CH.
1 SAMPLING AND RECONSTRUCTION
Reviews
1) Continuous-time Sinusoidal Signals
x( t)= Ae jt x* ( t)= Ae - jt
; = 2f
: radian frequency
jt - jt y( t)= A cos t = A } 2 {e + e * =1 2 {x( t)+ x ( t) }
2) Exponentially Decaying Sinusoid
x( t)= Ae ( - + j)t = Ae - te jt x* ( t)= Ae ( - - j)t = Ae - te - jt
; = 2f
: radian frequency
- t jt - jt y( t)= Ae - tcos t = 1 } 2 Ae {e + e * =1 2 {x( t)+ x ( t) }
3) Freqnency Response of Linear Systems
y( t) = h( t- )x( ) d = x(t- )h( ) d
-
Y ( s) = H( s)X ( s) x( t) = e jt,
-
s= j
Y ( ) = H( )X ( )
4) Steady-State Sinusoidal Response if
y( t) = e j( t - )h( ) d
= e jt = e jtH( )
h( )e - j d
- 1 -
By principle of superposition,
x( t) = A 1e j t + A 2e j t
1 2
y( t) = A 1e j tH( 1 )+ A 2e j tH( 1 )
1 2
6) Fourier Series
x( t) :
a periodic function with period T
x ( t) = c n e jn t
n =- T T
; 0=
where
1 cn = T
/2
- /2
x( t)e - j n tdt
0
7) Fourier Transform
x( t) :
X ( ) = x( t)e - jtdt
-
an energy signal
1 x( t) = 2 =
X ( )e jtd
X ( f )e j2ftdf
- 2 -
FT(Fourier Transform)
x(t )
~ x (t )
-T
-T / 2
T /2
x( t) :
an energy signal
Consider the Fourier Series representation:
x( t)=
c n e j nt n =-
0
(*)
-
where
1 cn = T 1 =T
T x( t)e j ntdt T x( t) e j ntdt
/2
0
- /2
() x( t) = x( t), if |t |T x( t) = 0, otherwise.
=>
From Eq. (*)
x( t)
X ( ) the envelope of Tc n = x( t)e - j tdt
-
1 X( n) cn = T 0
1 X ( n )e j nt = T 0 n =- = 21 X ( 0n )e j nt 0, n =-
0 0
1 = 0 ) (T 2
As T
, 0.
0
x( t) = 21
Therefore,
X ( )e jtd
///
Equivalently, FT can be represented in terms of the frequency:
1 x( t) = 2 = X ( ) =
X ( )e jtd
where
X ( f )e j2ftdf x( t)e - jtdt
- 3 -
Sampling Theorem
If the highest frequency contained in an analog signal is sampled at a rate sample values.
f s 2f max = 2B ,
then
x( t)
x( t)
is
f max = B
and the signal ///
can be exactly recovered from its
Example
speech audio video
f max =4 kHz f max =20 kHz f max =4 MHz
fs fs fs
8 kHz 40 kHz 8 MHz
The minimum sampling rate allowed by the sampling theorem: f s,min = 2f max ; Nyquist rate For arbitrary values of
f s/2 [ - f s/2, f s/2 ]
fs
, ; Nyquist frequency or folding frequency ; Nyquist interval
- 4 -
DSP Frequency Units
f
: ordinary frequency [ hertz = cycle/second ] : nomalized frequency [ cycle/sample ]
= 2f : radian frequency [ rad/second ]
f/f s
= 2f/f s : digital frequency in [ rad/sample ]
- 5 -
Spectra of Sampled Signals
x( t): sampled signal of x( t)=
x( t)
x( nT) ( t - n T ) -
DTFT(Discrete-Time Fourier Transform)
Def: the Fourier Transform of the sampled signal
X( f )
=
=
x( t)e - j2ftdt
x( nT) ( t - nT ) e - j 2ftdt - n =-
= x( nT)
n =- n =-
= x( nT)e - j 2fnT
Properties
( t - nT ) e - j 2ftdt
(*)
; DTFT
X ( f ) is computable from the knowledge of the sampled values. 1)
