mode failure is 0.05, then what should be crosstalk.
the probability that the sense/switch mecha-
nism does not fail in dormant mode such that 17. For the Problem 16, if the ground pins are in
the overall reliability of the system is greater between the two signal pins, then estimate the
than the individual units. crosstalk.
| ANSWERS
Fill in the Blanks
1. performance; cost; life 5. R(t) = 1 – F(t) 9. decreases
2. electrical; temperature 6. time 10. constant
3. probability; time 7. unity 11. increases
4. failure 8. burn-in 12. stress-related
Chapter 14.indd 833 3/20/2009 7:26:36 PM

13. exponential 17. continues; failure 21. lesser

14. all 18. instantaneous 22. reliability; maintainability
15. any 19. Poisson 23. decrease; rated
16. detects; isolates 20. demands; conception 24. increases
Chapter 14.indd 834 3/20/2009 7:26:36 PM

Reliability Prediction
Learning Objectives
CHAPTER
15
understand and apply part stress co-variate method for reliability prediction.
iterate the design to achieve the specified reliability goals.
R eliability prediction is the process of quantitatively assessing whether a proposed or actual equipment/
system design will meet a specified reliability requirement. Reliability prediction does not contribute to
system reliability but does constitute decision criteria for selecting courses of action which affect reliability.
The primary objective of reliability prediction is to provide guidance, relative to expected inherent reli-
ability of a given design. Reliability predictions are most useful and economical during the early phase of a
system design, before the hardware is constructed and tested. During design and development, predictions
serve as quantitative guides by which design alternatives can be judged for reliability. Basically, the purpose
of reliability prediction includes feasibility evaluation, comparison of alternative configurations, identifica-
tion of potential problems during design review, logistics support planning and cost studies, determination
of data deficiencies, trade-off decisions and apportioning/allocation of requirements.
The prediction of the reliability of the electronic product has the following uses:
1. Determination of reliability requirements in planning, preliminary design specifications and determina-
tion of the feasibility of proposed reliability requirement. Comparison of the determined reliability
requirement with state of the art, feasibility for guidance in budget and schedule decisions.
2. Identification and ranking of potential problem areas and the suggestion of possible solutions.
3. Allocation of reliability requirements among the sub-systems and lower level items.
4. Evaluation of the choice of proposed parts, materials, units and processes.
5. Evaluation of the design for prototype fabrication during the development phase.
6. Provides a basis for trade-offs and compromises.
Therefore, reliability prediction is a key to system development and allows reliability to become an integral
part of the design process. To be effective, the prediction technique must relate engineering variables to reli-
ability variables. In general, reliability prediction depends on the
1. depth of knowledge of the design;
2. availability of historical data
Chapter 15.indd 835 3/19/2009 2:51:23 PM

In the initial stages of development, the data available is qualitative which progressively becomes quantita-
tive at later stages. Therefore, there are reliability prediction techniques that can be used at various stages of
the product development.
15.1 Similar Equipment Technique
T he equipment under consideration is compared with similar equipments of known reliability in esti-
mating the probable level of achievable reliability. This is one of the simplest technique used in the early
stages of the system design which involves a simple estimate of the equipment reliability in terms of mean
time between failures (MTBF), failure rate, etc. based on experience gained from similar operational equip-
ments with similar function.
EXAMPLE 15.1 If the reliability of a previously designed DSP board is available, then another DSP
board with DSP from another manufacturer and similar features would result in a
new product with reliability in the same order of magnitude as the old board.
15.2 Similar Complexity Technique
T he reliability of a new design is estimated as a function of the relative complexity of the subject item
with respect to a typical item of similar type. Generally, a nomo-graph between MTBF and the number
of active components in an electronic product is used for the initial reliability prediction. The nomo-graph
between MTBF and the number of active components is developed from past design experience with vari-
ous products. A plot of the MTBF of the various products with respect to the number of active components
can be used as a graphical tool for predicting reliability. In order to have a better estimate of the reliability, a
family of plots of the MTBF versus number of active components with active device rating as a parameter
could be developed.
EXAMPLE 15.2 Consider that past history exists in terms of specification, achieved reliability and
number of active devices for a 5 KVA UPS system. If the requirement is to design
a 50 KVA UPS system, then it may be deduced that the design complexity is
similar to the 5 KVA UPS system; however, the active devices will handle different
electrical and thermal stresses. The nomo-graph of the MTBF versus active device
rating may be a helpful guide to estimate the pre-design reliability of the system.
15.3 Prediction by Function Technique
T his refers to a prediction technique which relates the expected reliability with the functional character-
istics of the equipment or sub-system. This is also an approximate method of reliability prediction that
is useful in pre-design stages of the equipment development. The technique is based upon a statistical cor-
relation between significant functional characteristics and the observed operational reliability of an equipment.
Chapter 15.indd 836 3/19/2009 2:51:24 PM

Reliability Prediction 837
The result is a series of regression equations which relate the more significant equipment functions (from the
reliability viewpoint) to the expected reliability.
EXAMPLE 15.3 Consider the case of an analog-to-digital converter (ADC) card. The mean time to
failure (MTTF) can be related to the most significant functional specifications as
follows:
MTTF = f (number of channels, resolution, accuracy, sampling time).
This relationship is again obtainable from past history or experience. The MTTF
as a function of various significant functional specifications is similar to a curve-fit
algorithm to statistical data.
15.4 Parts Count Technique
T his technique is used when one has a feel for the number of component parts by class or type that will
be used in the equipment but does not have enough data with respect to the stress that each part will
be subjected in the final design. It involves counting the number of parts of each class or type, multiplying
this number by the generic failure rate for each part class and summing these products to obtain the failure
rate for the equipment.
This technique of parts count analysis is generally used in the preliminary design stages where the gen-
eral circuit diagram is fixed but the values of the circuit components are not yet determined. It gives the
worst case value for failure rate for a given number of components assuming series model, that is, all the
components must work for satisfactory performance of the circuit.
The general expression for the item failure rate is
n
λitem = ∑ N i (λGπ Eπ Q )i
i =1
where litem is the total item or equipment failure rate; lGi the generic failure rate of the ith generic part; pE
the environment adjustment factor or environment co-variate; pQ the quality adjustment factor or quality
co-variate; Ni the quantity of the ith generic part; n the number of different generic part categories. The fail-
ure rates litem and lG are expressed in terms of failure units called FITs. FITs is a sort of acronym for Failure
unITs. 1 FIT means 1 failure in 10 9 hours.
There are many standards in use to determine the co-variates and the generic or the base failure rates of
the items. Typical standards are the MIL-217 of US military, the BELL (BELCORE) standards for commer-
cial and communication equipments from AT & T Bell, HRD standards from British Telecom and the
French CNET standards. The MIL-217 is one of the most popular standards used worldwide. This chapter
illustrates the reliability prediction examples using the MIL-217 standards. The relevant failure rate data are
available in the MIL-HAND BOOK. It should, however, be noted that whichever standard is being used,
the same standard should be used as a platform for comparing various other products from the reliability
point of view.
Chapter 15.indd 837 3/19/2009 2:51:24 PM

Problem 15.1
For the series-pass linear regulator shown in Figure 15.1, perform a parts count reliability prediction.
Q1
R1
R3
Vg = 100 V, 10 A
R2
Vout = 75 V, 10 A
Q2
Vg RL Vo Q1 dissipates around (100 − 75) × 10
= 250 W
R4
Dz
Solution
Table 15.1 Parts count failure rate estimation

n
λitem = ∑ N i (λGπ Eπ Q ) Nl
i =1
Components N FITs FITs R(t)
Resistors: 1/4 W cc 5 27.1 135.5 e–135.5t
Transistors: High power 1 16.68 16.68 e–16.68t
Low power 1 3.92 3.92 e–3.92t
Diode: Zener 1 21.68 21.68 e–21.68t
177.78
ltotal = ΣNl = 177.78 failures/109 hours

R(t) = (e–135.5t) (e–16.68t) (e–3.92t) (e–21.68t) = e–177.78t
where a unit time t represents 109 h.
Note: The above data can be obtained from the MIL-HANDBOOK or by executing the MATLAB-based tool-
box for reliability prediction [Reliability for Electronic Circuit (REC)] given in the accompanying CD. The
failure rates are calculated for 50% temperature and electrical stresses. The environment considered is ground
benign (GB) and the quality considered is of quality type “LOWER”.
Chapter 15.indd 838 3/19/2009 2:51:24 PM

Part Count with Non-Series Model

In case if some of the components are non-series in the reliability block diagram, those can be calculated
using techniques discussed in evaluating non-series blocks as discussed in the Section 14.2. They can be
treated as a module (unit). One cannot simply add the failure rates as the equivalent non-series block system
will not have an exponential distribution. In such cases, simply multiply the reliabilities of all generic parts
including the non-series modeled unit.
R(t) = R1(t)R2(t)…Rn(t)
The MTBF can be evaluated using
∞
MTBF = ∫ R (t )dt
0
15.5 Part Stress Technique
T he equipment failure rate is determined as an additional function of all part failure rate and operational
stress levels and derating characteristics of each part. Here the stresses in the components are also taken
into account. This is the most accurate method of reliability prediction that is currently in use.
The part stress technique is very much similar to the part count technique except that the thermal and
electrical stress are evaluated for each part and accounted for in the failure rate model as temperature and
electrical stress co-variates. Here
1. Base failure rate for each part has to be determined. This is determined by failure rate models, appropri-
ate tables or graphs.
2. The values of the adjustment factors or co-variates are to be determined.
3. The part failure rate is calculated using the base failure rate and the co-variates.
The remaining procedure is similar to the parts count technique where the series model is assumed. Here the
item failure rate is given by
n
λitem = ∑ N i (λGπ Eπ Q π Sπ T )i
i =1
where there are two additional co-variate factors: pS the electrical stress factor and pT the temperature stress
factor as compared to the parts count method.
Problem 15.2
Consider the three-phase full-bridge circuit as shown in Figure 15.2. The inverter is used to drive a three-phase
motor which has the following ratings.
Motor rating = 3 kW
Power factor, pf = 0.8
Efficiency, h = 0.9
Evaluate the reliability of the MOSFET bridge. The MOSFETs are known to handle 20 W of power dissipation
and have an ON-resistance of 0.1 Ω. The rise and the fall times can together be taken to be 1 ms. The Vdclink
voltage is 600 V and the switching frequency of the inverter is 20 kHz. However, it should be
Chapter 15.indd 839 3/19/2009 2:51:24 PM

Vdclink
Q1 Q3 Q5
Vg1 Vg3 Vg5
C
B
A
Q2 Q4 Q6
Vg2 Vg4 Vg6
noted that if space-vector modulation is used, the frequency at which the MOSFETs of the inverter will switch
is half of 20 kHz, that is, 10 kHz.
Note: The inverter bridge circuit shown in Figure 15.2 is implemented in practice with IGBTs especially for grid
voltages. This example is only representative and used to illustrate the reliability concepts.
Solution
3 kW
VA = = 4.166 kW
0.8 × 0.9
VA = 3VL I L
VA
IL = = 6 A rms
3 × 400
The peak device current = 2 × 6 = 8.5 A = I m
The current through each switch is a half sinusoid and therefore the root mean square (rms) current
through each switch is approximately Im/2 (neglecting the small periods during diode freewheeling). If each
inverter switch consists of single MOSFET device, then the conduction loss is given as
2 2
⎛I ⎞ ⎛ 8.5 ⎞
Pcond = I ms
2
RDson = ⎜ m ⎟ RDson = ⎜ ⎟ × 0.1 = 1.8 W
⎝ 2 ⎠ ⎝ 2 ⎠
Each MOSFET is switching at a frequency of 10 kHz.
Chapter 15.indd 840 3/19/2009 2:51:24 PM

⎛V I f (t + t ) ⎞
Pswitching = ⎜ dclink aV r f ⎟ × 0.5
⎝ 2 ⎠
(0.5 is used here because only half sine wave flows through each MOSFET in a fundamental period).
2I m
I aV = = 5.4 A
π
Using the appropriate values, Pswitching = 16.2 W.
Ploss = 1.8 + 16.2 = 18 W per MOSFET
Ploss rated = 20 W
From the reliability tables of MIL-Handbook or from the accompanying REC toolbox, the failure rate for
the MOSFET can be computed. The environment is considered as ground benign, and the quality level
is considered as lower. The heat sinks for the MOSFETs are selected such that the operating junction tem-
perature is 100°C for a maximum ambient of 50°C.
The failure rate for each MOSFET is calculated as
litem = 1100 FITs
If for each inverter switch, two MOSFETs are connected in parallel as shown in Figure 15.2, then:
2
⎛ 8.5 / 2 ⎞
Pcond = ⎜ ⎟ × 0.1 = 0.45 W
⎝ 2 ⎠
Pswitch = 8.1 W
Ploss = 8.55 W per MOSFET
The calculated failure rate is now
litem = 743.94 FITs
The reliability of the parallel combination of two MOSFETs is given by (see Chapter 14)
⎛ 2λ2 ⎞ −( λ2 +λ02 )t − λ1t
Rsw = e −2λ2t + ⎜ ⎟ (e −e )
⎝ 1 λ2 − λ02
λ − ⎠
where l1 is the failure rate of the MOSFET when it is sharing the 100% load; l2 the failure rate of the
MOSFET when it is sharing 50% load during parallel operation; l02 the probability that the MOSFET fails
in open-circuit mode during parallel operation = 10% of l2. Statistically it is found that 10% of the time,
the MOSFET initially fails in open-circuit failure mode.
Thus,
Rsw = e −1488t + 5.2841(e −818.4t − e −1100t )
There are six such paralleled MOSFET units in the bridge which are considered as series units from the reli-
ability point of view. Therefore, the overall reliability of the MOSFET bridge would be
Roverall = Rsw
6
= [e −1488t + 5.2841(e −818.4t − e −1100t )]6
The MTTF is obtained by integrating the reliability Roverall from 0 to ∞.
Chapter 15.indd 841 3/19/2009 2:51:25 PM

15.6 Environments
W hile predicting the reliability of the products, the environment in which the product ultimately oper-
ates has a significant bearing on the overall MTTF of the product. All reliability standards define
many standard environments in which a particular product is designed to operate. Each environment type
is associated with a symbolic notation indicating the type of standard environment. The MIL-STD defines
the following types of standard environments and their associated notations:
1. Ground benign, GB.
2. Ground fixed, GF.
3. Ground mobile, GM.
4. Naval sheltered, NS.
5. Naval unsheltered, NU.
6. Air-borne cargo, AIC.
7. Air-borne fighter, AIF.
8. Air-borne unmanned cargo, AUC.
9. Air-borne unmanned fighter, AUF.
10. Air-borne winged, ARW.
11. Space flight, SF.
12. Missile flight, MF.
13. Missile launch, ML.
14. Cannon launch, CL.
15.7 Reliability Prediction Process
R eliability prediction involves the use of nomo-graphs, tables, curve-fit data models and failure rate
equations. Therefore, it is cumbersome to calculate the reliability even for a simple circuit, let alone a
complex one, by manual process. There is a need for reliability evaluation and design tool to aid the iterative
system design implementation. This section formalizes the framework of the reliability prediction process
that can be implemented on program environments like C, MATLAB or SciLAB. However, the reliability
prediction toolbox included in the accompanying CD is with respect to the MATLAB environment.
The design of a circuit with reliability aspects integrated into the design is a three-step process. The
following steps illustrate the circuit-design process:
Step 1: Build the Circuit Template
After finalizing the circuit to be designed, generate a circuit template file that contains all the compo-
nents that go into the circuit. Every component is treated as an object comprising electrical and reli-
ability based variables and constants. The component naming convection and the object definition is
followed as per the discussion in the next section.
Step 2: Perform Functional Design using the Circuit Template
Based on the target specifications of the circuit, the functional design of the circuit is performed. From
the functional design parameters, the reliability aspects like the temperature stress due to the power dis-
sipation in the component and the electrical stresses should be computed to complete the definition
and selection of a particular component object.
Chapter 15.indd 842 3/19/2009 2:51:25 PM

Step 3: Display Results

The design of the circuit should be evaluated by a computer program and the target specifications, func-
tional design parameters, component selection parameters and the failure rate estimates are displayed.
If the predicted MTTF or MTBF value for the circuit does not satisfy the circuit’s life specification, then
one or more of the following actions may be performed in sequence,
1. Without changing the topology of the circuit, derate the component or components that contribute the
most to the overall failure rate of the circuit.
2. Modify portions of the circuit that contribute to large failure rate by simplifying the circuit or using
de-stressing circuits like paralleling of MOSFETs and snubber circuits.
3. If it is not possible to meet the reliability specifications by 1 and 2, then use the redundant topologies
as discussed in the previous chapter.
It should be noted that at the circuit level, all the components are essential, which implies that circuits are series
models from the reliability point of view. The exception to this are circuits like paralleling of MOSFETs and
paralleling of diodes as discussed in examples in the previous chapter. The redundant configurations are used
at a higher level of system integration like sub-system level or even at the system level.
15.8 Component Naming Convention
T he following is the list of generic electronic components used in the reliability analysis and reliability
prediction toolbox (REC). The first letter for a particular component variable is pre-defined (also in the
toolbox) and is indicated in the list given as follows:
1. BJT, Q.
2. Capacitor, C .
3. Connections, SC.
4. Connectors, J.
5. Diode, D.
6. Electronic filters, EF.
7. Fuses, FU .
8. Inductance, L.
9. Lamps, LA .
10. Laser, LAS .
11. Meters panel, MEP .
12. Microcircuits, U .
13. Opto-electronics, OPT.
14. Potentiometer, POT.
15. Quartzcrystals, QZ.
16. Resistor, R.
17. Relays, RY.
18. Rotating device, ROT.
Chapter 15.indd 843 3/19/2009 2:51:25 PM

19. Switches, SW .
20. Transformer, XFM .
21. Transistor-FET, M .
22. Tubes, TUB.
23. Thyristors, SCR.
24. Uni-junction transistor, UJT.
25. Variable capacitor, VC.
26. Miscellaneous, MIS.
27. HYBRID micro, HYB.
28. User defined, USR.
The failure rate or hazard rate is represented as “Hp”. In the case of hybrid packages, the components within
the hybrid package are made up of the standard discrete components. To indicate that the components like
say a BJT is used in the hybrid package, the notation is the component notation followed by ‘H’, that is,
QH. For capacitance used in the hybrid package, the notation is ‘CH’. Thus an ‘H’ is appended to the
normal component notations if they are used within a hybrid package.
15.9 Component Object Notation
A n electronic component as mentioned earlier is an object comprising both electrical and reliability
parameters. Specifying all these parameters would define the specific component. Each category of
component, say a resistor, may contain many types of resistors like film, carbon compound, etc. Each type
may require different parameters to define the resistor object. In the object notations, among the parameters,
the more common parameters are: “type” used to indicate the device type, “PE_type” used to indicate the
environment type, “PQ_type” used to indicate the quality type, “PA_type” used to indicate the application
type, “Tj” used to indicate the operating junction temperature, “Vr” used to indicate the rated voltage, “S”
used to indicate the electric stress ratio and “Pr” used to indicate the rated power. “Hp” is used to indicate
the part hazard rate or the failure rate. All the object parameters for the various components are described in
the Appendix VI. The component object notation is given as follows, using an illustrative example:
Q(1).type
where Q is the device symbol; 1 the device index and type the parameter.
The first letter is the device symbol as discussed in Section 15.8. The number within the parenthesis is the
device index. It can be any integer that uniquely identifies the device within the device class. The parameter
that needs to be quantified is indicated following the period (.). The REC toolbox also follows the component
object notation discussed in this section. It generates the circuit-design template file based on this compo-
nent object notation.
It should be noted that all the parameters within the component object should be defined either by a
relevant constant or an equation based on the circuit specification in order to wholly specify the component.
Based on such a component specification, a component with appropriate rating from any particular manu-
facturer may be selected. A sample list of some major component objects is given below.
Chapter 15.indd 844 3/19/2009 2:51:25 PM

(Refer to Appendix VI for a complete description of the component parameters.)

