Arduino library for the INA226 power sensor.

Experimental library for the INA226 power sensor. Not all functionality is tested / investigated.

Read datasheet for details.

==> USE WITH CARE

The INA226 is a voltage, current and power measurement device. A few important maxima, see datasheet, chapter 6.

Version 0.5.0 introduced a breaking change. You cannot set the pins in begin() any more. This reduces the dependency of processor dependent Wire implementations. The user has to call Wire.begin() and can optionally set the Wire pins before calling begin().

• Ω == Ohm = ALT-234 (Windows)
• µ == micro = ALT-0181 (Windows)

• https://www.ti.com/product/INA226#tech-docs
• https://www.ti.com/product/INA226#params
• https://www.ti.com/document-viewer/INA226/datasheet
• https://github.com/RobTillaart/INA219
• https://github.com/RobTillaart/INA226
• https://github.com/RobTillaart/INA236

The sensor can have 16 different I2C addresses, which depends on how the A0 and A1 address lines are connected to the SCL, SDA, GND and VCC pins.

See table - from datasheet table 2, page 18.

To be elaborated, example sketch available.

(From Datasheet)
The INA226 supports the transmission protocol for fast mode (1 kHz to 400 kHz) and high-speed mode (1 kHz to 2.94 MHz). All data bytes are transmitted most significant byte first.

Calibration with setMaxCurrentShunt() is mandatory to get getCurrent() and getPower() to work.

Some initial tests shows that the readings do not 100% add up. I expect this is caused by fluctuations in my power supply used and more important that the ADC is multiplexed so there is time between the bus voltage measurement and the shunt voltage measurement. If the current has changed a bit these values are not necessary in line.

Did some measurements with a load of 194 ohm and a shunt of 0.002 ohm that is a factor 10e5 Being on the edge of the sensitivity of the ADC measurements of current were up to ~9% too low. Possible cause is that some maths is done in 16 bit so numbers are truncated, not rounded.

(see issue #2) Sensors may have a different shunt resistor than the 0.002 I have. You should always check and verify what is on the shunt and even verify with a DMM that this value is correct. With the calibration function setMaxCurrentShunt() one can just set the actual value and even compensate slightly if readings are structural too low or too high.

I noted that the getPower() function does not always equal getBusVoltage() times getCurrent(). Cause is rounding/trunking maths and time of measurement. You might prefer to multiply those values yourself to get extra digits. Please be aware that more digits is not always more exact (think significant digits).

The example sketch INA226_setMaxCurrentShunt.ino switches between two calibration modes. It shows the INA266 sensor needs time to accommodate to this change. In practice you should call setMaxCurrentShunt() only once in setup().

Also see #30 for another typical deviation problem.

#include "INA226.h"

• INA226(const uint8_t address, TwoWire *wire = Wire) Constructor to set the address and optional Wire interface.
• bool begin() initializes the class. returns true if the INA226 address is on the I2C bus. Note: one needs to set Wire.begin() before calling begin().
• bool isConnected() returns true if the INA226 address is on the I2C bus.
• uint8_t getAddress() returns the address set in the constructor.

Note the power and the current are not meaningful without calibrating the sensor. Also the value is not meaningful if there is no shunt connected.

• float getShuntVoltage() idem, in volts.
• float getBusVoltage() idem. in volts. Max 36 Volt.
• float getCurrent() is the current through the shunt in Ampere.
• float getPower() is the current x BusVoltage in Watt.
• bool isConversionReady() returns true if conversion ready flag is set.
• bool waitConversionReady(uint32_t timeout = INA226_MAX_WAIT_MS) active waiting for ready flag. Polling for max timeout time, default 600 milliseconds, for wake up time.

The library has helper functions to convert above output to a more appropriate scale of units.

Helper functions for the milli scale.

• float getBusVoltage_mV() idem, in milliVolts.
• float getShuntVoltage_mV() idem, in milliVolts.
• float getCurrent_mA() idem, in milliAmpere.
• float getPower_mW() idem, in milliWatt.

Helper functions for the micro scale.

• float getBusVoltage_uV() idem, in microVolts.
• float getShuntVoltage_uV() idem, in microVolts.
• float getCurrent_uA() idem, in microAmpere.
• float getPower_uW() idem, in microWatt.

Note: The internal conversions runs in the background in the device. If a conversion is finished the measured value is stored in the appropriate register. The last obtained values can always be read from the registers, so they will not block. Result can be that you get the very same value if no new data is available yet. This is especially true if you increase the number of samples. (See also discussion in INA219 issue 11).

Using more samples reduces the noise level, but one will miss the faster changes in voltage or current. Depending on your project needs you can choose one over the other.

As a rule of thumb one could take the time between two I2C reads of a register as an upper limit. This would result in a fresh measurement every time one reads the register. NB it is always possible to average readings fetched from the device in your own code.

