Analog track parallel to PWM: crosstalk issues?

Question

I'm designing a PCB for my caving robot, which mainly consists of a STM32F723ZE and 16 motor controllers, with current measurement.

My question is: can I route the PWM signals going to the motor controllers just next to the analog voltage from the current measurement? Or will I get "significant" cross-talk?

In detail:

• track space: 0.15 mm (I could probably increase a little bit if needed, but not much).

• the analog track: the H-bridge has a "VpropI" pin with voltage proportional to current (I expect about 0.5 V). I add a RC filter (R = 100 kΩ, C = 10nF): the output of the RC filter goes to the ADC pin of the STM32 (about 6 cm long track). If needed, I can easily change for R = 10 kΩ, C = 100 nF.

• the tracks are parallel for 5-6 cm.

• the H-bridge supports up to 100 kHz PWM (I haven't chosen the PWM frequency yet, I was planning just to try different frequencies once I get the PCB and attach the motors to it).

• let's say that 20 mV of noise are fine (50 mV would still be OK I think with software filtering).

So, do I have to dig into how to compute cross-talk/reduce it, or can I just go ahead like that?

Is there any simple rule of the thumb to know when I can just ignore cross-talk and when it needs further investigation?

EDIT: as suggested by I did a LTspice simulation:

The blue signal is the "middle point" between the 2 capacitors (that in real PCB is connected to ADC). The green signal is a 3.3V PWM signal, with 40 ns period, and 1 ns or 0.1 ns rise/fall times (NB: I plotted 0.5V+V_PWM/1000 (green signal) in order to get it on a similar scale as the noise).

With 1 ns rise/fall time:

With 0.1 ns rise/fall time:

@Neil_UK: is this the modelisation you where refering to?

If this is the correct way to model it, then with both capacitive and inductive effects, I get at most 5 mV of noise with 0.1 ns rise/fall time. So as long as the rise/fall time is >0.1 ns, I should be fine (crosstalk<5 mV). Correct?

EDIT 2 : new circuit based on the last comment Neil_UK (L1 in series with L1) : So with L1 in series with C1 (nb : I know divide the PWM only by 100 for plotting). With rise/fall time of 100 ns

I have to admit I don't undestand why L1 should be in series with C1 and not with C2, but if so, then results are far worse : with 1ns rise, I get several volts of noise ; with 100 ns rise time, I'm still having about 45mV of noise

Answer 1

Motor current will have a strong ripple component at the PWM frequency.

If you drive the chip with PWM plus direction signal, and the motor receives either VCC or 0V, with polarity determined by direction... then VPROPI will only output a valid voltage when current actually goes through the sense resistor, which means when the PWM is in the ON state. So, instantaneous voltage on VPROPI will be proportional to the product of actual current and sense resistor when the PWM is ON, but it will be zero when it's off. Your filter will average that to a voltage that is proportional to the product of motor current, resistor, and duty cycle. This means at low duty cycle, you probably won't measure anything useful.

If you drive the chip with two independent PWMs, and the motor receives either +VCC or -VCC, which puts the neutral at 50% duty cycle, then current will always flow in the sense resistor and VPROPI will no longer be proportional to duty cycle.

Since the former is preferable, personally I would remove the filter completely and synchronize the ADC to the PWM peripheral (it's somewhere in STM32 manual, I don't remember where) so the ADC samples a small delay after the PWM turns on. To get this delay, add up the H-bridge switching time, current sense amp settling time, etc.

The time to sample sets a minimum duty cycle, but that shouldn't matter for motors, since a very low duty cycle means they won't turn at all anyway. It also sets a maximum switching frequency, which shouldn't matter either, because motors don't need to be driven at 100kHz, more like a few kHz.

If done right, this will result in less noise than the filter option, because when sampling is synchronized to PWM, it always samples at the same point in the current sawtooth waveform, which means it ignores the ripple. And since it samples after the switching is done, it will also ignore switching noise. The ideal solution is to sample all the motor channels while none of them are switching. Time to put that fast ADC to good use.

It also gets rid of the filter delay, so if you want to implement a feedback loop, less phase lag is always a plus.

The datasheet doesn't mention settling time or bandwidth for the current sense amp, which is suspicious. I would strongly recommend you measure it, because if it doesn't have enough bandwidth or settling is too slow, then synchronized sampling won't work. If the current sense amp limits slew rate, even the average with the filter may not work at all.