BJT component object D(1).PC =
Q(1).type = D(1).Vo =
Q(1).PE_type = D(1).Vr =
Q(1).PQ_type = D(1).Hp=
Q(1).Tj =
Q(1).VCE = Electronic flter object
Q(1).Vr = EF(1).PE_type =
Q(1).Pr = EF(1).PQ_type =
Q(1).PA_type = EF(1).Hb_type =
Q(1).Hp= EF(1).Hp=
Capacitor component object Fuse object
C(1).type = FU(1).PE_type =
C(1).PE_type = FU(1).Hp=
C(1).PQ_type = Inductor object
C(1).T = L(1).type =
C(1).S = L(1).PE_type =
C(1).C = L(1).PQ_type =
C(1).Hp= L(1).THS =
Connection object L(1).Hp=
SC(1).type = Lamp object
SC(1).PE_type = LA(1).PE_type =
SC(1).PQ_type = LA(1).PA_type =
SC(1).N1 = LA(1).PU =
SC(1).N2 = LA(1).Vr =
SC(1).P = LA(1).Hp=
SC(1).Hp=
Laser component object
Connector object
LAS(1).type =
J(1).type =
LAS(1).PE_type =
J(1).PE_type =
LAS(1).HM_type =
J(1).PQ_type =
LAS(1).HC_type =
J(1).C =
LAS(1).Hp=
J(1).TA =
J(1).G = Meter panel object
J(1).i = MEP(1).PE_type =
J(1).Hp= MEP(1).PQ_type =
MEP(1).PF_type =
Diode component object
MEP(1).PA_type =
D(1).type =
MEP(1).Hp=
D(1).D_type =
D(1).PE_type = Microprocessor object
D(1).PQ_type = U(1).PE_type =
D(1).Tj = U(1).PQ_type =
Chapter 15.indd 845 3/19/2009 2:51:25 PM

U(1).pkg = U(5).CFT =
U(1).y = U(5).Tj =
U(1).Np = U(5).pt =
U(1).family = U(5).mpt =
U(1).N = U(5).VTH =
U(1).Tj = U(5).XS =
U(1).Hp= U(5).DA1 =
U(5).Hp=
Linear IC object
U(2).PE_type = Digital IC object
U(2).PQ_type = U(6).PE_type =
U(2).pkg = U(6).PQ_type =
U(2).y = U(6).y =
U(2).Np = U(6).pkg =
U(2).N = U(6).Np =
U(2).Tj = U(6).family =
U(2).Hp= U(6).N =
Logic array object U(6).Tj =
U(3).PE_type = U(6).Hp=
U(3).PQ_type =
U(3).pkg = Opto electronic component object
U(3).y = OPT(1).type =
U(3).Np = OPT(1).PE_type =
U(3).family = OPT(1).PQ_type =
U(3).N = OPT(1).Tj =
U(3).Tj = OPT(1).Hb_type =
U(3).Hp= OPT(1).Hp=
Memory component object Potentiometer object

U(4).PE_type = POT(1).type =
U(4).PQ_type = POT(1).PE_type =
U(4).pkg = POT(1).PQ_type =
U(4).y = POT(1).R =
U(4).Np = POT(1).S =
U(4).family = POT(1).T =
U(4).B = POT(1).Ntaps =
U(4).MT = POT(1).R1 =
U(4).C = POT(1).Pa =
U(4).ECC = POT(1).VR =
U(4).Tj = POT(1).PC_type =
U(4).Hp= POT(1).Hp=
VLSI component object Quartz crystal component object
U(5).PE_type = QZ(1).PE_type =
U(5).PQ_type = QZ(1).PQ_type =
U(5).Np = QZ(1).F =
U(5).pkg = QZ(1).Hp=
Chapter 15.indd 846 3/19/2009 2:51:25 PM

Resistor component object MOSFET component object

R(1).type = M(1).type =
R(1).PE_type = M(1).PE_type =
R(1).PQ_type = M(1).PQ_type =
R(1).T = M(1).Tj =
R(1).S = M(1).Hb_type =
R(1).Pd = M(1).Hp=
R(1).Hp=
Tube component object
Relay component object TUB(1).type =
RY(1).type = TUB(1).PE_type =
RY(1).PE_type = TUB(1).T =
RY(1).PQ_type = TUB(1).Hb_type =
RY(1).Hb_type = TUB(1).PK_type =
RY(1).Hp= TUB(1).Hp=
Rotating device object Thyristor component object
ROT(1).type = SCR(1).PE_type =
ROT(1).T_type = SCR(1).PQ_type =
ROT(1).M_type = SCR(1).Tj =
ROT(1).T = SCR(1).VBo =
ROT(1).LC = SCR(1).VBr =
ROT(1).Hp= SCR(1).Ifrms =
Switch component object SCR(1).Hp=
SW(1).type = Unijunction transistor object
SW(1).PE_type = UJT(1).PE_type =
SW(1).PQ_type = UJT(1).PQ_type =
SW(1).PC_type = UJT(1).Tj =
SW(1).PL_type = UJT(1).Hp=
SW(1).S =
Variable capacitor object
SW(1).Hp=
VC(1).type =
Transformer object VC(1).PE_type =
XFM(1).type = VC(1).PQ_type =
XFM(1).PE_type = VC(1).T =
XFM(1).PQ_type = VC(1).S =
XFM(1).THS = VC(1).Hb_type =
XFM(1).Hp= VC(1).Hp=
Example Hybrid component object

%-------------HYBRID CIRCUIT PACKAGE BEGINS-------------
HYB(1).m_type = [1 4 5 6 7 8 10];
HYB(1).DA_type = 2;
HYB(1).DA =
HYB(1).PD =
HYB(1).TC =
HYB(1).cdn = 0;
Chapter 15.indd 847 3/19/2009 2:51:25 PM

HYB(1).Tj=Tj_hybrid(HYB(1).m_type,HYB(1).DA_type,HYB(1).DA,HYB(1).
PD,HYB(1).TC,HYB(1).cdn);
QH(1,1).type = 1;
QH(1,1).Tj = HYB(1).Tj;
QH(1,1).VCE =
QH(1,1).Vr =
QH(1,1).Pr =
QH(1,1).PA_type = 2;
QH(1,1).Hp=hybrid_BJT(QH(1,1).type,QH(1,1).Tj,QH(1,1).VCE,QH(1,1).
Vr,QH(1,1).Pr,QH(1,1).PA_type);
CH(1,2).type = 1;
CH(1,2).T = HYB(1).Tj;
CH(1,2).S =
CH(1,2).Hb_type = ;
CH(1,2).C =
CH(1,2).Hp=hybrid_capacitor(CH(1,2).type,CH(1,2).T,CH(1,2).
S,CH(1,2).Hb_type,CH(1,2).C);
DH(1,3).type = 1;
DH(1,3).D_type = 1;
DH(1,3).Tj = HYB(1).Tj;
DH(1,3).PC = 1;
DH(1,3).Vo =
DH(1,3).Vr =
DH(1,3).Hp=hybrid_diode(DH(1,3).type,DH(1,3).D_type,DH(1,3).
Tj,DH(1,3).PC,DH(1,3).Vo,DH(1,3).Vr);
MH(1,4).type = 2;
MH(1,4).Tj = HYB(1).Tj;
MH(1,4).PA_type = 1;
MH(1,4).P =
MH(1,4).F =
MH(1,4).PM_type = 3;
MH(1,4).Hp=hybrid_FET(MH(1,4).type,MH(1,4).Tj,MH(1,4).PA_
type,MH(1,4).P,MH(1,4).F,MH(1,4).PM_type);
HYB(1).CKT = 5;
HYB(1).PE_type = 1;
HYB(1).class = 16;
HYB(1).Y =
HYB(1).Hp_=hybridmicro(HYB(1).CKT,HYB(1).PE_type,HYB(1).
class,HYB(1).Y);
Hc = [QH(1,1).Hp CH(1,2).Hp DH(1,3).Hp MH(1,4).Hp ];
HYB(1).Hp = sum(Hc)*HYB(1).Hp_;
%------------HYBRID CIRCUIT PACKAGE ENDS-------------
Chapter 15.indd 848 3/19/2009 2:51:25 PM

15.10 Example Reliability Prediction Process

Steps to Design a Circuit
1. Generate the design template for the circuits.
2. Enter the specifications and the design equations for all the parameters of all the component objects.
3. Execute the program to compute the functional parameters and the reliability parameters and display
the results.
4. If the MTTF of the system is not satisfactory, then iterate the design by appropriately changing the rating.
Problem 15.3
A Simple BJT Base Drive: Design/select the components of the circuit in Figure 15.3 according to the functional
specification given below and predict the reliability after 1 year of operation.
ic
Vb R1 ib
Q
R2
Solution
Circuit Description: This base drive circuit is the basic building block for many complex switching tran-
sistor drives. This is the simplest of all base drive circuits. For low-power and low-frequency applications,
this base drive circuit is well suited. This circuit does not provide the starting surge base current ib+ during
the turn-ON time. The resistor R2 provides the discharge path for removing the base charges in the transis-
tor Q. As a rule of thumb, it is found that ibon = 2 (ic/hFEsat) and iR2 = ibon gives satisfactory performance.
The equations to determine the value of the resistors are
Vbesat Vbesat × hFEsat
R2 = = (15.1)
iR2 2ic
Vb − Vbesat (Vb − Vbesat ) × hFEsat

R1 = = (15.2)
2ibon 4ic
The following is the functional specification for the above circuit:
Specifications
Vb=5; //volts pulse amplitude
fs=20e3; //switching frequency in Hz
ic=1; //amps maximum during on time
Chapter 15.indd 849 3/19/2009 2:51:25 PM

Step 1: Generate the design template for the circuit. The template file is as follows.
//---------------------------------------------------
Q(1).type =
Q(1).PE_type =
Q(1).PQ_type =
Q(1).Tj =
Q(1).VCE =
Q(1).Vr =
Q(1).Pr =
Q(1).PA_type =
Q(1).Hp=
R(1).type =
R(1).PE_type =
R(1).PQ_type =
R(1).T =
R(1).S =
R(1).R =
R(1).Hp=
R(2).type =
R(2).PE_type =
R(2).PQ_type =
R(2).T =
R(2).S =
R(2).R =
Step 2: Enter the specifications and the design equations in the generated design template at the appro-
priate lines. This would actually be the functional design activity that the designer has to incorporate
into the design template. The updated circuit-design file will now be as follows.
//------------SPECIFICATION OF THE CIRCUIT-------------
Vb =5;
fs =20e-3;
ic =1;
Vbe_sat=0.8;
hfe_sat=50;
Rat_pow_tr= 0.8;
Rat_pow_res= 0.25;
Vce_sat=0.2;
Vce_rated = 80;
T_amb =28;
R_thermal =187; // Thermal Resistance junction to Ambient
ib_on = 2*ic/hfe_sat;
//---------------------------------------------------
Q(1).type = 1;
Q(1).PE_type = 1;
Chapter 15.indd 850 3/19/2009 2:51:25 PM

Q(1).PQ_type = 4;
Q(1).Tj = T_amb + R_thermal*(Vce_sat*ic*0.5); //Ave. Conduction
losses
Q(1).VCE = Vce_sat;
Q(1).Vr = Vce_rated;
Q(1).Pr = Rat_pow_tr;
Q(1).PA_type = 2;
Q(1).Hp=BJT(Q(1).type,Q(1).PE_type,Q(1).PQ_type,Q(1).Tj,Q(1).
VCE,Q(1). Vr,Q(1).Pr,Q(1).PA_type);
R(1).type = 1;
R(1).PE_type = 1;
R(1).PQ_type = 6;
R(1).T = T_amb;
R(1).power =(Vb - Vbe_sat)*2*ib_on*0.5; // Avg power calculated from
the switching waveform.
R(1).S = R(1).power /Rat_pow_res;
R(1).R = (Vb - Vbe_sat)/(2*ib_on);
R(1).Hp=resistor(R(1).type,R(1).PE_type,R(1).PQ_type,R(1).T,R(1).
S,R(1).R);
R(2).type = 1;
R(2).PE_type = 1;
R(2).PQ_type = 6;
R(2).T = T_amb;
R(2).power= Vbe_sat*ib_on*0.5;
R(2).S = R(2).power /Rat_power;
R(2).R = (Vbe_sat)/ib_on;
R(2).Hp=resistor(R(2).type,R(2).PE_type,R(2).PQ_type,R(2).T,R(2).S,
R(2).R);
Step 3: Execute the program and display the results. The result display will be as follows.
RELIABILITY PREDICTION FOR THE CIRCUIT
Part type PE_type PQ_type Tj VCE Vr Pr PA_type Hp

No.
Q1 1 1 4 46.7 0.2 80 0.8 2 0.19254
Part type PE_type PQ_type T power S R Hp
No.
R1 1 1 6 28 0.168 0.672 52.5 8.6894
R2 1 1 6 28 0.016 0.064 20 2.8429
Chapter 15.indd 851 3/19/2009 2:51:25 PM

CATEGORY-WISE HAZARD RATE

1. BJT, FITs = 0.19254
2. Resistor, FITs = 11.5323
TOTAL HAZARD RATE, FITs = 11.7249
MEAN TIME TO FAILURE (MTTF) = 9736.1778 years
This result is used as the basis for design iteration for derating and improving portions of the circuit
having high failure rates.
15.11 Physics of Failure (PoF)

C urrently, reliability prediction is based primarily on “Part Stress Analysis” or the co-variate approach.
Here the failure rates are probabilistic models that have been developed using accelerated failure induc-
tion. These models are valid for large samples. However, they cannot be used for predicting the failure of
individual components. If one has to focus on individual component failure, then the failure mechanisms of
the component has to be well understood. This means a physics of failure (PoF) model has to be evolved for
each individual component. Reliability prediction if based on PoF models would then be more accurate in
predicting the failures of even the individual components.
A central feature of the PoF approach is that reliability modeling (time-to-failure modeling) is based on
root cause failure processes or mechanisms. Failure mechanism models explicitly address the design, mate-
rial, process and environment parameters causing the failures. In this respect, PoF focuses on the following
three broad sections:
1. Identification of all possible failure mechanisms for the device.
2. Mathematical representation (model) of failure mechanisms as a function of time. In predicting the
reliability of systems, some parts will have the PoF models available and some may not have the PoF
models as yet available. In this context, how does one mix the co-variate approach and PoF approach
such that one obtains failure rates that are more accurate than using only co-variate approach.
The shortcomings of traditional approach to design for reliability are becoming prominent as the manufacturing
technology of electronic components improves. These design and reliability practices, based on stochastic models,
discourage the use of newer, state-of-the-art, low-cost components as the use of newer components would result
in lower life prediction because of less knowledge and confidence in them. The part stress methods for predicting
the product life are therefore being discredited. More deterministic and scientific methods are required for
assessing failure rates and identifying a validated model that considers all potential failure mechanisms. The PoF
methodology enhances prediction of reliability by addressing the failure root cause by modeling failure mecha-
nisms and the application environment. PoF approach enables prediction of failures within the operating lifetime
of the product and the designer can then eliminate such failures by design techniques.
The “physics-of-failure” approach to reliability prediction has the following problems that prevent it
being used currently:
1. All possible failure mechanisms are yet to be identified and therefore not available.
2. Different failure mechanisms result in different predicted failure times.
The most prominent device failure-rate prediction methodology is the US MIL-HDBK-217, which uses a
co-variate model. The basic concept of MIL-HDBK-217 is to use historical part failure data to build an esti-
mate of future system reliability. The handbook contains failure rate models for major electronic component
Chapter 15.indd 852 3/19/2009 2:51:25 PM

categories used in modern military systems. It contains two basic methods for performing a reliability pre-
diction – the parts count method and the parts stress method. Both methods are based on empirical obser-
vations of historical data. The parts stress method is applied at the component level based on an analysis of
component stresses by the system designer or reliability engineer. It uses a base failure rate for each compo-
nent, which is a function of the applied stress or circuit complexity or package used. This failure rate is mul-
tiplied by a number of factors which account for environmental effects, quality and learning curve.
The parts count method is a special case of the parts stress method which assigns typical default values
for the component stresses. Using either method, component failure rates are calculated for each component
and summed for all components to estimate an overall system failure rate.
MIL-HDBK-217 has a few drawbacks. One is that it relies on field failure data. These are not easily
available until well after a new technology has matured. Hence, it heavily penalizes new materials, structures
and technology.
The PoF approach is a significantly more complex approach compared to the traditional MIL-HDBK-217
approach. This is because each and every potential failure mechanism must be analyzed to determine the
MTTF. The failure mechanism with the shortest calculated life then becomes the weak link. Using this
approach, the MTTF for the system is the same value as the calculated life of the weakest component
making up the system. The PoF failure mechanism models are theoretically sound from the standpoint that
they have been developed from existing principles of physics, chemistry, mechanics and materials science,
however, they are not exhaustive.
EXAMPLE 15.4 Failure Mechanisms for Resistors