2) periodic in
with period of
f s:
X( f + m f s)
= x( nT)e - j2 ( f + mf s)nT
n =-
= x( nT)e - j2fnTe - j2f smnT
n =-
= x( nT)e - j2fnT n =- = X( f)
3) Inverse DTFT
;( ) e - j2mnf sT =1
X( f) Eq (*) may be thought of as a Fourier Series expansion of the periodic function
in the frequency domain. Thus,
x( nT ) = f1
f s f
s/2
- s /2
X ( f )e
j 2ff n s
df
4) Numerical Approximation
- 6 -
X ( f) = x( t)e - j2ftdt
-
x( nT)e - j 2fnTT = T X( f)
n =-
( f ) X ( f ) = lim TX T0
5) Approximation by keeping only a finite number of time samples
x( nT ) :
X( f)
X L ( f ) = x( nT)e - j 2fnT
n=0
L-1
; time windowing
6) Relationship with the Z-transform Z-transform of sequence
x( n ) : X ( z ) = x( n )z - n
n =-
X( f)
= x( nT)e - j 2fT n
n =-
= x( nT)e
n =-
f -j 2 f n
s
= x( nT)e - j n n =- = X ( z )| z = e j
- 7 -
Spectrum Replication
x( t): sampled signal of x( t)
x( t)
= x( nT) ( t - n T )
n =-
= x( t) ( t - nT ) n =- = x( t)s( t)
where
s( t) =
n =-
( t - nT )
Using Fourier series expansion, we may rewrite as
s( t) = c n e
n =- n =-
j 2 T nt
j 1 =T e
2 nt T
where
1 cn = T
Then,
T T
/2
- /2
( t) e
-j 2 T nt
1 dt = T
s
x( t)= x( t)s( t) =
X ( f )=
x( t)e j2nf t T n =-
X ( f - nf s) T n =-
- 8 -
Aliasing and Anti-aliasing Prefilter
- 9 -
Example 1.5.2
x( t): 1) The frequency spectrum of the sampled signal
Its magnitude is
2) The windowed spectrum:
= 0.2,f s = 1
Hz and 2 Hz,L = 10
- 10 -
3) Alternatively,
Combining the two expressions, we may obtain the following identity:
4) As
T0 ,
- 11 -
Sol)
(a) x=y=15dB (b) y = 15 +
A stop = x 50
- 12 -
Analog Reconstructors
Two reconstructors: - ideal reconstructor - staircase reconstructor
The sampled input:
y( t)=
y( nT) ( t - nT ) n =-
The reconstructed analog output:
ya ( t) = y( t)* h ( t) = y( )h ( t- )d
= = y( nT)
n =-
y( nT) ( t- nT ) h ( t- )d - n =-
= y( nT)h ( t - nT )
n =-
( t - nT ) h ( t - )d
Also note that
Y a ( f )= H( f ) Y( f )
Y ( f )=
Y ( f - nf s) T n =-
- 13 -
1) Ideal Reconstructor
H( f ) = T,
Then, for
{ 0,
|f |f s/2 o.w.
[- f s/2, f s/2] ,
1 Y( f) = Y( f) Y a ( f )= H( f ) Y ( f) = T T
The impulse response:
h ( t) = - H( f)e j2ftdf = - f /2 Te j2ftdf T ( e jf t - e - jf t) = 1 2j sin ( f t) = j2 s j2f st t sin ( f st) = = sinc( f st) f st
f s/2
s s s
- 14 -
2) The Staircase Reconstructor
h ( t) = u ( t)- u ( t - T ) H( f) = h( t)e - j 2ftdt = 1e - j 2ftdt
1 ( e - j 2fT -1) = 1 e - j fT ( e jfT - e - j 2fT ) = -j 2f j 2f 1 fT ) e - j fT = j 2 f e - j fT2j sin ( fT) = T sin (fT = T sinc( fT) e - j fT
- 15 -
Anti-image Postfilter:
Equalizer filter:
fT jfT H EQ ( f) = HT = ( f) sin ( fT ) e ,
for [- f s/2, f s/2]
- 16 -
- 17 -
Digital Equalization Filter and Anti-image Postfilter
A Typical DSP System
- 18 -
