Tradeoff Sampling rate / bandpass filter design

Dear all, I'm usually sampling my data with 20 kHz and want to filter the data with a butterworth filter. My question: if I filter the data from e.g. 30-90 Hz. If I use the simple b,a filter logic

nyquist=10000; 
Wn=[30, 90]/nyquist;   
[b,a]=butter(4,Wn,'bandpass'); 
filtered_signal=filtfilt(b,a,signal);

Is it still a good filter design, or is the order of 4 or even 3 to small to gain a valid filter result? Was I read in the threads, there is a tradeoff between the sampling rate, the filter order and the narrowness of the passband....

Thanks for your experience. Best, Peter

Answer 1

[Star Strider]

I suggest using the buttord function to calculate the filter order, and then use the second-order-section implementation in the actual filter operation. The full discussion is here.

[Star Strider]

I’m not sure you need this line:

hd=dfilt.df2sos(sos,g);

I would just use the ‘sos,g’ results in the filter.

Filter design is heuristic. I don’t know what you want to do or what is not working, since you didn’t describe it.

Experiment with different passband and stopband values, since that is the most difficult part of bandpass (or bandstop) filter design. I usually begin specifying the stopband with [0.5, 1/0.5]*Wp (with ‘Wp’ the desired passband in a bandpass filter), and then increase the rolloff to [0.8, 1/0.8]*Wp or so until I get the behaviour I want.

Alternatively, consider a different filter design, for instance Chebyshev or something with a sharper rolloff, if that’s the problem.