Two important failure mechanisms for the resistor component are described here.
Electro-migration: Electro-migration is the movement of metal atoms of a metal line
in the direction of the current flow through that metal line. This is a common failure
mechanism in film resistors. As such, during electro-migration, metal atoms move
from the starting end of the metal line and accumulate at the other end, forming
voids at the entrance and hillocks at the exit of the metal line. Thus, electro-migra-
tion can result in open circuits (due to the voids) or line-to-line short circuits (due
to the hillocks). The time to failure due to electro-migration is given as
t = CJ −n e ( Ea / KT )
where C depends on metal properties; J the current density; n varies between 1
and 7 depending on material; Ea the activation energy; K the Boltzmann constant;
T the temperature in K.
Corrosion: This failure mechanism is observed in wire-wound resistors. Corrosion
is metal degradation due to chemical or electrolytic reactions in the presence of
moisture, contaminants and bias. The time to failure due to corrosion is given as
t = AV − n H − m e ( Ea / KT )
where V is the voltage across the resistor; H the relative humidity.
The problem with this approach is that all failure mechanisms have not been
studied and/or modeled. Hence, the predicted life may be greater than the actual
life if an unidentified failure mechanism were to be activated. Secondly, the method
Chapter 15.indd 853 3/19/2009 2:51:25 PM

is much more computationally complex than the co-variate approach. A third

problem is that failure can occur in multiple ways, while PoF is deterministic. If
PoF were to be believed, failure could occur only by that mechanism with the least
predicted time to failure, but empirical observations show otherwise.
As PoF models are as yet in the nascent stage of development, only some of
the PoF models for some of the failure mechanisms are available. Therefore, the
best that can be done for now is to include deterministic knowledge of some of
the PoF mechanisms into the co-variate models to improve the reliability predic-
tion for such devices.
The previous chapter discussed the methods for tive action if the predicted failure rate does not meet
obtaining a reliability model for a given electronic the required life specifications. The corrective actions
system. In this chapter, the issue of predicting the to improve the failure rate of the circuit is by (a)
reliability of an electronic circuit is addressed. The derating the critical parts; (b) reduction of stress by
life specification of a system is in terms of the failure using snubbers; (c) modify the circuit and (d) use
rate or mean time to failure. The main function of redundancy.
the reliability prediction is to estimate the failure While reliability model is performed in a top-
rate of the designed system. The predicted failure down manner from the system to the device level,
rate is used to validate the circuit design in terms of the reliability prediction is performed in a bottom-
the life specifications. up manner from the device to the system level.
The reliability prediction has the added function However, both need to be performed in order to
of providing the necessary feedback to apply correc- quantify the reliability of a given system.
1. Study the user’s manual of the REC toolbox 6. How does one estimate the number of gates
(reliability for electronic circuits). within a digital IC?
2. Study the sample circuits designed using the 7. How does one estimate the number of transis-
circuit template files generated by the REC tors in a linear IC?
toolbox that is included in the accompanying
8. Design a sample hybrid microcircuit and esti-
CD.
mate its reliability using REC toolbox.
3. Observe the failure rate contribution of the
Term study task:
various classes of the system.
The additional information of the physics of
4. Observe the failure rate contribution of the failure model can be incorporated along with
individual devices of the system. the co-variate model to form a hybrid reliabil-
ity prediction method. Such a hybrid method
5. Adjust the operating electrical and/or tempera-
will provide an estimate of the reliability that
ture stresses and re-execute to observe the
will not be as accurate as the deterministic
changes in the failure rate of the system.
physics of failure model, but will definitely be
Chapter 15.indd 854 3/19/2009 2:51:25 PM

better than the stochastic parts stress co-variate (b) Evolve methods to combine the informa-
model. In this regard, tion of the physics of failure model with
(a) Study the various failure mechanism from the co-variate prediction method.
the physics of failure point of view.

1. Parts count technique is used when the general is stress co-variate and
diagram is finalized but the stress co-variate.
of the circuit components are not yet deter-
6. Most of the electronic products are designed
mined.
for the environment.
2. 1 FIT means one failure in hours.
7. Reliability prediction involves the use of
3. The components of the electronic circuits are , data models and failure
modeled as circuit. rate equations.
4. Parts stress technique is used considering all 8. In part stress analysis method, all the compo-
operational . nents are and therefore from the reli-
ability point of view the circuit is a
5. Parts stress technique uses two extra co-variates
model.
as compared to the parts count technique that
1. What is reliability prediction? 9. The part stress technique is very much similar
to the part count technique except that the
2. Why does one need to perform reliability pre-
thermal and electrical stresses are evaluated for
diction?
each part.
3. What are the different methods of reliability
10. What are the standard environments defined
prediction that are applicable at different stages
by the MIL-STD?
of the product development?
11. Explain the steps in the reliability prediction
4. Write short notes on: (a) similar equipment
process.
technique, (b) similar complexity technique
and (c) similar function technique. 12. Discuss the component naming convention.
5. What is generic or base failure rate? 13. Discuss the component object notation.
6. Discuss the environment co-variate. 14. What is physics of failure model for components?
7. Discuss the quality co-variate. 15. What are the advantages and limitations of
physics of failure model?
8. What are the different standards available used
to determine the co-variates and the base failure 16. Why is the physics of failure model not as pop-
rates? ular as the co-variate model?
Chapter 15.indd 855 3/19/2009 2:51:26 PM

| PROBLEMS
Use the reliability toolbox provided in the 2. The specifications are as follows:
accompanying CD to estimate the reliability of a. Input voltage: Vin = 15 V
the following circuits (Figures 15.4–15.6). b. Output voltage: Vo = 22 V
Find out the class of components that contrib- c. Output current: Io = Vo/Ro = 100 mA
ute most to the failure rate of the system. Also d. Switching frequency: fs = 22.22 kHz
identify the critical components of the system e. Ripple in output voltage: 14 mV
in terms of failure rate.
3. The specifications are as follows:
1. The specifications are as follows: a. Output voltage: Vo = 5 V DC
a. Output voltage: 15 V DC b. Output current: Io = 10–250 mA
b. Output current: 50–500 mA c. Input voltage: Vin = 15–40 V DC
c. Input AC voltage: 230 Vrms ± 20%; 50 Hz
1N4007
D5
LM350
IN OUT
ADJ
1N5408
1N5408
100 Ω
D1 D3 U1 R1
XFM1
2200 μF
2200 μF
+ +
0.1 μF
1 kΩ
C1 C2 POT1 C4 Ro
+
4.7 μF
230:48 C3
1N5408
1N5408
560 Ω
D2 D4 R2
Figure 15.4 Circuit 1.
1.7 mH FR304
L1
D1
1.5 kΩ R1
Vin 220 Ω
+ Q1 + Ro
15 VDC 100 μF C1
R2 Q2N3019
−
560 Ω R3
From TL494
Chapter 15.indd 856 3/19/2009 2:51:26 PM

D1
L1
FR304 +
C1 Ro
Vcc
220 μF
N1:N2
1.5 kΩ R1
+
Vin
− M1
27 Ω
IRF840
R2
R3 5.6 kΩ
From TL494
d. Maximum peak-to-peak ripple in f. Operating core flux density: Bm = 0.25 T

output voltage: Vr = 0.05 V g. The control power supply voltage:
e. Switching period = 45 μs Vcc = 15 V
| ANSWERS
Fill in the Blanks
1. circuit; values 4. stresses 7. nomo-graphs; curve-fit
2. 109 5. temperature; electrical 8. essential; series
3. series 6. ground benign
Chapter 15.indd 857 3/19/2009 2:51:26 PM

Chapter 07.indd 421 3/13/2009 7:08:49 PM
Appendix I
Area Product for the Saturable-Core Transformer

Refer to Chapter 7 for a discussion on the area product approach to design of the transformers.
K w Aw = N p1ap1 + N p2 ap2 + N b1ab1 + N b2 ab2 + N s as (A1.1)
where ai is the wire cross-section area.

ip1
ap1 = = ap2
J
ib1
ab1 = = ab2
J
is
as =
J
dφ 2Bm
E p1 = N p1 = N p1 Ac = 4 N p1 Ac Bm f s (A1.2)
dt Ts / 2
E p1
N p1 = N p2 = (A1.3)
4 Ac Bm f s
Vb Nb1
Ep1 Np1
Ns Es
Np2
Nb2
Figure A1 Saturable-core transformer.
Appendix I.indd 859 3/23/2009 7:37:25 PM

Es
Ns = (A1.4)
4 Ac Bm f s
Vb
N b1 = N b2 = (A1.5)
4 Ac Bm f s
2 E p1ip1 2Vbib1 E sis
K w Aw J = + + (A1.6)
4 Ac Bm f s 4 Ac Bm f s 4 Ac Bm f s
Simplifying Eq. (A1.6) we get
1 ⎛ I in I ⎞
K w Aw J = ⎜ 2Vin + 2nbVin in + E sis ⎟ (A1.7)
4 Ac Bm f s ⎝ 2 hFE 2 ⎠
where Vin is the DC-link voltage for push–pull; Iin the current through Vin connected to center-tap of
push–pull transformer.
Vb N b2 N b1
nb = = =
Vin N p2 N p1
1 ⎛ I in ⎞
ib1 = ⎜ ⎟
2 ⎝ hFEmin ⎠
Simplifying Eq. (A1.7), one obtains
1 ⎛ nb ⎞
K w Aw J = ⎜ 2 ⋅ Pin + 2 ⋅ Pin + Po ⎟ (A1.8)
4 Ac Bm f s ⎝ hFEmin ⎠
where Po = E sis . From Eq. (A1.8), the area product for the saturable-core transformer of Figure A1 is given by
Po ⎛ 2 2 ⋅ nb ⎞
Ap = ⎜ + + 1⎟ (A1.9)
4 K w JBm f s ⎝ η η ⋅ hFEmin ⎠
Appendix I.indd 860 3/23/2009 7:37:26 PM

Appendix II
Properties of Few Ferrite Cores

Mean
magnetic Core cross- Window
Cores without Mean length length lm section area area, Area product
air gap per turn (mm) (mm) Ac (mm2) Aw (mm2) Ap (mm4)
Pot Cores
P18/11 35.6 26 43 27 1161
P26/16 52 37.5 94 53 4982
P30/19 60 45.2 136 75 10200
P36/22 73 53.2 201 101 20301
P42/29 86 68.6 264 181 47784
P66/56 130 123 715 518 370370
EE Cores
E20/10/5 38 42.8 31 47.8 1481
E25/9/6 51.2 48.8 40 78 3120
E25/13/7 52 57.5 55 87 4785
E30/15/7 56 66.9 59.7 119 7104.3
E36/18/11 70.6 78 131 141 18471
E42/21/9 77.6 108.5 107 256 27392
E42/21/15 93 97.2 182 256 46592
E42/21/20 99 98 235 256 60160
E65/32/13 150 146.3 266 537 142842
UU Cores
UU 15 44 48 32 59 1888
UU 21 55 68 55 101 5555
UU 23 64 74 61 136 8296
UU 60 183 184 196 1165 228340
UU 100 29.3 308 645 2914 1879530
Toroids
T 10 12.8 23.55 6.2 19.6 121.52
T 12 19.2 30.4 12 44.2 530.4
T 16 24.2 38.7 20 78.5 1570
T 20 25.2 47.3 22 95 2090
T 27 34.1 65.94 42 165.1 6934.2
T 32 39.6 73 61 165.1 10071.1
T 45 54.7 114.5 93 615.7 57260.1
Appendix II.indd 861 3/23/2009 7:37:41 PM

Standard Types for Transformers and Choke Laminations

Type No. Ac (mm2) Aw (mm2) Ap (mm4)
17 161 120 19320

12A 252 188.4 47476.8
21 252 332 83664
10 252 443 111636
10A 252 443 111636
1 278.9 654.3 182484.27
74 306.3 228 698364
23 363 272 98736
11 363 726 263538
11A 363 907 329241
2 363 1089 395307
30 400 300 120000
31 494 371 183274
45 494 371 183274
15 645 484 312180
44 645 484 312180
14 645 655 422475
4 645 1585 1022325
33 784 588 460992
3 1008 754 760032
13 1008 1411 1422288
4A 1112 1026 1140912
16 1450 1089 1579050
5 1450 1273 1845850
6 1450 1935 2805750
7 2580 1895 4889100
8 2580 5123 13217340
34 250 668 167000
9 440 910 400400
9A 490 790 387100
4AX 560 1307 731920
75 650 1526 991900
35A 1450 3934 5704300
8B 5810 6552 38067120
100 10320 11610 119815200
Appendix II.indd 862 3/23/2009 7:37:42 PM

Appendix III
Wire Size Table

Diameter with Area of bare
SWG enamel (mm) conductor (mm2) R/km @20oC W Weight (kg/km)
45 0.086 0.003973 4,340 0.0369

44 0.097 0.005189 3,323 0.0481
43 0.109 0.006567 2,626 0.061
42 0.119 0.008107 2,127 0.075
41 0.132 0.009810 1,758 0.0908
40 0.142 0.011675 1,477 0.1079
39 0.152 0.013700 1,258 0.1262
38 0.175 0.018240 945.2 0.1679
37 0.198 0.023430 735.9 0.2202
36 0.218 0.029270 589.1 0.2686
35 0.241 0.035750 482.2 0.3281
34 0.264 0.042890 402 0.3932
33 0.287 0.050670 340.3 0.465
32 0.307 0.059100 291.7 0.5408
31 0.33 0.06818 252.9 0.6245
30 0.351 0.07791 221.3 0.7121
29 0.384 0.09372 184 0.8559
28 0.417 0.11100 155.3 1.014
27 0.462 0.13630 126.5 1.245
26 0.505 0.16420 105 1.499
25 0.561 0.20270 85.1 1.851
24 0.612 0.24520 70.3 2.233
23 0.665 0.29190 59.1 2.655
22 0.77 0.39730 43.4 3.607
21 0.874 0.51890 33.2 4.702
20 0.978 0.65670 26.3 5.939
19 1.082 0.81070 21.3 7.324
Appendix III.indd 863 3/23/2009 7:37:56 PM

Diameter with Area of bare

SWG enamel (mm) conductor (mm2) R/km @20oC W Weight (kg/km)
18 1.293 1.16700 14.8 10.537

17 1.501 1.589 10.8 14.313
16 1.709 2.075 8.3 18.678
15 1.92 2.627 6.6 23.64
14 2.129 3.243 5.3 29.15
13 2.441 4.289 4 38.56
12 2.756 5.48 3.1 49.22
11 3.068 6.818 2.5 61
10 3.383 8.302 2.1 74
9 3.8 10.51 1.6 94
8 4.219 12.97 1.3 116
Appendix III.indd 864 3/23/2009 7:37:57 PM

Appendix IV
Laplace Transforms and z-Transforms of Few Time Domain Signals

Time function Laplace transform z-transform
δ (t ) 1 1
δ (t − kT ) e − kTs z −k
u(t) – unit step 1 z
s z −1
T 1 Tz
s2 ( z −1)2
t2 1 T 2 z ( z + 1)
2 s3 2( z − 1)3
e − at 1 z
s +a z − e − aT
te − at 1 Tze − aT
( s + a )2 ( z − e − aT )2
sin ωt ω z sin ωT
s 2 + ω2 z 2 − 2 z cos ωT + 1
cos ωt s z 2 − z cos ωT
s + ω2
2
z − 2 z cos ωT + 1
2
F(s + a)
e − at f (t ) f (e aT z )
Appendix IV.indd 865 3/23/2009 7:38:12 PM

Appendix V
Induction Motor Specifications
3 phase, 400 V rms, 50 Hz, 4 pole machine (p = 2)
Rs = 0.19 Ω
Rr = 0.125 Ω
M = 36.9 mH
Lss = 38.51 mH
Lrr = 37.56 mH
Nominal drive torque = 98 Nm
J = 0.1 Nms2/rad
B = 0.01 Nms/rad
Appendix V.indd 866 3/23/2009 8:24:59 PM

Appendix VI
Component Parameter Description

BJT Q
--------------------------------------------------------------------
SELECT THE BJT TYPE
1. Transistors,low frequency,bipolar
2. Transistors,low noise,high frequency,bipolar
3. Transistors,high power,high frequency,bipolar
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 1 : Transistors,low frequency,bipolar
Hp=BJT(type,PE_type,PQ_type,Tj,VCE,Vr,Pr,PA_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Env. factor : GB GF GM NS NU AIC AIF AUC AUF ARW SF MF ML CL
PQ_type: 1 2 3 4 5
PQ_type=Quality factor: JANTXV JANTX JAN LOWER PLASTIC
Tj=Junction temperature(C)
VCE=Operating collector to emitter voltage
Vr=Rated collector to emitter voltage,base open
Pr=Rated power in watt
PA_type= 1 2
PA=Application factor= linearamplifcation switching
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 2 : Transistors,low noise,high frequency,bipolar
Hp=BJT(type,PE_type,PQ_type,Tj,VCE,Vr,Pr)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4
PQ_type=Quality factor; JANTXV JANTX JAN LOWER
Vr=Rated collector to emitter voltage,base open
--------------------------------------------------------------------
Appendix VI.indd 867 3/23/2009 7:40:44 PM

--------------------------------------------------------------------
Type 3 : Transistors,high power,high frequency,bipolar
Hp=BJT(type,PE_type,PQ_type,Tj,VCE,Vr,PA_type,DF,P,F,PT_type,PM_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4
PQ_type=Quality factor: JANTXV JANTX JAN LOWER
Vr=Rated collector-emitter breakdown voltage with base shorted to
emitter(volts)
PA_type =1 For CW application
2 For pulsed application
DF=Duty factor=[1 5 10 15 20 25 30]
P=Output power(W)
F=Frequency(GHz)
PT_type=1 For gold metallization
2 For aluminum metallization
PM_type= 1 2 3
PM=Matching network factor=Input and output Input None
--------------------------------------------------------------------
Capacitor C
function Hp =capacitor(type,PE_type,PQ_type,T,S,C,Rc,Vc),
SYNTAX:
Hp =capacitor(type,PE_type,PQ_type,T,S,C,Rc,Vc)
--------------------------------------------------------------------
Capacitor Spec
type style MIL-C- Description
--------------------------------------------------------------------
1. CP 25 Fixed, paper DC (hermetic seal in
metal case)
2. CA 12889 Bypass, Radio interference rejection
paper, AC/DC (hermetic,metal case)
3. CZ,CZR 11693 Feed through, Radio interference
reduction, AC/DC (hermetic, metal
case) Est.R and non-Est. R
4. CQ,CQR 19978 Fixed, plastic/paper plastic
(hermetically sealed in metal/

