ACS712

Library for the ACS712 Current Sensor - 5A, 20A, 30A and compatibles.

Description

The ACS712 is a chip to measure current, both AC or DC. The chip has an analogue output that provides a voltage that is linear with the current. The ACS712 library supports only a built in ADC by means of analogRead().

However since 0.3.4 there is an experimental setADC().

The library has 4 core functions:

• float mA_peak2peak(frequency = 50, cycles = 1)
• float mA_DC(cycles = 1)
• float mA_AC(frequency = 50, cycles = 1)
• float mA_AC_sampling(frequency = 50, cycles = 1)

The parameter cycles is used to measure multiple cycles and average them.

To measure DC current a single analogRead() with conversion math is sufficient to get a value. To stabilize the signal analogRead() is called at least twice.

To measure AC current a blocking loop for 20 milliseconds (50 Hz, 1 cycle) is run to determine the peak to peak value which is converted to the RMS value. To convert the peak2peak value to RMS one need the so called crest or form factor. This factor depends heavily on the signal form, hence its name. For a perfect sinus the value is sqrt(2)/2 == 1/sqrt(2). See Form factor below.

For a 60 Hz environment the blocking is ~16.7 milliseconds, still pretty long.

The mA_AC_sampling() calculates the average of the sumSquared of many measurements. This function should be used when the form factor is not known.

Note to make precise measurements, the power supply of both the ACS712 and the ADC of the processor should be as stable as possible. That improves the stability of the midpoint and minimizes the noise.

Resolution

Sensor	mVperA	LSB 10bit	LSB 12bit	LSB 16bit
5 A	185	26.4 mA	6.6 mA	0.41 mA
20 A	100	48.9 mA	12.2 mA	0.76 mA
30 A	66	74.1 mA	18.5 mA	1.16 mA

getmAPerStep();
mA LSB = (5000 mV / maxADC) / mVperA * 1000.0;
mA LSB = (1000 * 5000 mV) / (maxADC * mVperA);

Although no 16 bit ADC built in are known, it indicates what resolution could be obtained with such an ADC. It triggered the experimental supporting of external ADC's with this library.

Calibration and accuracy

The library has no means to calibrate the output or use an offset. However sort of calibrating can relatively easy be done by using the MultiMap library. MultiMap approaches a non-linear mapping by multiple linear mappings.

See https://github.com/RobTillaart/MultiMap.

Tests

The library is at least confirmed to work with the following boards:

Device	Voltage	ADC steps	Notes
Arduino UNO	5.0V	1024	tested with RobotDyn ACS712 20 A breakout.
Arduino UNO	5.0V	1024	tested with Open-Smart ACS712 5 A breakout.
Arduino NANO	5.0V	1024	#18
ESP32	3.3V	4096	#15
Promicro	5.0V	1024	#15

Please let me know of other working platforms / processors (and failing ones!).

Compatibles

Robodyn has a breakout for the ACS758 - 50 A. - See resolution below. This sensor has versions up to 200 Amps, so use with care!

AllegroMicro offers a lot of different current sensors that might be compatible. These include bidirectional and unidirectional ones. The unidirectional seem to be for DC only.

https://www.allegromicro.com/en/products/sense/current-sensor-ics/current-sensors-innovations

Devices that could be compatible:

	ACS720	ACS724	ACS725	ACS732	ACS733	ACS758	ACS772	ACS773	ACS780	ACS781
tested			#44

	ACS37002	ACS37003	ACS71240	ACS3761X	ACS37800	ACS72981
tested

If you have tested a compatible sensor, please share your experiences. (can be done by opening an issue to update documentation)

Resolution ACS758

Not tested, but looks compatible - same formula as above

Sensor	mVperA	LSB 10bit	LSB 12bit	LSB 16bit	directional
50 A	40	122.2 mA	30.5 mA	1.91 mA	bi
50 A	60	81.5 mA	20.3 mA	1.27 mA	uni
100 A	20	244.4 mA	61.0 mA	3.81 mA	bi
100 A	40	122.2 mA	30.5 mA	1.91 mA	uni
150 A	13.3	367.5 mA	91.8 mA	5.74 mA	bi
150 A	26.7	183.1 mA	45.7 mA	2.86 mA	uni
200 A	10	488.8 mA	122.1 mA	7.63 mA	bi
200 A	20	244.4 mA	61.0 mA	3.81 mA	uni

