Digital Lowpass Filters, basic questions

Question

Forgive the newbie questions, I'm just starting out and wanted some basic information.

I have a signal coming from a digital gyro, most of the useful information is below 90hz but I am seeing significant noise spikes from mechanical motor vibration at 200-400hz (the spike moves depending on motor RPM). This needs to be filtered in realtime to reject the noise with as little delay as possible.

1. When the sampling rate is increased, say from 1000hz up to 8000hz, does this increase the delay?
2. When sample rate increases, is the number of samples/cpu load per timestep for the same filter increased?
3. In the above situation, what is a good way to increase rejection of signals above 90hz while keeping the delay low? ie: filter type, raise/lower sampling rate...etc.
4. An existing system I've seen uses an LPF followed by 2 bandstop/notch filters positioned over the noise spikes. The designers say this 3 filter approach creates less latency than doing a single more aggressive LPF, does this sound right?

Answer 1

1. If you want as little delay as possible, I recommend notch filters at your resonance frequencies. If your resonance frequencies are fixed, it should be easy. Otherwise, a low-pass filter at about 100 Hz should do the trick.

2. No your delay depends mostly on the cut-off frequency, the order of the filter and the filter type. An order-2 IIR filter will have a smaller delay than a 30-tap FIR filter . Your sample rate will not affect (or barely) the delay. Typically, the lower the cut-off frequency, the higher the delay and vice versa. Also, an order-2 low-pass IIR will typically have a smaller delay than an order-4 low-pass IIR for the same cut-off frequency.

3. If you can live with non-constant group delay, an IIR should do the trick.
4. Yes it does, but it assumes that you know what your resonance frequency is. You can maybe adapt your notch filter to the resonance frequency in real-time.. it is feasible but can be a lot of work.