Anti-aliasing filter in driver boards / how to avoid aliasing

1

Hello FOCfriends,

Background to my question:

I’ve built a balancing bot, which generally already works, but I’ve noticed some optimization potential regarding the standstill capabilities.

So I want to start with the inner torque control loop and analyze and optimize its behavior in terms of control theory.

I wanted to check the bandwidth of the current controller, the digital low-pass filter, and maybe optimize the FOC loop time, which gives me the sample time of the system and defines the Nyquist frequency. In the next step I want to check the cutoff frequency of an analog low-pass filter in front of the current-sensing ADCs to avoid anti-aliasing (aliasing could results in folding back high frequencies into the low-frequency area and the controller wants to compensate for something that isn’t present, which could be the problem of my standstill performance).

But my driver board - and also 4 others I’ve checked - doesn’t have any anti-aliasing filters for current measurement.

So my question is, do I miss something regarding ADs and anti-aliasing fitlers, or are there badly designed driver boards on the market? I think the first assumption is true, but how do the boards avoid aliasing?

I’ve also read about sigma-delta ADCs. They work with oversampling and decimation, which results in a possibly simpler anti-aliasing filter.

But also with the sigma delta ADs, they are needed.

Thanks BEdyn

2

Let me add more specific details to my setup.
This is the actual configuration in terms of bandwidth/cutoff frequencies:

  • INA181A1 current sense amplifier bandwidth => 350 kHz
  • loop time 0.0006 s = 1666 Hz => Nyquist frequency = 833 Hz
  • digital LPF time constant = 0.01 s => cutoff frequency = 14.6 Hz

So when the signal gets sampled, it should not have any frequencies above 833 Hz to avoid aliasing. But the bandwidth of the current sense amplifier is much higher, and the Nyquist theorem is hurt.

The slow digital low-pass filter can’t heal the situation in my opinion: at the time the signal gets sampled by the ADC the frequencies over the Nyquist frequency get folded back in the low-frequency area, and the signal is corrupted.

Any meanings/help?

Thx