Interface

#include ACS712.h

Base

• ACS712(uint8_t analogPin, float volts = 5.0, uint16_t maxADC = 1023, float mVperAmpere = 100) constructor. It defaults a 20 A type sensor, which is defined by the default value of mVperAmpere. See table below. Volts is the voltage used by the (Arduino) internal ADC. maxADC is the maximum output of the internal ADC. The defaults are based upon an Arduino UNO, 10 bits ADC. These two ADC parameters are needed to calculate the voltage output of the ACS712 sensor.
• float mA_peak2peak(float frequency = 50, uint16_t cycles = 1) blocks ~21 ms to sample a whole 50 or 60 Hz period. Returns the peak to peak current, can be used to determine form factor. The mA_peak2peak() can also be used to measure on a zero current line to get an indication of the lowest detectable current. Finally this function is used internally to detect the noiseLevel in mV on a zero current line.
• float mA_AC(float frequency = 50, uint16_t cycles = 1) blocks ~21 ms to sample a whole 50 or 60 Hz period. Note that a lower frequency, or more cycles, will increase the blocking period. The function returns the AC current in mA. Its working is based upon multiplying the peak2peak value by the FormFactor which must be known and set.
  • 0.2.2 frequencies other integer values than 50 and 60 are supported.
  • 0.2.3 floating point frequencies are supported to tune even better.
  • 0.2.8 the parameter cycles allow to average over a number of cycles.
• float mA_AC_sampling(float frequency = 50, uint16_t cycles = 1) blocks ~21 ms to sample a whole period. The function returns the AC current in mA. (Note it returns a float). Its working is based upon sampling a full period and take the square root of the average sumSquared. This function is intended for signals with unknown Form Factor.
  • 0.2.8 the parameter cycles allow to average over a number of cycles.
• float mA_DC(uint16_t samples = 1) blocks < 1 ms (Arduino UNO) as it calls analogRead() twice. A negative value indicates the current flows in the opposite direction.
  • 0.2.8 the parameter samples allow to average over a number of samples.
  • 0.3.9 calls yield() every 2nd iteration to improve behaviour under RTOS.

mA_AC_sampling performance trick.

A trick to sample faster is to set the frequency to 2 times the actual frequency so to 100 or 120 Hz. This results in sampling only half a period and the same current will be measured. Advantage is that the function only blocks for ~10 ms @ 50Hz (8.5 @ 60Hz). The drawback is about 4x as many variation. So only use if the performance (or less blocking) is needed.

In a similar way one can increase the accuracy (reducing the variation) by setting the frequency a factor 2 lower (25 and 30 Hz). Drawback is a far longer blocking time.

Use with care!

See - https://github.com/RobTillaart/ACS712/issues/38

Midpoint

The midpoint is the (raw) zero-reference for all current measurements. It is defined in steps of the ADC and is typical around half the maxADC value defined in the constructor. So for a 10 bit ADC a number between 500..525 is most likely.

Since 0.3.0 all midpoint functions return the actual midPoint.

• uint16_t setMidPoint(uint16_t midPoint) sets midpoint for the ADC conversion. Parameter must be between 0 and maxADC/2, otherwise midpoint is not changed.
• uint16_t getMidPoint() read the value set / determined.
• uint16_t incMidPoint() manual increase midpoint, e.g. useful in an interactive application. Will not increase if midpoint equals maxADC.
• uint16_t decMidPoint() manual decrease midpoint. Will not decrease if midpoint equals 0.
• uint16_t resetMidPoint() resets the midpoint to the initial value of maxADC / 2 as in the constructor.
• uint16_t autoMidPointDC(uint16_t cycles = 1) Auto midPoint for DC only. Assuming zero DC current. To reduce the noise cycles must be increased even up to 100. This method is typically much faster for DC than the autoMidPoint(freq, cycles) for the same number of cycles. (See issue #35)
• uint16_t autoMidPoint(float frequency = 50, uint16_t cycles = 1) Auto midPoint, for any AC current or zero DC current.
  For DC one can use a high frequency e.g. 1000 Hz to reduce the time blocking. The function takes the average of many measurements during one or more full cycles. Note the function therefore blocks for at least 2 periods which is about 40 ms for 50 Hz. By increasing the number of cycles the function averages even more measurements, possibly resulting in a better midPoint. Idea is that noise will average out. This function is mandatory for measuring AC.
  • 0.2.2 frequencies other than 50 and 60 are supported.
  • 0.2.8 the parameter cycles allow to average over a number of cycles.

Since version 0.3.0 there is another way to determine the midPoint. One can use the two debug functions. (milliseconds > 20 to get at least a full cycle)

• uint16_t getMinimum(uint16_t milliSeconds = 20)
• uint16_t getMaximum(uint16_t milliSeconds = 20)

and take the average of these two values. In code:

uint16_t midpoint = ACS.setMidPoint(ACS.getMinimum(20)/2 + ACS.getMaximum(20)/ 2);

See - ACS712_20_AC_midPoint_compare.ino

The ACS712 has a midPoint level that is specified as 0.5 * VCC. So autoMidPoint() can help to detect voltage deviations for the ACS712. The library does not support this yet.