Appendix VI 869
ceramic/glass cases) Est.R and

non-Est.R
5. CH 18312 Fixed, metallized (paper, paper
plastic plastic flm), DC
(hermetically sealed in
metal cases)
6. CHR 39022 Fixed, metallized (paper, paper
plastic plastic flm)
7. CFR 55514 Fixed, plastic (or metallized
plastic) DC in non-metal
cases
8. CRH 83421 Fixed, supermetallised plastic flm
(DC, AC or DC and AC)
hermetic sealed in metal
cases, Est. R
9. CM 5 Fixed, mica dielectric
10. CMR 39001 Fixed, mica dielectric, Est.R
11. CB 10950 Fixed, mica dielectric, button
style
12. CY 11272 Fixed, glass dielectric
13. CYR 23269 Fixed, glass dielectric, Est.R
14. CK 11015 Fixed, ceramic, general purpose
15. CKR 39014 Fixed, ceramic, general purpose,
Est.R
16. CC,CCR 20 Fixed, ceramic (temperature
compensating)Est.R and
non-Est.R
17. CDR 55681 Chip, Multiple layer, fxed,
ceramic dielectric, Est. R
18. CSR 39003 Fixed, electrolytic (solid
electrolyte), Tantalum,
Est.R
19. CWR 55365 Fixed, electrolytic (Tantalum)
chip, Est.R
20. CL 3965 Fixed, electrolytic (non-solid
electrolyte)Tantalum
21. CLR 39006 Fixed, electrolytic (non-solid
electrolyte)Tantalum, Est.R
22. CRL 83500 Fixed, electrolytic (non-solid
electrolyte)Tantalum cathode
23. CU,CUR 39018 Fixed, electrolytic (Aluminum oxide
Est.R and non-Est.R
24. CE 62 Fixed, electrolytic (DC, Aluminum,
Dry electrolyte, polarized)

25. CV 81 Variable, ceramic (Trimmer)

26. PC 14409 Variable, (piston type, tubular
trimmer)
27. CT 92 Variable, air dielectric(Trimmer)
28. CG 23183 Fixed or variable, vacuum dielectric
--------------------------------------------------------------------
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5 6 7 8 9
Quality factor: D C S,B R P M L MIL-C-xx,non-Est.R. Lower
T=Capacitor ambient temperature(C)
S=Ratio of operating voltage to rated voltage
C=Capacitance in micro farad
Rc=Effective resistance between capacitor and power supply for
Tantalum CSR style capacitors only
Vc=Voltage applied to capacitor for Tantalum CSR style capacitors
only
--------------------------------------------------------------------
Connections SC
function Hp=connections(type,PE_type,PQ_type,N1,N2,P),
--------------------------------------------------------------------
type:
1. Hand solder, w/o wrapping
2. Hand solder, w/ wrapping
3. Crimp
4. Weld
5. Solderless wrap
6. Clip termination
7. Refow solder
8. Spring contact
9. Terminal block
INTERCONNECTIONS
10. Printed wiring assembly/PCBs with PTHs
11. Discrete wiring with electroless deposited PTH (<=2 levels of
circuitry)
--------------------------------------------------------------------
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type : 1 2
Quality factor: MIL-SPEC Lower

Appendix VI 871
N1 : Number of PTHs by automated techniques (Quantity of wave

infrared (IR) or vapour phase soldered functional PTHs)
N2 : Number of hand soldered PTHs
P : Number of circuit planes or layers (2<= P <= 18)
--------------------------------------------------------------------
Connectors J
function Hp=connectors(type,PE_type,PQ_type,a1,a2,a3,a4)
--------------------------------------------------------
type: type of connector
1. Circular/Cylindrical
2. Card Edge (PCB)
3. Hexagonal
4. Rack and panel
5. Rectangular
6. RF Coaxial (low power)
7. RF Coaxial (high power)
8. Telephone
9. Power
10. Triaxial
IC SOCKETS
11. Dual-in-line package
12. Single-in-line package
13. Chip carrier
14. Pin Grid Array
15. Relay socket
16. Transistor socket
17. Electron tube, CRT socket
--------------------------------------------------------------------
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type : 1 2
Quality factor: MIL-SPEC Lower
G : Contact Gauge : 1. 32 Gauge contact
2. 30 Gauge contact
3. 28 Gauge contact
4. 24 Gauge contact
5. 22 Gauge contact
6. 20 Gauge contact
7. 18 Gauge contact
8. 16 Gauge contact
9. 12 Gauge contact

i : Amperes per contact

TA : connector ambient (C)
C=Mating/unmating cycles(per 1000 hours)
N=Number of active contacts (for IC socket connectors)
--------------------------------------------------------------------
Diode D
function Hp=diode(type,D_type,PE_type,PQ_type,Tj,a1,a2,a3),
--------------------------------------------------------------------
SELECT THE DIODE TYPE
1.Low frequency diode

2.High frequency(microwave,RF) diode
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 1: Low frequency
Hp=diode(type,D_type,PE_type,PQ_type,Tj,PC,Vo,Vr)
D_type
1 General purpose analog
2 Switching
3 Power rectifer,fast recovery
4 Power rectifer,schottky power diode
5 Power rectifer with high voltage stacks
6 Transient suppressor/varistor
7 Current regulator
8 Voltage regulator and voltage reference(avalanche and
zener)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5
Quality factor: JANTXV JANTX JAN LOWER PLASTIC
Tj=Operating junction temperture(C)
PC=1 For metallurgical bonded
=2 For nonmetallurgical bonded & spring loaded constants
Vo= Operating peak inverse voltage (PIV), volts
Vr= Rated peak inverse voltage (PIV), volts
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 2 : high frequency(microwave,RF)
Hp=diode(type,D_type,PE_type,PQ_type,Tj,PA_type,Pr)

Appendix VI 873
D_type
1 Si IMPATT(<=35GHz)
2 Gunn/bulk effect
3 Tunnel and back(including mixer,detectors)
4 Pin
5 Schottky barrier(including detectors)and point contact
(200MHz<=frequency<=35GHz)|
6 Varactor and step recovery
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5
Quality factor: JANTXV JANTX JAN LOWER PLASTIC
Tj=Operating junction temperature (C)
Application factor - PA_type: 1 For varactor, voltage control,
2 For varactor multiplier,
3 All other diode=1.0
Pr=Rated power, watts
--------------------------------------------------------------------
Electronic Filters EF
function Hp=electronicflters(PE_type,PQ_type,Hb_type),
--------------------------------------------------------------------
Hp=electronicflters(PE_type,PQ_type,Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ=Quality factor: PQ_type : 1 For SPEC
2 For lower
Hb=Base failure rate;
Hb_type =1 For MlL-F-l5733,Ceramic-Ferrite Construction
(Styles FL 10-16,22,24,30-32,34,35,38,41-43,45,47-50,
61-65,70, 81-93,95,96)
2 For MlL-F-l5733, Discrete LC Components, (Styles FL 37,
53, 74|
3 For MlL-F-l8327, Discrete LC Components (Composition 1)
4 For MIL-F-18327, Discrete LC and Crystal Components
1.9.7. Fuses-FU
function Hp=fuses(PE_type),
--------------------------------------------------------------------
Hp=fuses(PE_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Inductance L
function Hp=inductance(type,PE_type,PQ_type,THS),
----------------------------------------------------------- COILS:-
Hp=inductance(type,PE_type,PQ_type,THS)
type: 1. Fixed inductor or Choke
2. Variable inductor
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5 6
PQ_type=Quality factor: S R P M MIL-SPEC Lower
THS=Hot spot temperature(C)
Lamps LA
function Hp=lamps(PE_type,PA_type,PU,Vr),
--------------------------------------------------------------------
Hp=lamps(PE_type,PA_type,PU,Vr)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PA:Application factor:PA_type:1 For alternating current
:2 For direct current
PU=Utilization factor:Utilization(illuminate hrs/eqpt operate
hrs)
Vr=Rated voltage(volts)
Laser LAS
function Hp=Laser(type,PE_type,a1,a2,a3,a4,a5,a6,a7,a8,a9,a10,a11,a12)
--------------------------------------------------------------------
SELECT THE LASER TYPE
1 Helium and argon
2 Carbon dioxide,sealed
3 Carbon dioxide,fowing
4 Solid state,ND:YAG and ruby rod
--------------------------------------------------------------------
Type 1 : helium and argon
Hp=Laser(type,PE_type,HM_type,HC_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
HM=Lasing media failure rate:- HM_type: 1 2 3
type: He/Ne He/Cd Argon
HC=Coupling failure rate:- HC_type:1 For helium
2 For argon

Appendix VI 875
--------------------------------------------------------------------
Type 2 : carbon dioxide,sealed
Hp=Laser(type,PE_type,I,POS,OF,BVI)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
I=Tube current(mA) 10<=I<=150
POS=Optical surface factor:POS=1 For 1 active optical surface
=2 For 2 active optical surface
OF=CO2 overfll percent(0 to 50 )
BVI=Percent of ballast volumetric increase ( 0 to 200 )
--------------------------------------------------------------------
Type 3 : carbon dioxide,fowing
Hp=Laser(type,PE_type,POS,P)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
=2 For 2 active optical surfacev
P=Avg power output in KW. 0.01<=P<=1.0
--------------------------------------------------------------------
Type 4: solid state,ND:YAG and ruby rod
Hpump_type=Pump Pulse Failure Rate:-1 For xenon Flashlamps.
it is the failure rate contribution of the xenon fashlamp or
fashtube.
The fash lamps evaluated herein are linear types.
2 For krypton fashlamps.
it is the failure rate contribution Of the krypton fashlamp or
fashtube.
The fashlamps evaluated herein are the continuouswave(CW) type.they
are approximatety 7mm in diameter and 5 to 6 inches long.|
--------------------------------------------------------------------
Type=4:- Hpump_type : 1 for xenon Flashlamps
Hp=Laser(type,PE_type,POS,PC_type,Hpump_type,Pcool,L,HM_type,PPS1,
Ej,d,t,F,PPS2)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PC_type=Coupling cleanliness factor type:-
PC_type
:1 For Cleanliness level:regorous cleanliness procedures and
trained maintenance personnel.bellows provided over optical train

:2 For Cleanliness level:Mlnimal precautions during opening,

maintenance, repair, and testing.bellows provided over optical
train
maintenance, repair, and testing.no bellows provided over optical
train
Hpump_type=1
Pcool=Is the cooling factor due to various cooling media immediately
surrounding the fashlamp or fashtube,Pcool=1.0 for any air or inef gas
cooling,Pcool =0.1 for all liquid cooled designs.
Default value:Pcool=0.1,lquid cooled.
L=Is the fashlamp or fashtube arc length in inches.
Default value: L = 2.
HM_type:Media failure rate:-1 For ND:YAG
2 For ruby
PPS1=It is the repetition pulse rate in pulses per second.
Typical values range between 1 & 20 pulses per second.
Ej=It is the fashlamp or fashtube input energy per pulse, in joules.
Its value is determined from the actual or design input energy. For
values less than 30 joules.use Ej=30 Default value:Ej=40.
d=It is the fashlamp or fashtube inside diameter, in millimeters.
Default value: d = 4
t=It is the truncated pulse width in microseconds.
Use t -100 microseconds.for any truncatd pulse width exceeding 100
microseconds. For shotier duration pulses, pulse width is to be
measured at 10 percent of the maximum current amplitude.
Default value: t = 100
Only if HM_type=2 enter the value of F & PPS2.
F=Is the energy density in Joules per cm^2/pulse over the
cross-sectional area of the laser beam.which is nominally
equivalent to the cross-sectional area of the laser rod.
PPS2=Is the number of pulses per second.
--------------------------------------------------------
Type=4:- Hpump_type 2 :for krypton fashlamps
Hp=Laser(type,PE_type,POS,PC_type,Hpump_type,Pcool,L,HM_
type,P,F,PPS2)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PC_type=Coupling cleanliness factor type

Appendix VI 877
PC_type
:1 For Cleanliness level:regorous cleanliness procedures and
trained maintenance personnel.bellows provided over optical
train
maintenance, repair, and testing.bellows provided over optical
train
maintenance, repair, and testing.no bellows provided over
optical train
Hpump_type=2
Pcool=Is the cooling factor due to various cooling media
immediately surrounding the fashlamp or fashtube.
Pcool=1 for any air or inert gas cooling.
Pcool=0.1 for all liquid designs.
Defautt value:- Pcool=0.1,liquidcooled.
L=Is the fashlamp or fashtube arc length in inches.
Defautt value: L =2
HM_type:Media failure rate:-1 For ND:YAG
2 For ruby
P=Is the average inputpower in Kilowatts.Default
value: P =4
only if HM_type=2 Enter the value of F & PPS2.
F=Is the energy density in Joules per cm^2/pulse over the
cross-sectional area of the laser beam.which is
nominally equivalent to the cross-sectional area of the
laser rod.
PPS2=Is the number of pulses per second.
Meters Panel MEP

function Hp=meterspanel(PE_type,PQ_type,PF_type,PA_type),
-------------------------------------------------------------
Hp=meterspanel(PE_type,PQ_type,PF_type,PA_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ=Quality factor:PQ_type : 1 For MIL-M-10304
: 2 For lower
PF=Function factor:PF_type : 1 For ammeter
: 2 For voltmeter
: 3 Other
PA:Application factor:PA_type:1 For direct current
2 For alternating current

Microcircuits: Bubble Memories

function Hp=microcircuits_bubblememories(PE_type,PQ_type,y,pkg,Np,
Tc,NC,N1,N2,D1,D2,RW)
-------------------------- -----------------------------------------
Hp=microcircuits_bubblememories(PE_type,PQ_type,y,pkg,Np,Tc,NC,
N1,N2,D1,D2,RW)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:
PQ_type= 1 For class S categories:a.procured in full accordance
with MIL-M-38510.class S requirements.
2 For class B categories: a.procured in full
accordance with MIL-M-38510.class B requirements.
3 For class B-1 category:-procured to a in
production MIL drawing. DESC drawing or the
government approved documentation.
4 Other commercial or unknown screening levels
y=Number of years the device has been in production
pkg :Package type:1 For hermetic:dips w/solder or weld seal,pin
gnd array(PGA)1.SMT(leaded and nonleaded)
2 For DIPs with glass
3 For fatpacks with axial leads on 50 Mil
centers
4 For cans
5 For nonhermetic DIPs,PGA,SMT(Leaded and
nonleaded)
Np=Number of functional pins
Tc=Case temperature(C)
NC=Number of bubble chips per packaged device
N1=Number of dissipative elements on a
chip(gates,detectors,generators..etc)N1<=1000
N2=Number of bits,N2<=9*10^6
D1=Avg device data rate
D2=Mfg. max. rated data rate D=(D1/D2) <=1
RW=No. of reads per write
Microcircuits Digital
function Hp=microcircuits_digital_ic(PE_type,PQ_type,y,pkg,Np,
family,N,Tj),
--------------------------------------------------------------------
Hp=microcircuits_digital_ic(PE_type,PQ_type,y,pkg,Np,family,
N,Tj)

Appendix VI 879
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:
2 For class B categories:a.procured in full accordance
with MIL-M-38510.class B requirements.
3 For class B-1 category:-procured to a in production
MIL drawing. DESC drawing or dher government
approved documentation.
4 Other commercial or unknown screening levels|
centers
4 For cans
nonleaded)
family : 1 For TTL,STTL,CML,HTTL,FTTL,DTL,FCL,ASTTL
2 For BiCMOS,LSTTL,LTTL,ALSTTL
3 For III,ISL
4 For digital MOS,CMOS
N=Number of gates in the IC
Tj=Device junction temperature (calculate using Ta + Rthja*P)
(C)
Microcircuits GaAs
function Hp=microcircuits_GaAs(PE_type,PQ_type,pkg,y,Np,DT,C,Tj),
--------------------------------------------------------------------
Hp=microcircuits_GaAs(PE_type,PQ_type,pkg,y,Np,DT,C,Tj)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:
with MIL-M-38510.class S requirements
2 For class B categories: a.procured in full accordance
with MIL-M-38510.class B requirements
MIL drawing. DESC drawing or the government approved
documentation.


centers
4 For cans
nonleaded)
DT : Device type :- MMIC devices:-
1 For low noise & low power(<=100mW)
2 For driver & high power(>100mW)
3 For unknown
Digital devices:-
4 For Digital devices
C:Complexity(No. of elements)
(C)
Microcircuits Linear
function Hp=microcircuits_linear_ic(PE_type,PQ_type,pkg,y,Np,N,Tj),
--------------------------------------------------------------------
Hp=microcircuits_linear_ic(PE_type,PQ_type,pkg,y,Np,N,Tj)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:
documentation.
centers
4 For cans
nonleaded)

Appendix VI 881

N=Number of transistors in the IC
Tj=Device junction temperature (calculate using Ta +
Rthja*P)(C)
Microcircuits Logic Array

function Hp=microcircuits_logic_array(PE_type,PQ_type,pkg,y,Np,
family,N,Tj),
--------------------------------------------------------------------
Hp=microcircuts_logic_array(PE_type,PQ_type,pkg,y,Np,family,
N, Tj)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:
documentation.
centers
4 For cans
nonleaded)
family : Bipolar PLA/PAL
1 For TTL,STTL,CML,HTTL,FTTL,DTL,FCL,ASTTL
3 For III,ISL, MOS, PLA/PAL
4 MOS, CMOS
N=Number of gates in the IC
Rthja*P)(C)

Microcircuits Memories
function Hp=microcircuits_memories(PE_type,PQ_type,pkg,y,Np,family,
B,MT,C,ECC,Tj),
--------------------------------------------------------------------
Hp=microcircuits_memories(PE_type,PQ_type,pkg,y,Np,family,B,MT,C,
ECC,Tj)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:
documentation.
centers
4 For cans
nonleaded)
family : MOS:-
1 For ROM
2 For PROM,UVEROM,EEPROM & EAPROM
3 For DRAM
4 For SRAM(MOS & BiCMOS)
BIPOLAR:-
5 for ROM & PROM
6 for SRAM
B=Memory size,B(bits)= 1 For up to 16K
2 For 16K<B<=64K
3 For 64K<B<=256K
4 For 256K<B<=1M
MT=Memory device type:-1 For fotox
2 For textured poly

