Re: filter raw EEG by frequency bands in time domain

[Robert T. Short]

On 5/3/20 1:28 PM, Sergei Steshenko via Help-octave wrote:
> On 03/05/2020 23:09, Nicklas Karlsson wrote:
> > fft transform from time domain to frequency domain, you get 
> > amplitudes for different frequencies. ifft is inverse transform, 
> > guess you do not need this.
>
>
> "fft transform from time domain to frequency domain, you get 
> amplitudes for different frequencies. ifft is inverse transform, guess 
> you do not need this." - huh ? The OP explicitly stated he needed 
> result in time domain, so if FFT processing is to be used, ifft is 
> absolutely necessary.
>
> And specifically regarding "you get amplitudes for different 
> frequencies" - it's an incomplete statement. FFT produces array of 
> complex numbers from real inputs. Real part represents the 'cos' 
> component of the frequency bin, and imaginary part represents the 
> 'sin' component of the frequency bin.
>
>
> --Sergei.
From experience you really don't want an FFT.  The uniform bins don't 
play nice with EKG/EEG signals.  That is less true if you are not doing 
things real time, but still a consideration.  Of course you can build 
any filter you want with overlap add/save techniques but that is way 
overkill for the relatively trivial problem you are up against.

Linear phase is not a major concern here, but still desirable.

So.  Build a bank of bandpass linear-phase FIR filters covering the 
frequencies of interest.  These don't need to be real tight, but it is 
so easy to do tight filters you might as well go for it.  As with any 
system design you need to know what is required before you build it, and 
if you don't know about FIR filters there are lots of good texts.  That 
stuff has been around about as long as I have so there is no lack of 
good information.

DB