Form factor

The form factor is also known as the crest factor. It is only used for signals measured with mA_AC().

• void setFormFactor(float formFactor = ACS712_FF_SINUS) manually sets the form factor. Must typical be between 0.0 and 1.0, see constants below.
• float getFormFactor() returns current form factor.

The library has a number of predefined form factors:

definition	value	approx	notes
ACS712_FF_SQUARE	1.0	1.000
ACS712_FF_SINUS	1.0 / sqrt(2)	0.707	default
ACS712_FF_TRIANGLE	1.0 / sqrt(3)	0.577
ACS712_FF_SAWTOOTH	1.0 / sqrt(3)	0.577

It is important to measure the current with a calibrated multimeter and determine / verify the form factor of the signal. This can help to improve the quality of your measurements.

Please let me know if other crest factors need to be added.

Since version 0.3.0 the form factor can be determined by

float formFactor = 2.0 * mA_AC_sampling() / ACS.mA_peak2peak();

See - ACS712_20_determine_form_factor.ino

Noise

Default = 21 mV (datasheet)

• void setNoisemV(uint8_t noisemV = 21) sets the noise level, is used to determine zero level e.g. in the AC measurements with mA_AC().
• uint8_t getNoisemV() returns the set value.
• float mVNoiseLevel(float frequency, uint16_t cycles) determines the mV of noise. Measurement should be taken when there is no AC/DC current or a constant DC current. The level will give a (not quantified yet) indication of the accuracy of the measurements. A first order indication can be made by comparing it to voltage / 2 of the constructor.

Noise on the signal can be reduced by using a low pass (RC) filter. Version 0.3.1 includes experimental code to take two sample and average them. The idea is that ((3 + 5)/2)^2 < (3^2 + 5^2)/2

In theory this should suppress noise levels however more investigation in software noise detection and suppression is needed.

• void suppressNoise(bool flag) experimental noise suppression.

mV per Ampere

Used for both for AC and DC measurements. Its value is defined in the constructor and depends on type sensor used. These functions allow to adjust this setting run-time.

• void setmVperAmp(float mVperAmpere) sets the milliVolt per Ampere measured.
• float getmVperAmp() returns the set value.

Typical values see "Resolution" section above, and the "voltage divider" section below.

Frequency detection

Experimental functionality for AC signal only!

• float detectFrequency(float minimalFrequency = 40) Detect the frequency of the AC signal.
• void setMicrosAdjust(float factor = 1.0) adjusts the timing of micros in detectFrequency(). Values are typical around 1.0 ± 1%
• float getMicrosAdjust() returns the set factor.

The minimum frequency of 40 Hz is used to sample for enough time to find the minimum and maximum for 50 and 60 Hz signals. Thereafter the signal is sampled 10 cycles to minimize the variation of the frequency.

The microsAdjust() is to adjust the timing of micros(). This function is only useful if one has a good reference source like a calibrated function generator to find the factor to adjust. Testing with my UNO I got a factor 0.9986.

Current version is experimental and not performance optimized.

setADC (experimental 0.3.4)

• void setADC(uint16_t (*)(uint8_t), float volts, uint16_t maxADC) sets the ADC function and the parameters of the used ADC. The library uses the internal analogRead() as default. Be sure to set the parameters of the ADC correctly.

The easiest way to implement an external ADC is to make a wrapper function as casting for function pointer is a no go area.

//  set to external ADC - 5 volts 12 bits
ACS.setADC(myAnalogRead, 5.0, 4096);

...

uint16_t myAnalogRead(uint8_t pin)
{
  return MCP.read(pin);  //  assuming MCP is ADC object.
}

To reset to the internal ADC use NULL as function pointer. Be sure to set the parameters of the ADC correctly.

//  reset to internal ADC - 5 volts 10 bits
ACS.setADC(NULL, 5.0, 1023);

• example ACS712_20_DC_external_ADC.ino

• https://github.com/RobTillaart/ACS712/issues/31

• example ACS712_ESP32_external_ADC.ino

• https://github.com/RobTillaart/ACS712/issues/46

Note that the use of an external ADC should meet certain performance requirements, especially for measuring ma-AC().

To 'catch' the peaks well enough one needs at least 2 samples per millisecond (2000 sps) for a 60 Hz signal. That gives 34 samples for 360 degrees => 10.6 degrees, which results in a max deviation of 5.3 degrees from peak => max 0.5% off.

As a 50 Hz signal is a bit slower, 2000 sps would give 40 samples for => 9 degrees, which results in a max deviation of 4.5 degrees from peak => max 0.4% off.