Appendix VI 883
3 For other devices(except fotox and

textured poly EEPROMS)
C=Total no of programming cycles over EEPROM life.C= 1 to
500000.
ECC:Error correction code options:-1 For no on-chip ECC
2 For on-chip hamming code
3 For two-needs-one
redundant cell approach
Rthja*P)(C)
Microcircuits uPs
function Hp=microcircuits_Microprocessor(PE_type,PQ_type,pkg,y,Np,
family,N,Tj),
--------------------------------------------------------------------
Hp=microcircuits_Microprocessor(PE_type,PQ_type,pkg,y,Np,
family,N,Tj)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:
documentation.
centers
4 For cans
nonleaded)
family : 1 For TTL,STTL,CML,HTTL,FTTL,DTL,FCL,ASTTL
3 For III,ISL
4 For digital MOS,CMOS

N=Number of Bits(up to 32 bits)

(C)
Microcircuits SAW
function Hp=microcircuits_SAW(PE_type,PQ_type),
--------------------------------------------------------------------
Hp=microcircuits_SAW(PE_type,PQ_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:screening level:-
PQ_type=1 For 10 temp cycles(-55C to 125C)with end point
electrical tests at temp extremes
2 For none beyond best commercial practices
Microcircuits VLSI
function Hp=microcircuits_VLSI(PE_type,PQ_type,NP,pkg,CFT,Tj,pt,mpt,
VTH,XS,DA1),
--------------------------------------------------------------------
VHSIC/VHSIC like and VLSI CMOS
Hp=microcircuits_VLSI(PE_type,PQ_type,NP,pkg,CFT,Tj,pt,mpt,VTH,
XS,DA1)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Quality class:
with MIL-M-38510.class B requirements.
MIL drawing. DESC drawing or dher government approved
documentation.
NP=Number of package pins
pkg:-Package type: 1 For DIP
2 For pin grid array
3 For chip carrier(surface mount
technology)
CFT:-Package type correction factor type:1 For hermetic
2 For nonhermetic
Rthja*P)(C)

Appendix VI 885
pt:-Part type:1 For logic and custom

2 For gate array and memory
mpt:-Manufacturing process type:
1 For QML(qualifed manufacturer list) or QPL(qualifed product
list)
2 For non QML or non QPL
VTH=ESD susceptibility(volts)
XS=Feature size(microns)(.80 to 1.25)
DA1:Total Scribed Chip Die Area in cm2
(die area conversion cm^2=MIL^2+155000)
Opto-electronics OPT
function Hp=optoelectronics(type,PE_type,PQ_type,Tj,Hb_type,a1,
a2,a3,a4,a5),
--------------------------------------------------------------------
Type 1 : Detectors,isolators & emitters
Hp=optoelectonics(type,PE_type,PQ_type,Tj,Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5
PQ_type=Quality factor; JANTXV JANTX JAN LOWER PLASTIC
Hb_type:-Base failure rate for optoelectonics
Hb_type Optoelectonics types
Photodetectors
1 Photo transistor
2 Photo diode
Opto-isolators
3 Photo diode output,single device
4 Phototransistor output ,single
device
5 Photodarlington output,single
device
6 Light sensitive resistor,single
device
7 Photodiode output,dual device
8 Phototransistor output,dual
device
9 Photodarlington output,dual
device
10 Light sensitive resistor,dual
device Emitters

11 Infrared light emitting

diode(IRLD)
12 Light emitting diode(LED)
--------------------------------------------------------------------
Type 2 : Alphanumeric display
Hp=optoelectonics(type,PE_type,PQ_type,Tj,Hb_type,C,Hic)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5
PQ_type=Quality factor; JANTXV JANTX JAN LOWER PLASTIC
Hb_type= 1. For segment displays
2. For diode array displays
C=Number of characters= 1 to 15
Hic=1 For displays with logic chip
=2 For displays without logic chip
--------------------------------------------------------------------
Type 3 : Laser diode
Hp=optoelectonics(type,PE_type,PQ_type,Tj,Hb_type,PA_type,I,
Pr,Ps,DC)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type=Quality factor; 1=Hermetic package
2=Nonhemertic with facet coating
3=Nonhemertic without facet coating
Hb=Base failure rate; Hb_type = 1 For GaAs/AI GaAs
2 For In GaAs/In GaAsP
PA_type=Application factor :-1 for CW & 2 for pulsed.
I=Forward peak current(Amps),<=25
Pr=Required optical power output(mW) 0<Pr/Ps<=0.95
Ps=Rated opticalpower output(mW)
DC=Duty cycle 0.1 to 1
Potentiometer POT
function Hp=potentiometer(type,PE_type,PQ_type,a1,a2,a3,a4,a5,a6,a7,
a8),
--------------------------------------------------------------------
Type 1 : Thermistor (RTH)
Hp=potentiometer(type,PE_type,PQ_type,Hb_type,R)

Appendix VI 887
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
PQ_type=Quality factor: MIL-SPEC Lower
Hb=Base failure rate: Hb_type:- 1 for Bead, 2 for Disk & 3 for
Rod.
R=Resistance in ohm
--------------------------------------------------------------------
Type : 2 Variable, Wirewound (RTR, RT)
Hp=potentiometer(type,PE_type,PQ_type,R,S,T,Ntaps,R1,Pa,VR)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5 6
PQ_type=Quality factor: S R P M MIL-R-27208 Lower
R=Resistance in ohm
S=Ratio of operating power to rated power
T=Ambient temperature(C)
Ntaps=Number of potentiometer taps including the wiper and
terminations
R1= Nominal Total Potentiometer Resistance
Pa = Power Dissipation
VR(rated)=40 Volts for RT 26 and 27
=90 Volts for RTR 12, 22 and 24; RT 12 and 22
--------------------------------------------------------------------
Type 3 : Variable, Wirewound, Precision (RR)
Hp=potentiometer(type,PE_type,PQ_type,R,S,T,Ntaps,R1,Pa,VR,
PC_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
R=Resistance in ohm
terminations

VR(rated)=250 Votts for RR0900, RRI 100,RR1300, RR2000,

RR3000,RR31OO, RR3200, RR3300,RR3400, RR3500|
VR(rated)= 423 Volts for RR3600, RR3700
PC=Construction class factor: construction class type:
PC_type=1 RR0900A2A9J103
2 3
3 4
4 5
--------------------------------------------------------------------
Type 4 : Variable, Wirewound, Semiprecision (RA, RK)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
R=Resistance in ohm
terminations
VR(rated)=50 Votts for RA1o
=75 Volts for RA20X-XC, F
=130 Volts for RA30X-XC, F
=175 Volts for RA20X-XA
=275 Volts for RK09
=320 Volts for RA30X-XA
--------------------------------------------------------------------
Type 5 : Variable, Wirewound, Power (RP)
PC_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
R=Resistance in ohm

Appendix VI 889

terminations
Pa(applied) = Power Dissipation
VR(rated)=250 Volts for RP06,RP10
=500 Volts for others
PC=Construction class factor:
construction class type:PC_type=1 Enclosed
2 Unenclosed
---------------------------------- Type 6 : Variable, Nonwirewound
(RJ, RJR)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5 6
PQ_type=Quality factor: S R P M MIL-R-22097 Lower
R=Resistance in ohm
terminations
VR(rated)=200 Volts for RJ and RJR26;RJ and RJR50
=300 Volts for All Others
--------------------------------------------------------------------
Type 7 : Variable, Composition (RV)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
R=Resistance in ohm
terminations


VR(rated)= 500 Volts for RV4X-XA & XB
= 500 Volts for 2RV7X-XA & XB
= 350 Volts for RV2X-XA & XB
= 250 Volts for RV1 X-XA & XB
= 200 Volts for All Other Types
--------------------------------------------------------------------
Type 8 : Variable, Nonwirewound, Film & Precision (RQ, RVC)
Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
R=Resistance in ohm
terminations
VR(rated)= 250 Votts for RQ090, 110, 150, 200,300
= 500 Vofts for RQ100,160,210
= 350 Volts for RVC5,6
Hb=Base failure rate:
Hb_type:1 For MIL-R-39023 (RQ)Variable, Nonwirewound, Film,
Precision
2 For MIL-R-23285 (RVC) Variable, Nonwirewound, Film
Quartzcrystals QZ
function Hp=Quartzcrystals(PE_type,PQ_type,F),
--------------------------------------------------------------------
Hp=Quartzcrystals(PE_type,PQ_type,F)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ=Quality factor:PQ_type : 1 For MIL-SPEC
: 2 For lower
F=Frequency(MHZ)

Appendix VI 891
Resistor R
function Hp=resistor(type,PE_type,PQ_type,T,S,Pd),
SYNTAX:
Hp=resistor(type,PE_type,PQ_type,T,S,Pd)
--------------------------------------------------------------------
type Resistor Spec Description
style MIL-R-
--------------------------------------------------------------------
1. RC 11 Fixed, composition(insulated)
2. RCR 39008 Fixed, composition(insulated)
Est.R
3. RL 22684 Fixed, flm, insulated
4. RLR 39017 Fixed, flm, insulated, Est. R
5. RN(R,C or N) 55182 Fixed, flm, Est. R
6. RM 55342 Fixed, flm, Chip, Est. R
7. RN 10509 Fixed, flm (high stability)
8. RD 11804 Fixed, flm (Power type)
9. RZ 83401 Fixed, flm, resistor networks
10. RB 93 Fixed, wirewound (accurate)
11. RBR 39005 Fixed, wirewound (accurate)
Est.R
12. RW 26 Fixed, wirewound (power)
13. RWR 39005 Fixed, wirewound(power)
Est. R
14. RE 18546 Fixed, wirewound(power,
chassis mnt)
15. RER 39009 Fixed, wirewound( power,
chassis mnt) Est. R
16. RTH 23648 Thermistor, insulated
17. RT 27208 Variable, wirewound (Lead
screw activated)
18. RTR 39015 Variable, wirewound (Lead
screw activated) Est.R
19. RR 12934 Variable, wirewound,
precision
20. RA 19 Variable, wirewound (low
operating temp.)
21. RK 39002 Variable, wirewound,
semi-precision
22. RP 22 Wirewound, power type
23. RJ 22097 Variable, non-wirewound
24. RJR 39035 Variable, non-wirewound
Est. R
25. RV 94 Variable, composition

26. RQ 39023 Variable, non-wirewound

precision
27. RVC 23285 Variable, non-wirewound
--------------------------------------------------------------------
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5 6
Quality factor: S R P M MIL-R-xx Lower
T=Resistor case temperature or ambient component temperature(C)
Pd=Operating power dissipation in resistor
Relays RY
function Hp=relays(type,PE_type,PQ_type,Hb_type,TA,S,PC_type,
PL_type,CYRenv,CYR,CT),
--------------------------------------------------------------------
Type 1 : mechanical
Hp=relays(type,PE_type,PQ_type,Hb_type,TA,S,PC_type,PL_type,
CYRenv,CYR,CT)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5 6 7
8
PQ_type=Quality factor: R P X U M L Non-est.rel
Commercial
Hb=Base failure rate: Hb_type:1 for T=85 C max rated
:2 for T=125 C max rated
TA=Ambient temperature (C)
S=Operating load current/rated resistive load current
PC_type: Contact form factor :1 for SPST
:2 for DPST
:3 for SPDT
:4 for 3PST
:5 for 4PST
:6 for DPDT
:7 for 3PDT
:8 for 4PDT
:9 for 6PDT
PL_type :Load stress factor for :1 for resistive
:2 for inductive
:3 for lamp

Appendix VI 893
CYRenv= quality type for cycle rate :1 for MIL-SPEC

:2 for lower quality
CYR=Cycle rate (cycles per hour)
CT: construction type – choose from the list given below
___________________________________________________________________
| | Construction type |
| Contact rating | Application type | CT
|_________________|__________________________|_____________________
| | | 1 Armature(long)
| | | 2 Dry reed
| | | 3 Mercury wetted
| Signal current | Dry circuit | 4 Magnetic latching
| (low mV and mA) | | 5 Balanced armature
| | | 6 Solenoid
|_________________|__________________________|_____________________
| | | 7 Armature(long)
| | General | 8 Balanced armature
| | Purpose | 9 Solenoid
| |__________________________|_____________________
| | | 10 Armature(long &
short)
| | Sensitive | 12 Magnetic latching
| | (0 - 100 mV) | 13 Meter movement
| | | 14 Balanced armature
| 0 - 5 Amp |__________________________|_____________________
| | | 15 Armature(short)
| | Polarized | 16 Meter movement
| |__________________________|_____________________
| | | 17 Dry reed
| | Vibrating reed | 18 Mercury wetted
| |__________________________|_____________________
| | | 19 Armature(balanced
&short
| | High speed | 20 Dry reed
| |__________________________|_____________________
| | Thermal time delay | 21 Bimetal
| |__________________________|_____________________
| | Electronic time | 22
| | delay(non thermal) |
| |__________________________|_____________________
| | | 23 Dry reed
| | Latching magnetic | 24 Mercury wetted
| | | 25 Balanced arma-
ture

|_________________|__________________________|_____________________
| | | 26 Vaccum (glass)
| | High voltage | 27 Vaccum (ceramic)
| |__________________________|_____________________
| 5 - 20 Amp | | 28 Armature(long &
short)
| | Medium power | 30 Magnetic latching
| | | 31 Mechanical latch-
ing
| | | 32 Balanced armature
| | | 33 Solenoid
|_________________|__________________________|_____________________
| | | 34 Armature( short)
| | | 35 Mechanical
latching
| 25 - 600 Amp | Contactors | 36 Balanced armature
| | (high current) | 37 Solenoid
|_________________|__________________________|_____________________
--------------------------------------------------------------------
Type 2 : solid state and time delay

Hp=relays(type,PE_type,PQ_type,Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
Hb=Base failure rate:Hb_type:-1 For solid state
2 For solid state time delay
3 For hybrid
Rotating Device ROT

function Hp=rotatingdevice(type,a1,a2,a3,a4,a5),
--------------------------------------------------------------------
Type 1 : Motors
Hp=rotatingdevice(type,T_type,M_type,T,LC,H)
T_type : 1. ambient temperature(C)
2. cycled temperature
M_type : 1. Electrical (general)
2. Sensor
3. Servo
4. Stepper

Appendix VI 895
T : for T_type=1 ->ambient temperature (deg.C)

for T_type=2 ->[temp1, temp2, temp3,....]
LC : system design life cycle in hours or
the motor preventive maintenance interval in hours
H : for T_type = 2 ONLY
[time @ temp1, time @ temp2, time @ temp3, ....]
--------------------------------------------------------------------
Type 2 : Synchros and resolvers
Hp=rotatingdevice(type,PE_type,PN,DT,SIZE,TF)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PN=Number of brushes 2, 3, 4
DT=Device type: 1 For synchro
: 2 For resolver
SIZE:1 For size 8 or smaller
:2 For size 10-16
:3 For size 18 or larger
TF=Frame temperature(C)
(if TF is unknown assume TF=40 C+ambient temperature)
--------------------------------------------------------------------
Type 3 : Elapsed time meters
Hp=rotatingdevice(type,PE_type,PT_type,Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PT=Temperature stress factor:- Operating T(C)/rated T(C)
PT_type
0 to 0.5 1
0.6 2
0.8 3
1.0 4
Hb=Base failure rate: type Hb_type
A.C. 1
Inverter driven 2
Commutator D.C. 3
Switches SW
function Hp=switches(type,PE_type,PQ_type,PC_type,PL_type,S),
---------------------------------------------------------------
type: switch type
1. Centrifugal
2. Dual in line package

3.
Limit
4.
Liquid level
5.
Microwave (waveguide)
6.
Pressure
7.
Pushbutton
8.
Reed
9.
Rocker
10.
Rotary
11.
Sensitive
12.
Thermal
13.
Thumbwheel
14.
Toggle
15.
Circuit breaker - (not used as power on/off switch)
(magnetic, thermal and thermal-magnetic switches)
16. Circuit breaker - (also used as power on/off switch)
(magnetic, thermal and thermal-magnetic switches)
--------------------------------------------------------------------
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
PC_type :Contact confguration factor :1 for SPST
:2 for DPST
:3 for SPDT
:4 for 3PST
:5 for 4PST
:6 for DPDT
:7 for 3PDT
:8 for 4PDT
:9 for 6PDT
PL_type :Load stress factor :1 For resistive load
:2 For inductive load
:3 For lamp load
S=Operating load current/rated resistive load current
Transformer XFM
function Hp=transformer(type,PE_type,PQ_type,THS)
--------------------------------------------------------------------
Transformers
Hp=transformer(type,PE_type,PQ_type,THS)
type: 1. Flyback (<20 volts)
2. Audio (15-20KHz)

Appendix VI 897
3. Low power pulse (Peak pwr < 300W and avg. pwr <5W)
4. High power, high power pulse (peak power >=300W
and avg.pwr >=5W)
5. RF(10K - 10M Hz)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2
THS=Hot spot temperature(C)
Transistor FET-M
function Hp=transistor(type,PE_type,PQ_type,Tj,a1,a2,a3,a4,a5,a6,a7,
a8),
--------------------------------------------------------------------
Type 1 : Transistors,low frequency,bipolar
Hp=transistors(type,PE_type,PQ_type,Tj,Pr,VCE,VCEO,PA_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5
VCE= Collector to emitter voltage
VCEO=Collector to emitter voltage,base open
PA=Application factor=>PA_type=>1 for linear amplifcation
2 for switching
--------------------------------------------------------------------
Type : 2 Transistors,low frequency,SI FET
Hp=transistors(type,PE_type,PQ_type,Tj,Pr,Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5
Hb_type=1 For MOSFET
2 For JFET

--------------------------------------------------------------------
Type 3 : Transistors,unijunction
Hp=transistors(type,PE_type,PQ_type,Tj)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4 5
--------------------------------------------------------------------
Type 4 : Transistors,low noise,high frequency,bipolar
Hp=transistors(type,PE_type,PQ_type,Tj,Pr,VCE,VCEO)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4
VCEO=Collector to emitter voltage,base open
--------------------------------------------------------------------
Type 5 : Transistors,high power,high frequency,bipolar
Hp=transistors(type,PE_type,PQ_type,Tj,PA_type,VCE,P,F,PT_type,
BVCES,… PM_type,DF)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4
PA_type =1 For CW application (DF=duty factor not required)
2 For pulsed application (DF=duty factor required)
P=Output power(W)
F=Frequency(GHz)
PT_type=1 For gold metallization
2 For aluminum metallization
BVCES=Collector-emitter breakdown voltage with base shorted to
emitter(volts)
PM_type=> 1 2 3
PM=Matching network factor=> Input and output Input None

Appendix VI 899
Only if PA_type=2 enter DF.

