In several projects I have had a need to measure current consumption on battery-powered devices, and there has never been any really good way, no apparatus or tools I found have proven sufficient. The typical use case is when you develop a low-power product with a battery that should last for a long time, but with potentially high peak currents, for instance during radio transmission bursts. You need to understand how much power your product uses during normal operation in order to optimize the hardware and firmware for lowest power consumption, but there is no way without having high resistors in series with your battery, which ruins operation if the device for instance needs to do a radio burst.

I did this as a simple measurement shield for Arudino Uno, the schematics and layout is in this GitHub repository, together with the Arduino firmware.

Tested and verified rev B of the schematics, works nicely. A lot of improvements remains (see below), but at least first shot is working now.

• 1 uA – 1 A measurement range
• No switching series resistors that could interfere with operation, the entire range measured using the same 50 mOhm resistor
• Data output on serial port to PC showing momentary voltage and current

• Display accumulated energy use (mAh or mWh) since measurement start
• Time stamped logging to SD-card
• Output on display shield for standalone operation
• Make calibration sticky (may not be neccessary to calibrate on every boot) - store in eeprom
• Increase max voltage range by having a high-impedance divider before the unity-gain OP instead of after. (hits maximum input voltage on the OP; VCC-1,5V)

All of those are quite easy, I just didn’t find the time yet.

The main difficulty here is to amplify say 1 uA over 50 mOhm (that's just 50 nV) to something measureable with reasonable accuracy. I tried using a low noise, very low offset drift OP (offset can always be calibrated for, but offset drift is difficult to manage), the OPA2335. It actually worked reasonably well.

Another pitfall was the series resistor (50 mOhm), which on my PCB turned out to be something like 54 mOhm, due to soldering and pad resistance (4 mOhm is about a mm of copper trace in my case). This was an oversight by me, I should have used a 4-terminal resistor. Instead I just calibrate for the extra 8% in software now, ugly but works.

Another thing I didn't consider was that the OP's can have negative offsets, which mine turned out to have, which ment the first part of each range was below zero. I therefor added a tiny positive offset to make sure I'm always on the positive side.

Apart from this the design is quite straight forward, a low-side current measurement resistor (50 mOhm) being amplified using three separate amplifiers and A/D converters, giving three ranges:

• 0 – 500 uA, 500 nA, theoretical resolution on the Arduino 10-bit A/D
• 500 uA – 20 mA, theoretical resolution 20 uA
• 20 mA – 1 A, theoretical resolution 2 mA

Also a fourth channel measures the voltage through a standard voltage divider. See the schematics on the measurement shield for details.

All amplifiers have a first order low-pass filter (~10 Hz) to average out any spikes higher than the Arduino sample frequency (100 Hz). This means that the measured data is 100% correct, but it's impossible to measure the voltage of spikes shorter than 10 Hz (100 ms).

The rest should be self-explanatory by looking at the schematics. If not just ask me and I'll clarify.

There are two revisions, where rev A was done in a hurry and had a few quirks, don't use that one. Rev B has the correct schematics but I never made a PCB from that one, it was easier to build rev B using the rev A board, with some patches.

The Arduino Uno samples these four channels every 10 ms (exactly!), to make counting energy (mAh) easy. It also chooses the range automatically and outputs the results on a serial port.

Note that since three amplifiers are used, the inaccuracy varies across the range – measurements close to the range breaking points are less accurate than other points, which means there is no true general accuracy statement. This is what I measured on the first sample:

(Reference multimeter | DUT | error)
Range 0:
9 uA | 8 uA | <12%
13 uA | 11 uA | <16%
19 uA | 17 uA | <11%
30 uA | 29 uA | <4%
46 uA | 46 uA | <1%
89 uA | 89 uA | <1%
212 uA | 211 uA | <1%
252 uA | 249 uA | <2%
325 uA | 320 uA | <2%
424 uA | 419 uA | <2%

Range 1:
505 uA | 516 uA | <3%
613 uA | 626 uA | <3%
917 uA | 921 uA | <1%
1197 uA | 1198 uA | <1%
1704 uA | 1714 uA | <1%
11,3 mA | 11,1 mA | <2%

Range 2:
20,8 mA | 21 mA | <1%
51,4 mA | 52 mA | <2%
96,0 mA | 95 mA | <2%
235 mA | 233 mA | <1%
0,53 A | 514 mA | <3%
0,71 A | 697 mA | <2%
0,88 A | 873 mA | <1%

Voltage:
285 mV | 275 mV | <4%
1,726 v | 1755 mV | <2%
2,327 V | 2375 mV | <1%
3,34 V | 3415 mV | <3%
3,60 V | 3680 mV | <3%

So in general, better than 4% accuracy can be expected in average. Accuracy is worse in the lower area of each range (due to limited resolution), and at some points it might theoretically be worse than 4%.

Accuracy on very low currents is bad, don't know why yet. Could just as well be my reference multimeter, I will double check this when I have a chance.

• step response, made sure LP filters on all three ranges has time constant well above the 10 ms sampling rate.
• tested that variations in supply voltage does not affect readings