The 16 bit I2C ADS1115 in continuous mode gives max 0.8 samples per millisecond. This will work perfect for high resolution mA-DC() but is not fast enough for doing mA-AC(). It will get an accuracy around ~2%.

The SPI based MCP3202 ao can do up to 100 samples per millisecond at 12 bit. These ADC's are perfect both mA-DC() and mA-AC().

• https://github.com/RobTillaart/ADS1X15
• https://github.com/RobTillaart/MCP_ADC

Voltage divider

As per issue #15 in which an ACS712 was connected via a voltage divider to the ADC of an ESP32.

Schema

ACS712 ----[ R1 ]----o----[ R2 ]---- GND
                     |
                     |
                ADC of processor

The voltage divider gave an error of about a factor 2 as all voltages were divided, including the "offset" from the midPoint zero current level.

By adjusting the mV per Ampere with setmVperAmp(float mva) the readings can be corrected for this "voltage divider effect".

Examples:

For a 20 A type sensor, 100 mV/A would be the normal value. After using a voltage divider one need to adjust the mVperAmp.

R1 (ACS)	R2 (GND)	voltage factor	mVperAmp corrected
10200	4745	4745 / (10200 + 4745) = 0.3175	100 * 0.3175 = 31.75
4745	10200	10200 / (10200 + 4745) = 0.6825	100 * 0.6825 = 68.25
10200	9800	9800 / (10200 + 9800) = 0.4900	100 * 0.4900 = 49.00

Note: setting the midPoint correctly is also needed when using a voltage divider.

Disconnect detection

(to be tested)

To detect that the ACS712 is disconnected from the ADC one could connect the analog pin via a pull-down to GND. A pull-up to VCC is also possible. Choose the solution that fits your project best. (Think safety).

mA_DC() and mA_AC_sampling() will report HIGH values (Out of range) when the ACS712 is disconnected. The other - peak2peak based functions - will see this as zero current (min == max).

Schema with PULL-UP.

                 ACS712 OUT
                     |
                     |
   VCC ----[ R1 ]----o       R1 = 1 M ohm.
                     |
                     |
                ADC of processor

The library does not support this "extreme values" detection.

RTOS

The library can be used in an RTOS environment, however a few functions of this library are blocking for relative long times.

In version 0.3.9 the mA_DC() calls yield() between every three calls of analogRead. This is done both for the external and intern ADC to prevent blocking of other threads.

For the mA_AC() and mA_peak2peak() a call to yield() is not desirable as the samples are all needed to make a decent measurement. For the applications that need proper scheduling one should put the sampling of the INA226 at least for AC in a separate thread.

There is no RTOS example. If you have and willing to share you are welcome.

ESPhome

For people who want to use this library for ESPhome, there exists a wrapper class for this ACS712 library.

• https://github.com/marianomd/acs712-esphome

As I do not have ESPhome know how, please share your experiences. This can be done by an issue.

Operation

The examples show the basic working of the functions.

Future

Must

• test more
  • other than the 20A module
  • 5, 10, 30, 50 ...
  • need to buy extra hardware

Should - 0.3.x

• investigate estimateMidPoint(confidence) See issue #35
  • is less blocking by spreading the sampling over many calls. returning a confidence level.
• investigate noise suppression #21 (0.3.1 and later)
• investigate blocking calls:
  • mA_AC() blocks for about 20 ms at 50 Hz. This might affect task scheduling on a ESP32. Needs to be investigated. Probably need a separate thread that wakes up when new analogRead is available?
  • RTOS specific class?
• investigate detectFrequency(float) blocks pretty long.

Could

• merge mA_AC() and mA_AC_sampling() into one. (0.4.0)
  • or remove - depreciate - the worst one
• add range check to (all) set functions?
• add unit test for autoMidPointDC() (needed?)
• setMidPoint()
  • Q: could midpoint be set beyond maxADC? is there any use case?

Won't (unless requested)

• investigate support for micro-Amperes. ACS.uA_DC()
  • need a very stable voltage
  • needs a 24 bit ADC
  • default noise is already ~21mV...
  • => not feasible in normal setup.
• Should the FormFactor not be just a parameter of mA_AC() it is the only function using it. ==> No unnecessary breaking API
• should cycles be an uint8_t ?
  • No, uint16 allows averaging in minutes range uint8_t just ~5 seconds
• midPoint can be a float so it can be set more exact.
  • extra precision is max half bit = smaller than noise?
  • math will be slower during sampling (UNO)
• split the readme.md in multiple documents?
  • which?
• setADC() to support > 16 bit?
  • uint32_t performance penalty?

Support

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,