DF=Duty factor=[1 5 10 15 20 25 30]
--------------------------------------------------------------------
Type 6 : Transistors,high frequency,GaAs FET
Hp=transistors(type,PE_type,PQ_type,Tj,PA_type,P,F,PM_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4
PA_type=Application factor: 1 For all low power and pulsed=1
2 For CW=4
P=Average output power(Watts)
F=Frequency(GHz)
PM_type=Matching network factor=>1 For input and output
2 For input
3 For none
--------------------------------------------------------------------
Type 7 : Transistors,high frequency,SI FET
Hp=transistors(type,PE_type,PQ_type,Tj,Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12 13 14
PQ_type: 1 2 3 4
Hb_type= 1 2
Hb=Base failure rate= MOSFET JFET
Tubes TUB
function Hp=tubes(type,PE_type,a1,a2,a3,a4,a5),
--------------------------------------------------------------------
SELECT THE TUBE TYPE
1. Tubes,all types except twt and magnetron
2. Tubes,traveling wave
3. Tubes,magnetron
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 1: Tubes, all types except twt and magnetron
Hb=Base failure rate:Hb_type:1 For pulsed Klystrons
2 For any other pulsed Klystrons

3 For CW
4 For any other CW
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 1:- Hb_type 1 : Pulsed Klystrons
Hp=tubes(type,PE_type,T,Hb_type,PK_type)
PE_type: 1 2 3 4 5 6 7 8 9 10
11 12 13 14 |
PE=Environment factor : GB GF GM NS NU AIC AIF AUC AUF ARW
SF MF ML CL |
T=Number of years since introduction to feld use
Hb_type=1
----------------------------------------------------------------
| PK_type | | Tube type
|---------|-----------------------------|-----------------------
| 1 | |Triode,tetrode,pentode
| 2 | Reciever |Power rectifer
| 3 | |CRT
| 4 | |Thyratron
|---------|-----------------------------|-----------------------
| 5 | Crossed feld amplifer |QK681
| 6 | |SFD261
|---------|-----------------------------|-----------------------
| 7 | |2041
| 8 | Pulsed gridder |6952
| 9 | |7835
|---------|-----------------------------|-----------------------
| 10 | |Triode,peak
pwr<=200KW,Avg
| | |Pwr<=2KW,freq<=200MHz
| 11 | Transmitting |Tetrode & pentode,peak
pwr
| | |<=200KW,Avg pwr<=2KW,
Freq<=200MHz| |
| 12 | |If any of the above
limit exceeded | |
----------------------------------------|-----------------------
| 13 | | Antimony
trisulfde(Sb2S3)
| | Vidicon | Photo conductive
material
| 14 | | Silicon diode array
| | | Photo conductive
material

Appendix VI 901
|---------|-----------------------------|-----------------------
| 15 | | VA144
| 16 | | VA145E
| 17 | Twystron | VA145H
| 18 | | VA913A
|---------|-----------------------------|-----------------------
| 19 | | 4KMP10000LF
| 20 | | 8568
| 21 | | L3035
| 22 | | L3250
| 23 | Klystron pulsed | -
| 24 | | SAC42A
| 25 | | VA842
| 26 | | Z5010A
| 27 | | ZM3038A
----------------------------------------------------------------
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 1 :- Hb_type 2 : For any other pulsed Klystrons
Hp=tubes(type,PE_type,T,Hb_type,P,F)
PE_type: 1 2 3 4 5 6 7 8 9 10
11 12 13 14 |
SF MF ML CL |
Hb_type=2
P=Peak output power in MW 0.01<=P<=25 and p<=490F^-2.95
F=Operating frequency in GHz 0.2<=F<=6
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 1:- Hb_type 3 : CW klystron
Hp=tubes(type,PE_type,T,Hb_type,PK_type)
PE_type: 1 2 3 4 5 6 7 8 9 10
11 12 13 14 |
SF MF ML CL |
Hb_type=3

-----------------------------------------------
| PK_type | Tube type
|---------|-----------------------------------|
| 1 | Klystron low power |
|----------------------------------------------
| | Klystron continuous wave |
| 2 | 3K3000LQ
| 3 | 3K50000LF
| 4 | 3K210000LQ
| 5 | 3KM300LA
| 6 | 3KM3000LA
| 7 | 3KM50000PA
| 8 | 3KM50000PA1
| 9 | 3KM50000PA2
| 10 | 4K3CC
| 11 | 4K3SK
| 12 | 4K50000LQ
| 13 | 4KM50LB
| 14 | 4KM50LC
| 15 | 4KM50SJ
| 16 | 4KM50SK
| 17 | 4KM3000LR
| 18 | 4KM50000LQ
| 19 | 4KM50000LR
| 20 | 4KM170000LA
| 21 | 8824
| 22 | 8825
| 23 | 8826
| 24 | VA800E
| 25 | VA853
| 26 | VA856B
| 27 | VA888E
-----------------------------------------------
-------------------------------------------------------------------’
--------------------------------------------------------------------
Type 1 :- Hb_type 4: For any other CW Klystrons
Hp=tubes(type,PE_type,T,Hb_type,P,F)
PE_type: 1 2 3 4 5 6 7 8 9 10
11 12 13 14 |
SF MF ML CL |
Hb_type=4

Appendix VI 903
P=Avg output power in KW 0.1<=P<=100 and p<=8.0*(10^6)*(F^-1.7)

F=Operating frequency in MHz 300<=F<=8000
-------------------------------------------------------------------’
--------------------------------------------------------------------
Type 2 : Tubes,traveling wave
Hp=tubes(type,PE_type,P,F)
PE_type: 1 2 3 4 5 6 7 8 9 10
11 12 13 14 |
SF MF ML CL |
P=Rated power in watts(peak,if pulsed) 0.001<=P<=40000
-------------------------------------------------------------------’
--------------------------------------------------------------------
Type 3 : Tubes,magnetron
Magnetron type :M_type: 1 For pulsed magnetron |
2 For CW magnetron(rated power<5KW) |
-------------------------------------------------------------------’
--------------------------------------------------------------------
Type 3 :- magnetron type :M_type: 1 for pulsed magnetron
Hp=tubes(type,PE_type,M_type,PC_type,R,P,F)
PE_type: 1 2 3 4 5 6 7 8 9 10
11 12 13 14 |
SF MF ML CL |
M_type=1
PC=Construction factor :PC_type 1. CW(rated power<5KW)
2. Coaxial pulsed
3. Conventional pulsed
R=Radiate hours/flament hours(0.0 to 1.0)
P=Output power in MW 0.01<=P<=5
-------------------------------------------------------------------’
--------------------------------------------------------------------
Type 3 : Magnetron type :M_type: 2 for CW magnetron(rated
power<5KW)
Hp=tubes(type,PE_type,M_type,PC_type,R)

PE_type: 1 2 3 4 5 6 7 8 9 10
11 12 13 14 |
SF MF ML CL |
M_type=2
PC=Construction factor :PC_type 1. CW(rated power<5KW)
2. Coaxial pulsed
3. Conventional pulsed
R=Radiate hours/flament hours(0.0 to 1.0)
Thyristors SCR
function Hp=thyristors(PE_type,PQ_type,Tj,VBo,VBr,Ifrms),
--------------------------------------------------------------------
Thyristors,SCRs & triacs
Hp=thyristors(PE_type,PQ_type,Tj,VBo,VBr,Ifrms)
PE_type: 1 2 3 4 5 6 7 8 9 10
11 12 13 14
Env. factor : GB GF GM NS NU AIC AIF AUC AUF ARW
SF MF ML CL
PQ_type: 1 2 3 4 5
VBo=Blocking voltage applied
VBr=Blocking voltage rated
Ifrms=RMS rated forward current(Amps)
UJT-UJT
function Hp=ujt(PE_type,PQ_type,Tj),
--------------------------------------------------------------------
Transistors,unijunction
Hp=ujt(PE_type,PQ_type,Tj)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12
13 14
Env. factor : GB GF GM NS NU AIC AIF AUC AUF ARW SF MF
ML CL
PQ_type: 1 2 3 4 5

Appendix VI 905
Variable Capacitor VC
function Hp=variablecap(type,PE_type,PQ_type,T,S,a1,a2),
--------------------------------------------------------------------
Type 1 : Variable ceramic(CV)
Hp=variablecap(type,PE_type,PQ_type,T,S,Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12
13 14
ML CL
PQ_type: 1 2
Hb=Base failure rate: Hb_type:1 For T=85 C max rated((MIL-C-81
Styles CV 11,14,21,31,
32,34,40,41)
:2 For T=125 C max rated(MIL-C-81
Styles CV 35, 36)
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 2 : Variable piston type(PC)
Hp=variablecap(type,PE_type,PQ_type,T,S,Hb_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12
13 14
ML CL
Hb=Base failure rate: Hb_type:1 For T=125 C max rated
(MIL-C-14409 Styles G, H, J,
L, T)
:2 For T=150 C max rated(MIL-C-
14409 Characteristic Q)
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 3 : Variable, Air trimmer
Hp=variablecap(type,PE_type,PQ_type,T,S)

PE_type: 21 3 4 5 6 7 8 9 10 11 12
13
14
Env. factor : GB
GF GM NS NU AIC AIF AUC AUF ARW SF MF
ML
CL
PQ_type: 1 2
--------------------------------------------------------------------
--------------------------------------------------------------------
Type 4 : Variable and fxed, gas or vaccum(CG)
Hp=variablecap(type,PE_type,PQ_type,T,S,Hb_type,PCF_type)
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12
13 14
ML CL
PQ_type: 1 2
Hb=Base failure rate: Hb_type:1 For T=85 C max rated (Styles CG
20,21,30,31,32,40-44, 51,60-
64,67)
:2 For T=100 C max rated (Styles
CG 65, 66)
:3 For T=125 C max rated (Style CG
50)
PCF=Confguration factor:PCF_type:-1 For fxed
2 For variable
HYBRID Micro-HYB
function Hp=hybridmicro(CKT,PE_type,class,y)
--------------------------------------------------------------------
Hp=hybridmicro(CKT,PE_type,class,Y)
CKT:Circuit type:- 1 For digital
2 For video,10MHz<1<1GHz
3 For microwave,f>1 GHz
4 For linear,f<10 MHz
5 For power

Appendix VI 907
PE_type: 1 2 3 4 5 6 7 8 9 10 11 12
13 14
ML CL
Class:-
1 For class S categories:-
a.procured in full accordance with MIL-M-38510.class S
requirements.
b.procured in full accordance with MIL-I-38535 and appendix
B thereto(class U).
c.hybrids:(procured to class S requirements(quality level K)
of MIL-H-38534).
2 For class B categories:-
a.procured in full accordance with MIL-M-38510.class B
requirements.
b.procured in full accordance with MIL-I-38535.(class Q)
c.hybrids:procured to class B requirements(quality level
H)of MIL-H-38534).
3 For class B-1 category:-procured to a in production MIL
drawing. DESC drawing or dher government approved
documentation. |
4 For TM1010 and TM2001 and TM5004 and TM1014 and TM2009
5 For TM1010 or TM2001,TM5004 and TM1014 and TM2009
6 For pre-bum in electricals TM1015 and TM5004 B level
7 For pre-bum in electricals TM1015 and TM5004 S level
8 For TM2020 particle impact noise detection
9 For TM5004(or 5008 for hybrids)
10 For TM2010/17
11 For TM1014
12 For TM2012
13 For TM2009
14 For TM5007/5013(GaAs)
15 For TM2023
y:years generic device type has been in production

Bibliography
1. Anderson, R.T. (March 1976) Reliability Design Handbook, Rome Air Development Center, NY, USA.
2. Billinton, R. and Allan, R. (1983) Reliability Evaluation and Design, Plenum Press, NY, USA.
3. Bose, B.K. (February 1982) Adjustable Speed AC Drives – A Technology Status Review. Proc. IEEE,
70 (2), 116–135.
4. Colonel McLyman, T.W. (1978) Transformer and Inductor Design Handbook, Marcel Dekker, Inc.,
NY, USA.
5. Crandall, S.H., Karnopp, D.C., Kurtz, E.F. (1968) Dynamics of Mechanical and Electromechanical Sys-
tems, McGraw-Hill Book Company, NY, USA.
6. Demidovich, B.P. and Maron, I.A. (1976) Computational Mathematics, Mir Publishers, Moscow,
Russia.
7. Doyle, J.C., Stein, G. (February 1981) Multivariable Feedback Design: Concepts for a Classical/
Modern Synthesis. IEEE Trans. Automat. Control, AC-26(1), 4–16.
8. Finney, D. (1980) The Power Thyristor and its Applications, McGraw-Hill Book Company, London, UK.
9. Franklin, G.F., Powell, J.D. and Workman, M.L. (1990) Digital Control of Dynamic Systems, 2nd edn,
Addison-Wesley, MA, USA.
10. Friedland, B. (1987) Control System Design – An Introduction to State Space Methods, McGraw-Hill
Book Company, Singapore.
11. Fuqua, N.B. (1987) Reliability Engineering for Electronic Design, Marcel Dekker, Inc., NY, USA.
12. Gottlieb, I.M. (1985) Power Supplies – Switching Regulators, Inverters and Converters, BPB Publica-
tions, New Delhi, India.
13. Johnson, H.W. and Graham, M. (1993) High Speed Digital Design – A Handbook of Black Magic,
Prentice Hall, NJ, USA.
14. Kitsum, K. (1984) Switched Mode Power Conversion – Basic Theory and Design, Marcel Dekker, Inc.,
NY, USA.
15. Klion, J. (1992) Practical Electronic Reliability Engineering, Van Nostrand Reinhold, NY, USA.
16. Kraus, J.D. (1984) Electromagnetics, 3rd edn, McGraw-Hill Book Company, NY, USA.
17. Krause, P.C. (1987) Analysis of Electric Machinery, McGraw-Hill Book Company, NY, USA.
18. Lander, C.W. (1981) Power Electronics, McGraw-Hill Book Company, NY, USA.
19. Leonhard, W. (1985) Control of Electrical Drives, Springer Verlag, Berlin, Germany.
20. Lewis, F.L. (1992) Applied Optimal Control and Estimation, Prentice Hall, NJ, USA.
21. Meyer, P.L. (1970) Introductory Probability and Statistical Applications, Oxford and IBH Publishing
Company Pvt. Ltd., New Delhi, India.
22. Military handbook: MIL-HDBK 217F (Notice 2).
23. Miller, I. and Freund, J.E. (1985) Probability and Statistics for Engineers, Prentice Hall of India,
New Delhi, India.
24. Mohan, N., Undeland, T.M. and Robbins, W.P. (1989) Power Electronics: Converters, Applications and
Design, John Wiley & Sons, NY, USA.
25. Mottershead, A. (1984) Electronic Devices and Circuits – An Introduction, Eastern Economy edn, Prentice
Hall of India, New Delhi, India.
PE_Bibliography.indd 909 3/23/2009 3:53:36 PM

910 Bibliography
26. Murphy, J.M.D. and Turnbull, F.G. (1988) Power Electronic Control of AC Motors, Pergamon Press,
NY, USA.
27. Nadkarni, M.A. and Bhat, S.R. (1985) Pulse Transformer: Design and Fabrication, CEDT series,
Tata McGraw-Hill Book Company, New Delhi, India.
28. Ogata, K. (1991) Modern Control Engineering, Prentice Hall of India, New Delhi, India.
29. Ohno, E. (1988) Introduction to Power Electronics, Oxford Science Publications, Oxford, UK.
30. Pearman, R.A. (1980) Power Electronics – Solid State Motor Control, Reston Publishing Co. Inc., VA, USA.
31. Press, W.H., Flannery, B.P., Teukolsky, S.A., Vetterling, W.T. (1986) Numerical Recipes, Cambridge
University Press, NY, USA.
32. Ramamoorthy, M. (1991) Thyristor and their Applications, Affiliated East-West Press Pvt. Ltd.,
New Delhi, India.
33. Rashid, M.H. (1988) Power Electronics, Circuits, Devices and Applications, Prentice Hall, NJ, USA.
34. Shepherd, W., Hulley, L.N. and Liang, D.T.W. (1995) Power Electronics and Motor Control, Cambridge
University Press, NY, USA.
35. Shinners, S.M. (1964) Control System Design, John Wiley & Sons, NY, USA.
36. Strang, G. (1986) Introduction to Applied Mathematics, Wellesley-Cambridge Press, MA, USA.
37. Tarter, R.E.(1985) Principles of Solid State Power Conversion, Howard W. Sams and Co., Inc., IN, USA.
38. Thorborg, K. (1988) Power Electronics, Prentice Hall, NJ, USA.
39. Umanand, L. and Bhat, S.R. (1992) Design of Magnetic Components for Switched Mode Power Converters,
Wiley Eastern Limited, New Delhi, India.
40. Unitrode seminar manual (1988) Switching Regulated Power Supply Design, Unitrode Corporation,
NH, USA.
41. Vas, P. (1990) Vector Control of AC Machines, Clarendon Press, Oxford, UK.
42. Vithayathil, J. (1995) Power Electronics – Principles and Applications, McGraw-Hill Book Company,
NY, USA.
43. Wildi, T. (1981), Electrical Power Technology, John Wiley & Sons, NY, USA.
44. William, B.W. (1992) Power Electronics: Design, Drives, Applications, 2nd edn, McGraw-Hill Book
Company, NY, USA.
PE_Bibliography.indd 910 3/23/2009 3:53:36 PM