• bool reset() software power on reset. This implies calibration with setMaxCurrentShunt() needs to be redone. Returns true upon success.
• bool setAverage(uint8_t avg = INA226_1_SAMPLE) see table below. (0 = default ==> 1 read), returns false if parameter > 7.
• uint8_t getAverage() returns the value set. See table below. Note this is not the count of samples.
• bool setBusVoltageConversionTime(uint8_t bvct = INA226_1100_us) see table below. (4 = default ==> 1.1 ms), returns false if parameter > 7.
• uint8_t getBusVoltageConversionTime() return the value set. Note the value returned is not a unit of time.
• bool setShuntVoltageConversionTime(uint8_t svct = INA226_1100_us) see table below. (4 = default ==> 1.1 ms), returns false if parameter > 7.
• uint8_t getShuntVoltageConversionTime() return the value set. Note the value returned is not a unit of time.

Note: times are typical, check datasheet for operational range. (max is ~10% higher)

Note: total conversion time can take up to 1024 * 8.3 ms ~ 10 seconds.

See datasheet.

Calibration is mandatory to get getCurrent() and getPower() to work.

• int setMaxCurrentShunt(float ampere = 20.0, float ohm = 0.002, bool normalize = true) set the calibration register based upon the shunt and the max Ampere. From these two values the current_LSB is derived, the steps of the ADC when measuring current. Returns Error code, see below.
• bool isCalibrated() returns true if CurrentLSB has been calculated by setMaxCurrentShunt(). Value should not be zero.
• float getCurrentLSB() returns the LSB in Ampere == precision of the calibration.
• float getCurrentLSB_mA() returns the LSB in milliAmpere.
• float getCurrentLSB_uA() returns the LSB in microAmpere.
• float getShunt() returns the value set for the shunt in ohm.
• float getMaxCurrent() returns the value for the maxCurrent which can be corrected.

To print these values in scientific notation use https://github.com/RobTillaart/printHelpers

setMaxCurrentShunt() will round the current_LSB to nearest round value (typical 0.001) by default (normalize == true).

• The user must check the return value == 0x000, otherwise the calibration register is not set.
• Normalization typically gives smaller steps => improve precision
• Normalization can cause that the maxCurrent passed cannot be reached any more. Solution is not to normalize if this max range is needed.

Note: in 0.5.1 the setMaxCurrentShunt() function is rewritten after it showed a bug when normalize flag was set to true. See #29 for details of the discussion.

See datasheet, partially tested.

Mode = 4 is not used, is also a shutdown() unknown if there is a difference with mode == 0.

• bool setMode(uint8_t mode = 7) mode = 0..7. The value 7 == ShuntBusContinuous mode.
• uint8_t getMode() returns the mode (0..7).

Descriptive mode functions (convenience wrappers).

• bool shutDown() mode 0 - not tested yet
• bool setModeShuntTrigger() mode 1 - not tested yet - how to trigger to be investigated
• bool setModeBusTrigger() mode 2 - not tested yet
• bool setModeShuntBusTrigger() mode 3 - not tested yet
• bool setModeShuntContinuous() mode 5
• bool setModeBusContinuous() mode 6
• bool setModeShuntBusContinuous() mode 7 - default.

See datasheet, not tested yet.

• bool setAlertRegister(uint16_t mask) by setting the mask one of five types of over- or underflow can be detected. Another feature that can be set is the conversion ready flag. Returns true if write to register successful.
• uint16_t getAlertFlag() returns the mask set by setAlertRegister().
• bool setAlertLimit(uint16_t limit) sets the limit that belongs to the chosen Alert Flag. Returns true if write to register successful.
• uint16_t getAlertLimit() returns the limit set by setAlertLimit().

The alert line falls when alert is reached.

• uint16_t getManufacturerID() should return 0x5449
• uint16_t getDieID() should return 0x2260

• uint16_t getRegister(uint8_t reg) fetch registers directly, for debugging only.

use at own risk In issue #26 a hack is made to scale the INA226 to 300A by using a very small shunt. The library has a minimal limit for the shunt of 0.001 ohm. This limit can be overruled to support other ranges like the one discussed in #26. Overruling can be done by patching the following value in the INA226.h file.

#define INA226_MINIMAL_SHUNT             (0.001)

Be aware that

• you should NOT do this unless you understand the implications.
• you do this at your own risk.
• the resistance of wires used affect measurements with very small shunts.
• solder might change the resistance too.
• you do this at your own risk.

• update documentation.
• keep in sync with INA219 where possible.

• test different loads (low edge).
• test examples.
• investigate alert functions / interface.
• disconnected load.
  • can it be recognized? => current drop?

• clean up magic numbers in the code

• lastError() do we need this?
  • no
• if BVCT SVCT is set to 6 or 7
  • does the long timing affects RTOS? ==> yield()
  • wait for issue
• expand unit tests possible?
  • need virtual device => too much work
• cache configuration ? ==> 2 bytes.
  • what is gained? updates are faster. footprint code?
  • how often operational?
  • 15 times used..
• can the calibration math be optimized?
  • integer only?
  • less iterations?
  • would cause rounding errors
• make defines of "magic" numbers
  • const floats (most used only once)
• default address 0x40 ?
  • the the user set it makes it always explicit.

If you appreciate my libraries, you can support the development and maintenance. Improve the quality of the libraries by providing issues and Pull Requests, or donate through PayPal or GitHub sponsors.

Thank you,