INA220 / INA220-Q1: Measuring current of PWM-driven load

Hi,

It appears you’re trying to measure the average of the “PWM current”. Theoretically it is doable as long as the averaging period is much greater than the PWM period. However, if the input common mode voltage is also PWM, it likely will cause problems. The device common mode is expected to be stable.

To understand the problem better, please post relevant schematic.

What will also help is to post differential input voltage (or load current being monitored) and common mode input voltage waveforms. Hand drawings are fine as long as they are properly labeled.

Regards, Guang

Hi Guang, thanks for the quick response.

VBUS is connected to the 12V rail, but you got me thinking about whether it is stable enough. As this is the same rail used to power the load, noise from the PWM switching is slightly coupled into the VBUS input. Could this explain some of the variability we've been seeing in the bus voltage measurements? Could it also affect the shunt measurements somehow? Would it be advisable to add a bypass capacitor or something to the VBUS input to help with this? I did not see any references to bypass capacitors or filters for the VBUS input in the datasheet.

Attached is a schematic of the setup. In my original post I noted that this was an inductive load, but I am seeing the same results with a resistive load. I tested the same load with the same driver configuration on a circuit with and without flyback diode protection and saw the same results, so my schematic shows the simplest circuit that exhibits the measurement problem.

Unfortunately I haven't been able to get useful shunt voltage measurements with the probes I have on hand. I do have a measurement taken at the low side of the load showing the PWM (low voltage means active, high means inactive), see below:

Here are some histograms showing the low variance of data at 100% Duty Cycle (DC) compared with other driver configurations (same 12Ω load each time). Even at 80% DC, the measurements are fairly noisy. At 30% DC the numbers are often negative so not really useful at all. The units on the horizontal axis are mA.

Here's some more info. I've tried using the 32-sample averaging mode as well as 128-sample averaging for both bus voltage and shunt voltage conversions.

I redid the histograms with common ranges so they can be more easily compared with each other, see below. I also graphed the bus voltage measurements to show how they become problematic when we use PWM to drive the load. At 80% DC 5kHz there is a lot of variability but they are somewhat close to the correct values. At 30% DC 25kHz the VBUS measurement is still in the ballpark of the correct value but the shunt measurement is very wrong. In the below graphs, current is shown in mA and voltage is shown in V. The left column shows current and voltage measured by the INA220 during my 30% DC tests, and so on.

I've played around with a few changes to the shunt low-pass filter, including switching the 10Ω with 100Ω resistors, and separately swapping the 1µF with 10µF, but it hasn't seemed to help.

Could you clarify how the INA220 ADC works w.r.t. non-DC signals? The 12-bit conversion time is 532µs but the sampling rate is 500kHz (2µs). I understand this to mean the signal is "oversampled" but over that 532µs how does it internally come to a conclusion on the voltage? At a PWM frequency of 5kHz, there are about 2.6 PWM periods for every 532µs conversion. Does the fact that the signal is changing so often cause the conversion data to be bad? If we use a PWM frequency of 25kHz, now there are about 13 PWM periods for each 532µs conversion. I feel like there could be a problem here but I don't understand enough about the delta-sigma ADC process to be sure.

One idea I had was to slow the heater driver PWM down to <1kHz to take advantage of the Nyquist Sampling Theory. The lower limit of the DRV103 is 500Hz which would be very close to the maximum necessary to meet the Nyquist sampling requirements. This wouldn't work well for the inductive loads though since they operate better with a higher frequency. Does the INA220 require the maximum frequency in the current signal to be <= 1/2 the conversion time in order to measure correctly? Or should the delta-sigma ADC front-end be able to accurately measure the average amplitude of higher frequencies?

Related to this, how should I read the frequency response given in the INA220 datasheet (see below)? It seems like there's large signal loss for frequencies higher than 1kHz, so is this device just unable to measure the average voltage amplitude of a 5kHz or 25kHz PWM signal?

Thanks!

Guang, thanks for pointing that out - I can see how that would be a problem!

We're now looking into the possibility of placing the shunt either below the DRV103 or above the load such that the shunt doesn't see large common mode voltage changes every PWM cycle.

Could it be said that my design ran into the limits of the INA's "common mode rejection ratio"? I see that the INA220 has a common mode rejection of 100-120dB although the frequency range isn't given in the datasheet.

I see the 2 additional capacitors you added to the differential lines. Do you have any recommended values for those capacitors or any resources where I can learn more about their purpose?

Edit: One other thing.. does this explain the poor repeatability of VBUS measurements? VBUS is connected to the 12V rail so is always nominally at 12V, although it sees a small amount of noise coupled with the PWM switching.

Thank you!