Index
A Bode’s theorem, 551, 552 three-level arm, 293

Absorptance, 755–756 Boltzmann constant, 4, 754 two-arm, 296
Ackerman’s formula, 592, 633 Boltzmann law, 4, 5 Buck-boost converter, 213, 214,
AC voltage, 95 Bond, energy flux in a, 454 217
Algorithm, Bond graph, 453–473 operation of, 214
alternate algorithm, 710 augmented, 462 schematic of a, 219
arithmetic, 707 construction, 458–462 steady state analysis of, 217
iterative, 709 drawbacks, 453 topology, 213
Kalman filter, 660 elements of, 457 waveforms of various signals
least squares, 660, 663 energy, 453 of, 215
multiplication, 707–709 half bond, 454 with non-idealities, 219
square root iteration, 712 no switch element in, 467 with power semiconductor
Aliasing, 521 of an SPDT switch, 470 switches, 216
Alpha parameter, 12 power, 453, 454
Ampere’s law, 369, 370, 373 Boundary limit C
Amp-second balance, 200 circular, 727–728 Capacitance
Analog controller, 499 hexagonal, 727, 728–729 drain-gate, 79
Anode, 6 British Telecom, 837 filter, 110
short, 44 Buck converter, 201, 202, 205, Capacitor
Aluminum 450, 471–472 current waveform, 206
anodized, 754 inductor current ripple of a, charging time constant, 69
polished, 754 222 energy of, 69, 112
rough surface, 754 operation of, 201 flux walking, 248, 249, 253
Area waveforms of various signals of input filter, 109
cross-section, 378–379 a, 203 selection, 114
product, 377, 379, 391 with non-idealities, 207 size, 212
window, 377, 378 with power semiconductor tantalum, 114
AT&T, 837 switches, 204 value of, 6, 206, 218
ZCS applied to, 267–271 Capacitor current curve, 218
B ZVS applied to, 271–275 negative portion of a, 218
Baker’s clamp, 63 Boost converter, 206, 211, 212, Causality assignment, 462–463
Bang-bang technique, 317 222, 472–473 sequence, 462
BELL, 837 consisting of an SPDT switch, limitations to selection of, 457
Belt-conveyor system, 806, 816 206 rules for selection of, 457, 462
BJTs (bipolar junction transis- efficiency of, 612 Cathode, 6
tors), 1 operation of a, 208 CCM (continuous conduction
Black box, 425 output value, 206 mode), 221, 222, 472
Bode diagram, 552 topology of, 206 Characteristic dimension, 743
constraints in, 556 waveforms of various signals of Charge balance, 200
controller design with, a, 209 Chopper, 194, 290
556–561, 567 with non-idealities, 213 applications of, 196
method, 550, 551–561 Bridge DC–DC converters using, 200
open-loop, 555 n-level, 295 dual-four quadrant, 199
shaping of, 557 three-level arm, 283 output voltage of, 196
PE_Index.indd 911 4/6/2009 6:17:11 PM

912 Index
output waveform, 192 major issues with, 549 push-pull, 226, 238–243, 393
primary side, 230 performance criteria, 549 phase-controlled, 95
schematic of, 194–195 robust, 679–687 single-phase full-wave con-
Chopper, dual state space, 550 trolled, 95
circuit schematic, 200 steps, 682–685 single-phase half-wave con-
using two SPDT switches, 199 techniques, 550–551 trolled, 95
Ciruit-averaging method, 447 transfer function based, 550 soft switching of, 265–275
Circuits Control systems step down, 201, 205
anti-saturation, 63 basics, 535–540 step-up, 206, 211, 216, 217
full-wave half-controlled, 138 design of, 649 three-phase full-wave full-con-
MOSFET drive, 70 dynamics and augmentation, trolled, 95
opto-isolated, 73 638–641 three-phase full-wave half-con-
S/H (sample and hold), 576 essential tools of, 501 trolled, 95
single-phase, 96 performance parameters of, three-phase half-wave con-
single-phase rectifier, 96 535 trolled, 95
snubber, 80, 243 SSS performance parameters unidirectional power flow, 613
three-phase, 10–109 of, 535 unity power factor, 613
three-phase capacitor-filter rec- stable, 535 utility power factor, 610–613
tifier, 115 unstable, 535 variants, 201
three-phase rectifier, 96 Convection, 736 ZCS, 266
transformer-isolated gate drive, forced, 742, 744–753 ZVS, 266
75 free, 742, 744–747 Copper, polished, 754
transistors drive, 55 heat transfer by, 742–752 Core
voltage scaling, 149 Converter circuits, problems with C-shaped, 404
CMOS (complementary metal implementing, 258 E-shaped, 404, 405
oxide semiconductor), 72 Converters I-shaped, 405
hex inverting buffer, 72 AC–DC, 628 loss, 412
logic integrated circuit, 72 basic, 200 planar, 405–406
non-inverting hex, 72 bidirectional power flow, 613 pot, 406
Conduction, 737 boost, 472–473 rod, 404
Conductors, buck, 201, 202, 222, 450 shapes, 404–406
air gaps between, 403 buck-boost, 213, 214 toroidal, 406
metallic, 6 current-controlled, 607–610 U-shaped, 404
Connectors, 822–828 Cuck, 259–263 Cost
Controllability, 582–583 DC–AC, 289 analysis, 770
Control principles, 540–543 DC–DC, 193 optimization, 770
Control index, 348, 357 DC–DC switched mode, 193 CRGO (cold rolled grain ori-
Controller flyback, 225, 253–257, 601 ented) cores, 373, 379
digital current, 634 forward, 226–228 Crosstalk, 823–825
optimal, 670 front-end, 613–619 between directly overlapping
PI, 717–719 full-bridge, 26, 248–253, 393 loops, 825
speed and current, 630 half-bridge, 226, 243–248 estimation of, 823
transfer function, 564, 565 high boost, 259, 263–265 problems, 824–825
Controller design input-output relationship of, Cuck converter, 259–263
classical techniques, 550 423 application of Kirchhoff’s volt-
guidelines for, 575–576 isolated, 225–257 age law to, 261, 262
PE_Index.indd 912 4/6/2009 6:17:12 PM

Index 913
isolated, 262–263 DC–AC converter, 289 doping, 3

non-isolated, 263 classification of, 289 forward-biased, 2
schematic of, 259 DC–DC converters, 200 freewheeling, 97
two operating mode, 259 area products for, 393–394 model, 10–11
Current practically realizable, 201 opto-coupler, 601
controlling, 607–619 using chopper, 200 opto-isolator, 602
drain, 79 using inductor, 200 parameters, 8
eddy, 410 using capacitor, 200 p-n junction, 3
emitter, 165 DC–DC linear regulators, 149 reverse-biased, 3
holding, 39 DCM (discontinuous conduction selection, 113
line side, 101, 103 mode), 221, 222 turn-OFF of, 7
load, 108, 109, 201 DC motor, 196, 465 turn-ON of, 8
magnetizing, 232, 400 armature-controlled, 434, 465 Zener, 36, 151, 163, 174, 180,
primary, 101, 103 constant field, 435 183, 570
protection, 164 examples of, 465–467 Disturbance propagation,
regulation, 163, 180 excited by armature supply, 609
reverse saturation, 6 466–467 Drive circuits, 55
secondary winding, 101, 103 first quadrant operation of a, base drive, 58
stator , 484 196 characteristics and classification
surge, 9 four quadrant operation of a, of, 58
tail, 37 197 examples of, 59–70
transformer, 64, 398–403 operating zones of a, 196 transistor, 55
turn-ON, 117–120, 128 second quadrant operation of Driven inverter, 304–308
Current, stator , 484 a, 196 Duty cycle, 65, 77, 312
direct axis, 481 separately excited, 465–466 Duty ratio, 65, 66
equation, 482 third quadrant operation of a, operating, 213
quadrature axis, 481 196 Dynamic compensator, 632
Current regulator, analysis of, DC series motor, 467 Dynamics and augmentation
180–184 DC servomotor, 434, 435 anti-alias filter, 636–638
Current, rotor DC steady state principles, 200 controller, 638–641
direct axis, 482 DC voltage, 95 plant, 636
equation, 482 Decoupling network, 630–631
quadrature, 482 Density E
Current transformer, 64, flow, 372 Early effect, 13
398–403 flux, 372–373, 407 Eber–Moll model, 12
design of, 401–403 Digital controller, 499 Eddy current loss, 410
Cylindrical solid advantages of, 499 Electric domain, 371
horizontal, 746 design of, 500 Electric stress rating, 221
vertical, 746–747 Digital integrator, 711 Electrolyte, aluminum, 114
Digital processor, 695 Electromagnetic torque
D Digit product, 708, 709 equation, 490
DAC (digital-to-analog conver- Diode, 2, 174 of an induction motor, 490
sion), 500 average current rating of a, 232 Elements
D’Alember’s principle, 428 body, 23 capacity or C–, 454
Darlington connection, 21 classification, 8 resistance or R–, 454
DC bus, 109, 110 diffusion current, 4 inertia or L–, 454
PE_Index.indd 913 4/6/2009 6:17:12 PM

914 Index
EMI (electromagnetic interfer- Exponent, 699 Fixed-point system,

ence), 111, 825–828 Exponential, 715–716 primary operations in, 703
due to large current loop, 827 problem in, 703
filter, 117 F tracking the base point, 703
Emittances for common radiation Failure, Flow, 372
surfaces, 754 mean time between, 776–778 density, 372
Enamel paint, 754 mean time to, 770, 776–777, Flux, 227, 228, 232, 374
Energy 795–796 change, 384
equation, 655 Failure mode density, 372–373, 374
flux, 454 dormant, 792 estimator, 630
storage, 375–376 immediate, 792 estimators for indirect field-
-storage device, 369 of switching network, 792 oriented control, 628
-transfer device, 369 Failure rate, 770, 771–773, 776 in a motor, 622
Environments, 842 allocation, 816 magnetic, 473
Equation basic, 837 rate, 386
algebraic, 463 curve, 774 state model, 622
electromagnetic torque, general model for, 774 swing, 411
489–490 periods in, 774–775 walking, 243, 248, 249
state, 463–467 variation in, 774 waveform, 230
Equilibrium hole density, 4 Failure time Flux density, 372–373, 374, 407
Error, residual, 715 probability density function, saturation, 373
Error constant, 537 773–776 Flux linkage, 484, 489
acceleration, 537 reliability density function, air-gap, 489, 490, 620
positional, 537 771 generalized, 490
velocity, 537 Faraday’s law, 369, 370, 375, 378, leakage, 386
Estimator, 591 384, 410, 473 rotor, 489, 490, 622
current, 594 Fault reset, 807 stator, 489, 490
design, 591–598 Feed rate set, 807 Flux state model
dynamics, 591 Field-oriented control, 623, 628 air-gap, 485, 486
equation of an, 591, 597 flux estimators for indirect, generalized, 485, 486–489
error dynamics, 592–593 628 rotor, 485
full-order state, 597 Field weakening region, 620 stator, 485, 486
gain, 591 Filer Flux walking capacitor, 248, 249,
Kalman, 670 capacitor input, 110 253
optimal, 664 common digital, 512 capacitance of, 253
prediction, 591 digital, 508–515 rms value of the current
reduced-order, 596–598 dual input range rectifier, 115 through a, 253
state, 591 finite impulse response, 508, Flyback converter, 253–257
Euler’s backward rectangular rule, 509 development of a, 254
527–528 first-order low-pass, 553 schematic of, 255
block schematic of, 527 infinite impulse response, 508, waveforms of various signals
illustration of, 527 510 of, 255
Euler’s forward rectangular rule, Kalman, 595 Forward converter
525, 526 rectifier-LC, 124–128 drawbacks of, 233
block schematic of, 526 types of, 515 dual switch, 236–238
illustration of, 525 First-level modeling, 166 operation of a, 233
PE_Index.indd 914 4/6/2009 6:17:12 PM

Index 915
topology, 232 Gauge of wire, 380, 383 Hysteresis, 373

waveforms, 237 Glass, 754 band, 317
with demagnetizing winding, Gummel –Poon model, 13 loop, 373, 390
233–236 Generator, 799 loss, 373, 388, 407
Forward generation, 199 Generation mode, 196
mode, 198, 199 forward, 196 I
Forward motoring mode, 198 reverse, 196 IGBT (insulated gate bipolar tran-
Fourier series, 124, 125, 319, sistor), 1, 30, 194, 212, 232
324, 356 H dynamic switching characteris-
a general waveform, 319 Half-bridge configuration, tics of an, 36
for line-to-line voltage wave- 306–307 remarks on, 37
form, 325 Half-bridge converter, 243–248 structure of an, 34
of a rectified waveform, 124, circuit, 245, 247 Inductance,
125 equation for, 246 leakage, 231, 232
of a waveform with quarter operation of, 243 magnetizing, 400
wave summary, 320 schematics of, 243 total series, 117
Four-wire connection, 162–163 Harmonic index, 318, 323, 325 Induction motor, 619–631, 636,
Fractional part, 715 Harmonic elimination, 350–358 806
Frame ways to implement, 354 closed loop V/f control,
arbitrary reference, 478, 480 with multiple notches, 351– 621–622
stator reference, 478, 479 352 control of, 619–631
French mathematician, 442 Heat sink, 740, 741 control with output feedback,
Frequency domain performance mounting of power devices on, 633–641
specifications, 681 760–761 direct field-oriented control of,
Frequency ratio, 355 size, 761 625–627
Full-bridge circuit, 249 volume, 761 example, 674–679
equation, 251 Heat transfer indirect field-oriented control
implementation, 251 best method for, 762–763 of, 627–631
topology, 250 by conduction, 738–742 open loop V/f control of, 621
waveforms of, 252 by convection, 742 scalar control, 619622
Full-bridge configuration, 308 by mass transport, 756 speed of an, 807
Full-state feedback, 584–587, by radiation, 753–756 steady-state model of an,
649 constants, 739 619–620
Function, general notes on, 758–763 strategies for, 677–679
cosine, 714, 715 mechanism, 735 Inductor, 369, 375
exponential, 715–716 mixed, 758–759 calculation of, 205
logarithm, 716–717 modeling of, 736–738 current, 429
sine, 714, 715 process, 735 design of, 379
steady-state, 736 multiple-winding, 382–383
G transient model for, 761–763 value, 205, 375, 379
Gain margin, 555 Hold Insulation factor, 404
Gate turn-OFF thyristor, 44 equivalence, 521, 530–533 Integer part, 715
structure of a, 45 time, 116 Integrated circuits, 158
symbol of, 45 Hold equivalence, 530–533 classification of, 158
turn-OFF characteristics of a, zero-order, 530 example of, 158
45 Hypotenuse, 713714 Inverse polar plot method, 550
PE_Index.indd 915 4/6/2009 6:17:12 PM

916 Index
Inversion layer, 23 Kinetic energy of a pendulum, efficiency of, 165

Inverter 443 operating principle,
classification of, 290 Kirchhoff’s voltage law, 65, 216, 150–151
composition of, 290 230, 261, 447, 654 parameters of, 166
constraints in voltage genera- application of, 434 performance quality, 165
tion by an, 319 topologies, 151–166
double-level, 291 L Line regulation, 182
driven, 304–308 Lagrange, 442 coefficient, 172, 182
features of storable core, 297– Lagrange’s method, 442–447 range of, 182
298 Lagrange’s equation, 442, 444, Litz wire, 413
multi-level, 292, 349–350 447 LMS algorithm, 660–664
practical, 296 Lagrange’s function, 442 Load
PWM (pulse-width modu- Lagrangian, current, 108, 109
lated), 315 partial derivatives of the, 444, delta, 313
quasi-square wave, 321, 322 447 regulation, 182
saturable base drive, 302–304 with respect to generalized resistance, 173
self-driven, 297 coordinates, 444, 447 star, 313
simple, 290 Laplace transformation, 431, 446, three-phase, 315
single-level, 290 530 voltage, 108
six-step, 314–315 Latent heat, 757 Load-regulation coefficient,
three-phase, 312–318 Law 172, 173
two-level, 629 Ampere’s, 369, 370, 373 Loci
topologies, 290–297 effort, 471 of the roots of the characteris-
Iron, rough surface, 754 Faraday’s, 369, 370, 374, 375, tic equation, 575
Isolated converters, 225–257 378, 384, 385, 410, 473 root, 575
benefits of, 225 flow, 471 Logarithm, 716–717
buck-based, 225 Lead-lag network, 554–556 Lorentz force, 473
buck-boost based, 225 Leakage inductance, 231
categories of, 225 issue of, 258 M
classification of, 226 problems, 238 Magnetic amplifier, 604
schematic of, 226 solutions to, 258 Magnetic components, 369
Least squares Magnetic domain, 371
J principle, 650 power dissipated in, 372
Jacobian matrix inverse, 355 recursive , 659–668 Magnetic field
Jury test, 549 solution, 655–658 energy, 413
Junction temperature, 735 weighted, 658–659 intensity, 388
Linear differential equation, first- rotating, 474
K order, 445 Magnetic loss, 407–412
Kalman estimator, 670 Linear quadratic Gaussian, 672 Magnetic retentivity, 407
Kalman filter, 595, 664–668 Linear quadratic regulator, 543, Magnetics,
algorithm, 660 670–672 general notes on, 403–415
implementation filter, 668 Linear quadratic tracker, theory of, 300
measurement update in, 667 672–674 Mantissa, 699
recursion steps in the, 665 Linear regulators, 149, 150, 158, Mason’s rule, 440–442
steady-state, 667 162, 163 MATLAB, 532, 534, 549, 551,
time update in, 667 classification of IC, 158 555, 575
PE_Index.indd 916 4/6/2009 6:17:12 PM

Index 917
Mean time steady state, 427, 448, 452 Nichols chart method, 550
between failure, 776–777 Modeling of systems n-Material, 3
to failure, 770, 776–777, d-q model of, 483–490 Normalized form, 699
795–796 modeling approaches, 424 Notch, 331
Measurement update, 595, 665 physical, 424 width, 356
Method system identification, 424 Notch angles, 331, 351, 355
adaptive controller, Modeling uncertainties, 681–682 determination of, 352–355,
bode diagram, 550 Modulation 357
coupled-inductor, 602 center-pulse, 327–329 n-Region, 4
full state feedback, 550 control index, 318, 322, 325 Negative rail, 156
full state feedback with estima- end-pulse, 324–326 regulator, 158
tor, 550 index, 331, 334, 357 terminal, 163
inverse polar plot, 550 over-, 334, 727–729 Normalization and scaling factor,
linear quadratic, 550 pulse-width, 318–350 704, 705
linear regulator, 602 quasi-square, 321 NPN transistor, 11
magnetic amplifier, 603 sinusoidal pulse-width, 329 equivalent circuit of an, 13
Nichols chart, 550 space vector, 339–350 operation of, 12
non-isolated converter, 603 MOSFET, 193 Number
Nyquist diagram, 550 drive circuits, 70–80 Nusselt, 743, 744, 745
optimal controller, 550 paralleling of, 29 Rayleigh, 743, 744
optimal controller with opti- precautions for, 29 Reynolds, 743, 744
mal estimator, 551 schematic characteristics of an Numerical integration method,
output feedback, 550 n-channel, 22 521, 523–530
robust controller, 551 simulated model for a, 31 Numeric format, 695–702
root locus, 550, 573–581 symbol of an n-channel, 22 classes of a, 696
self-tuning, 551 symbol of a p-channel, 22 features of a, 696
state space, 582–584 Motor, induction fixed-point, 696
Microcontroller, 695, 707 assumptions for, 479 floating-point, 699–702
Microprocessor, 695 electromagnetic torque of an, general notes on, 702
MIL-217, 837 490 problems with a, 696
Miler turn-ON, 79 modeling of, 479–483 Nusselt number, 743, 744,
MIMO (multiple-input multiple- voltage equation, 479 745
output), 433, 542 voltage supplied to the stator Nyquist diagram method,
Minimum energy principle, of, 479 550
653–655 Motoring mode, 196
mmf (magneto-motive force), forward, 196, 198 O
369, 371, 375 reverse, 196 Observability, 583–584
Model Multiplicand, 707, 708 Observer, 591
average large-signal, 448, 449, Multiplier, 707 Ohmic drop, 7
452 Mutual inductance, effect, op-amp (operational amplifier),
Eber–Moll, 12 823–825 154
equilibrium, 427 Operating duty ratio,
Gummel–Poon, 13 N 213
large-signal, 427, 447, 451 Nature’s minimum energy prin- Opto-coupler, 65, 600
linearized small-signal, 466 ciple, 413 Output feedback, 649
small-signal, 427, 449, 452 Newton’s law of motion, 442 Overlap angle, 140–142
PE_Index.indd 917 4/6/2009 6:17:12 PM

918 Index
P output of a, 636 Power system

Parabola, 660 state, 636 modeling-switched, 467–473
Paraboloid, 652 Plate Protection circuitry, 163
Part derating, 828 flat, 748 Proximity effect, 413–415, 414
Pendulum horizontal flat, 744–745 due to equal current, 414
equation of motion of an parallel flat, 747 due to unequal current, 414,
inverted, 444 rectangular, 745–746 415
inverted, 446 p-material, 3 Push-pull circuit, 239
kinetic energy of a, 443 Poles, 436, 536 Push-pull configuration, 304–
linearized dynamic model of a, Pole voltage, 312 306
445 Pole-zero map, 521, 522–523 Push-pull converters, 226, 238–
second-order differential equa- Polynomial, 714 243
tion, 445 Positive rail, 156 circuit schematic of, 239, 242
under equilibrium condition, regulator, 158 problem with, 243
445 Power schematic, 239
Performance index, 675 active, 121 waveforms for, 241
constraint, 674 circular, 745 PWM (Pulse-width modulation ),
Performance parameters reactive, 121 318–350, 500, 601, 630,
measures for, 538–540 Power circuits, single-phase, 720–724
SSS, 539 129–134 bipolar, 330, 331
Period Power converter, 194 classification of, 319
discrete, 774 Power electronics, sinusoidal, 332–334
infant mortality, 775 applications, 1 objectives of, 318
useful life, 775 integral part of, 369 space vector, 724–727
wear out, 775–776 system, 717 staircase, 355–357
Permeability, 376, 386 Power MOSFET, 22 through phase modulation,
infinite, 388 dynamic characteristics of, 25 337
Permeance, 375, 379, 382 model for a, 30 unipolar, 330
dependence of, 380 power dissipated in a, 30 waveform, 721
infinite, 388 static characteristics of, 25 with harmonic elimination,
Perturbation, 609 Power semiconductor switches, 1, 350–358
pf (power factor), 55, 121–124 265, 290, 297 without harmonic elimination,
Phase margin, 555 circuit viewpoint of, 1 319
Physics of failure, 853–854 drive viewpoint of, 2 Pynter, Prof. H.M., 453
approach for, 852 packaging viewpoint of, 2 Pythagoras theorem, 714
sections of, 852 physics viewpoint of, 1
PID (proportional integral deriv- protection viewpoint of, 1 Q
ative) controller, 562–573 Power switches, 55 Quadratic forms, 651–653
design, 567–568 electrical stress rating of, 221 practical significance of, 653
integrator windup in, 568 high, 55 types of, 653
parameters, 567 device of, 219 Quadratic function, 652
practical, 570 macro models of, 467 cases in, 652
Plant modulated bond graph ele- coefficient of, 652
dynamics, 636 ment in, 467 two-dimensional, 652
input to a, 636 selection of, 219–221 Quadratic performance index,
thermal stress rating, 221 673
PE_Index.indd 918 4/6/2009 6:17:12 PM

Index 919
Quasi-square wave, 309 Regulators, 585 power dissipation in, 232

Fourier series representation 78xx, 158 total series, 117
for, 310 79xx, 158 zero, 388
inverter, 309–312 efficiency of, 150 Resistor,
output quantity of, 309, 311 fixed, 158 bleeder, 127
regulation based on duty cycle IC linear, 158–160 burden, 399
control, 311 linear, 149, 150, 154 power rating of, 163
linear quadratic, 670–672 Ripple output, 127
R negative, 157, 158 Robust controller design, 679–
Radiation, 737 operating principle of, 150 687
Rail positive voltage, 158 Root
negative, 156, 157 series, 149, 154–156, 157 loci, 575
positive, 156, 157 shunt, 149, 151–154 locus method, 550, 573–581
Rayleigh number, 743, 744 three-pin, 158 Rotating reference frame, 619
Raynolds number, 743, 744 variable, 158 Rotor magnetizing current, equiv-
Rectifier Zener shunt, 151 alent, 628
circuits, 96 Regulator types, 158 Routh–Hurwitz criterion, 549
classification of, 95 fixed, 158 Rule
components, 112 variable, 158 Euler’s backward rectangular,
controlled, 95, 128–142 Relative error, 697, 699, 700 527–528
design of capacitor input filter, Relative density, 697, 699 Euler’s forward rectangular,
110–116 uniform, 700 525, 526
full-wave bridge, 96, 101–103 Reliability Tustin trapezoidal, 528–530
full-wave center-tapped, 96, component, 805
99–101 concepts, 771–777 S
half-wave, 96, 97–99 definition of, 771 Sample time, 635
polyphase, 127 design for, 804–813, 822–828 Sampling
silicon-controlled, 38 determination of, 771 effect of, 518–521
un-controlled, 95, 96 effect of circuit design on, 805 Nyquist, 520
Rectifier circuit, 115 equipment, 777 Shannon, 520
three-phase capacitor-filter, environmental effect on, 805 time, 501
115 factors affecting, 805 Saturable-core inverters,
three-phase controlled, 134 inter-connection wiring in, 298–302
Rectifier filter, dual input range, 805 behavior of, 300
115 mathematical representation designing of, 300
Recursive least squares, 659 of, 771 features of storable core,
drawbacks in, 659 modeling for, 777–788 297–298
Redundant systems, 779, network model, 777 Saturable base drive inverter,
787–799 prediction of, 770, 835 302–304
approach in, 799 prediction process, 842–843 SciLAB, 532, 534, 549, 551, 555,
switching in, 791–795 requirements, 770, 814 575
with non-identical component, target specifications of, 805 Second-level modeling, 174,
799–804 Reliability prediction, 770, 835 183–184
Reflectance, 755–756 uses of, 885 Self-driven inverters,
Regulation loop, 163 depends on, 835 297–304
Regulator parameters, 172 Resistance, Semiconductor devices, 1
PE_Index.indd 919 4/6/2009 6:17:12 PM

920 Index
Semiconductor switches, 204, Source State equation

210, 216, 228 single-phase sinusoidal, 127 discretizing, 521, 533–534
Input-output relationship of, three-phase, 127 extraction, 458, 463–467
204 Space factor, 403 State space form, 484
steady state analysis of, 204 Space vector, 340, 473, 475 State space method, 542
with ideality constraints, 204 amplitude of a, 345 Steady state
Sensitivity flux linkage, 489 accuracy, 536–538, 549
co-, 670 general, 476 error, 538
complementary, 680 in a coplanar vector space, 343 Kalman filter, 667–668
Series inductance effect, 825– in orthogonal coordinates, 476 Step-down converter, 205
828 in polar coordinates, 477 Step-up converter, 206, 208, 211,
Series regulator modeling, 473–490 214, 222
analysis of a, 166–180 PWM, 724–727 inductor current ripple of a,
equivalent circuit of, 168 resultant, 475 222
linear, 167 resultant stator current, 475 inductor current waveform of
Shunt regulator, rotation of, 344 a, 222
modified, 154, 165 transformation, 476–478 schematic of a, 219
power rating of, 163 Space vector modulation, 339– waveforms of various signals of
Signal 350 a, 209, 215
modulating, 329 features of, 339 with non-idealities, 213, 219
switching, 329 principle, 341, 344 with power semiconductor
Signal processing SPDT (single-pole double throw), switches, 216
analog, 499 193, 201, 212, 213, 216, 228, Step-up/down converter, 213,
digital, 499 232, 235, 236, 290 217
Silicon-controlled rectifier, 38 Speed controller synthesis, topology, 213
biasing a, 39 685–687 Stimuli and response, 806, 807
circuit model of a, 42 block schematic of, 685 of control units, 808–809
Sine and cosine, 714–715, 719 Square root, 711 of subsystems, 808
Sinusoid, peak of a, 103, 108 reciprocal of, 713 Switch
SISO (single-input single-out- Stability robustness, 681, 682 single-pole double throw
put), 433, 542, 649 Standards mechanical, 193, 213
Skin effect, 412–413 AT&T, 837 SPDT, 201, 201, 212, 216
Skin depth, 413 BELL, 837 Switched junction, 468
Snubber, BELLCORE, 837 application of, 470–471
circuits, 243 British Telecom, 837 Switches
series, 83–85 CNET, 837 bidirectional, 42
shunt, 80–83 HRD, 837 high-power, 55
Snubber circuits, 80 MIL-217 of US Military, 837 ideal, 2
turn-OFF, 80 most popular, 837 mathematical representation
turn-OFF dissipative, 80 Standby redundancy, 797–799 of, 2
turn-ON, 80 analysis of, 797 semiconductor, 210, 216, 228
turn-ON dissipative, 80 multiple-survivor configura- switches, 204
SOAR (safe operating area), 19 tion, 798 uncontrolled, 3
forward, 19 with imperfect switching, 799 Switching network, 791
reverse, 19 with multiple failures, 799 dormant failure mode, 792
Solid cylinder, 748–749 Star load, 312 immediate failure mode, 792
PE_Index.indd 920 4/6/2009 6:17:12 PM

Index 921
functions performed by, 791 Tesla, 300 practical, 230

Surge limiting, 117–120 Theorem secondary, 383
System Pythagoras, 714 Transformer, secondary
complex mechanical, 442 Thermal coefficient, 738, 739 single, 383
equilibrium model of a, 427 Thermal conductivity of materi- multiple, 383
input to a, 425 als, 739 Transistors, 1, 174
large-signal model of a, 427 Thermal constants, 739 bidirectional triode, 38
linear time invariant, 542 Thermal resistance of varying bipolar junction, 1, 11
mesh, 786–788 cross sections, 759–760 bipolar power, 11
parallel, 778–786 Thermal resistivity, 738, 739 calculation of power dissipa-
redundant, 779, 787–799 Thermal sense, 807, 809 tion of, 1
portraying, 453 Thermal stress rating, 221 downstream switching, 128
response of a, 425 Thermal viewpoint, 9–11 dynamics characteristics of,
series, 777–778 Thevenin theorem, 167, 174, 14
small-signal model of a, 427 179 gate turn-OFF, 44
speed of response of a, 538 Thevenin voltage, 179 insulated gate bipolar, 1, 30
steady-state model of a, 427 Time delay, 635 MOSFET (metal oxide semi-
System design Time update, 594, 665, conductor field effect), 1,
cost, 769 in Kalman filter, 667 193
good, 770 state-error covariance, 666 NPN, 11
life, 769 value, 666 output characteristics of,
optimal compromise, 770 Torque, 434, 435 14
performance, 769 drive, 490 Paralleling of, 20
primary constraints in, 769– equation, 489–490 PNP, 11
770 Tracker, 585–587, 599–600 static characteristics of, 13
block diagram of a, 599 turn-OFF of, 15
T control system of a, 631 turn-ON of, 16
Tacho, 807 linear quadratic, 672–674 Transmittance, 755–756
Taylor’s series, 426 Transfer function, 435, 510, 511 Thyristors, 1
Technique by hold equivalence, 531 bidirectional, 43
agree apportionment, 814 discrete, 531 family, 37
ARINC apportionment, 815 fourth-order, 446 GTOs (gate turn-OFF), 1
equal apportionment, 814 Transformation MOS-controlled, 46
part count, 837–839 factor, 478 parasitic, 31
part stress, 839–842 stator to synchronous reference Trust factor, 658
prediction by function, frame, 630 Tungsten, 754
836–837 three-phase to two-phase, Turn ratio, 385, 386
similar complexity, 836 630
similar equipment, 836 Transformer, 369 U
Temperature characteristics of an ideal, 388 Uncertainties
effect, 173 class, 455 Additive, 681
influence of, 179 current, 64, 398–403 modeling, 681
junction, 735 ideal, 230, 388 multiplicative, 681, 682
parameters depending on, 179 impedance, 455 Utility power factor converter,
Temperature-regulation coeffi- magnetic, 230 610–613
cient, 172 potential, 383, 398 block diagram, 611
PE_Index.indd 921 4/6/2009 6:17:12 PM

922 Index
V negative, 156–158 Z
Variable negative output, 158 ZCS (zero-current switching),
across, 427 negative voltage, 158 259
effort, 427, 454 Voltage, stator, 484 Zener, 152
flow, 427, 454 direct axis, 481 breakdown resistance, 177,
scalar, 447 quadrature, 481 183
state, 429 Volt-second balance, 200, 204, breakdown voltage, 153
through, 427 217, 265, 374 current, 151–152, 183
Vector principle of, 374 diode, 36, 151, 152, 166, 174,
control, 622–631 Volt-second product, 374 180, 183
input, 485, 486, 488 ideal, 151
output, 485, 486, 488 W model, 183
state, 485, 486, 488 Waveform, 228 non-ideal, 153
Voltage capacitor current, 206, 218 voltage reference, 157
AC, 95 flux, 230 Zener diode, 36, 151, 153, 163,
bidirectional, 199 for a push-pull circuit, 241 166, 174, 180, 183, 570
controlling, 600–607 half-wave rectifier, 10 ideal static characteristics of a,
DC, 95 forward converter, 229 152
forward break-over, 39 inductor current, 222 piecewise linear static charac-
half-wave control output, of various signals of a buck teristics of a, 152
136 converter, 203, 209 static characteristics of, 153
handling capacity, 230, 231 output voltage, 134 Zero-current ripple, 261
input DC, 149 PWM, 721 buck converter with, 259
input source, 611 quasi-square, 350 operative mode of, 267
load, 108 rectifier output voltage, 138 Zeros, 436
minimum capacitor, 109 six-step, 315 Zero vector, 347
minimum input, 109 triangular, 232, 236, 720 choice of, 348
output DC, 149 Weighted factor, 658 periods, 347
pole, 312 Weighted least squares, 658–659 ZVS (zero-voltage switching),
protection, 164 Weight feedback, 807 259
reverse, 31 Winding applied to buck converter,
ripple, 109, 124, 128 primary, 383 271–275
Voltage regulation secondary, 383 operative modes of, 272
output, 175 Window utilization factor, 392 z-Transform, 502–508
Voltage regulator Wood, 754
PE_Index.indd 922 4/6/2009 6:17:12 PM

Chapter 07.indd 421 3/13/2009 7:08:49 PM

Power Electronics Essentials and Applications

Uploaded by

Copyright:

Available Formats

Power Electronics Essentials and Applications

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Power Electronics Essentials and Applications

Uploaded by

Copyright:

Available Formats

Chapter 07.

indd 421 3/13/2009 7:08:49 PM

Other Wiley Editorial Offices:

First Edition: 2009

PE_Preface.indd v 3/24/2009 7:20:16 PM

PE_Preface.indd vii 3/24/2009 7:20:16 PM

PE_Preface.indd viii 3/24/2009 7:20:16 PM

PE_Preface.indd ix 3/24/2009 7:20:16 PM

1 Power Semiconductor Switches 1

PE_TOC.indd xi 3/24/2009 7:51:18 PM

PE_TOC.indd xii 3/24/2009 7:51:18 PM

4 DC–DC Linear Regulators 149

PE_TOC.indd xiii 3/24/2009 7:51:18 PM

Second-Level Modeling 174

5 DC–DC Switched-Mode Converters 193

PE_TOC.indd xiv 3/24/2009 7:51:18 PM

Descriptive Questions 283

6 DC–AC Switched-Mode Converters 289

7 Design of Magnetics 369

PE_TOC.indd xv 3/24/2009 7:51:18 PM

7.2 Inductor 375

8 Modeling of Systems 423

PE_TOC.indd xvi 3/24/2009 7:51:18 PM

9 Control System Essentials 499

PE_TOC.indd xvii 3/24/2009 7:51:18 PM

9.8 Control Principles 540

10 Digital Controller Design 549

PE_TOC.indd xviii 3/24/2009 7:51:18 PM

10.14 Induction Motor Control with Output Feedback 633

11 Optimal and Robust Controller Design 649

12 Discrete Computation Essentials 695

PE_TOC.indd xix 3/24/2009 7:51:18 PM

12.2 Tracking the Base Point in the Fixed-Point System 703

13 Thermal Aspects 735

PE_TOC.indd xx 3/24/2009 7:51:18 PM

Concluding Remarks 763

14 Reliability Modeling 769

15 Reliability Prediction 835

PE_TOC.indd xxi 3/24/2009 7:51:18 PM

15.4 Parts Count Technique 837

PE_TOC.indd xxii 3/24/2009 7:51:18 PM

Chapter 01.indd 1 3/12/2009 3:47:46 PM

1.1 Ideal Switch

Chapter 01.indd 2 3/12/2009 3:47:47 PM

Figure 1.1 Schematic diagram of p–n junction.

Chapter 01.indd 3 3/12/2009 3:47:47 PM

Chapter 01.indd 4 3/12/2009 3:47:48 PM

Figure 1.3 Charge densities near a biased p–n junction.

Chapter 01.indd 5 3/12/2009 3:47:48 PM

Chapter 01.indd 6 3/12/2009 3:47:48 PM

Chapter 01.indd 7 3/12/2009 3:47:48 PM

1. Rectiﬁer diodes (trr in microsecond range).

1. Average current, I Fav

Chapter 01.indd 8 3/12/2009 3:47:49 PM

1.3 Surge Current

1.4 Thermal Viewpoint

Chapter 01.indd 9 3/12/2009 3:47:49 PM

Figure 1.6 Diode model.

Chapter 01.indd 10 3/12/2009 3:47:49 PM

1.5 Bipolar Power Transistors