[Eeglablist] Filtering trouble

[Andreas Widmann]

Hi Erin,

> I am trying to filter a set of 128-channel EEG data and am not sure
> that the filter tool is working correctly -- and I'm not sure how I
> can tell.  I read that using a bandpass filter is unpredictable in
> EEGLab, so I started with a lowpass filter and then moved onto the
> highpass, checking the spectral content of the data before and after
> each filter.  However, my spectrum diagram winds up looking not much
> like the example diagram: rather than a few distinguishable lines
> there are swaths of color.  I imagine this is due to the large number
> of channels?  When I filter, the shapes of some of these swaths
> change, but others show no change, leaving me confused about whether
> the filter worked.  I have attached spectral plots taken before and
> after the lowpass filter (threshold = 45 Hz) to illustrate the
> problem.
The "swaths" slightly resemble stopband ripple (which is not a problem 
per se), however with a lowpass cutoff frequency of 45 Hz there should 
be no stopband in this frequency range.

Would you mind posting the commands and parameters used for filtering 
(the script or the output of "eegh" after filtering in case you use GUI)?

Have you already tried the firfilt plugin?

> Additionally, it seems I am unable to use 0.1 Hz as the lower edge of
> my accepted band -- when I try to do the highpass step, I get the
> following error:
> Filter error: continuous data portion too narrow (DC removed if highpass only)
> Processing continuous data (307201:450306)
> eegfilt() - performing 7680-point highpass filtering.
> eegfilt() - highpass transition band width is 0.01 Hz.
0.1 Hz is an extreme filter. The shorter the transition band the higher 
the filter order/length, and the EEGLAB built-in fir filter requires the 
epoch to be longer than some multiples of filter length. Do you really 
need the cutoff frequency to be so low? Using 0.5 Hz might already help.

Best,
Andreas

> In case it matters: I am using EEGLab v. 6.01b on Matlab student
> version 7.0.0.27 (Release 14), and my data were recorded using
> Biosemi.
> 
> Thanks for any insight you can offer, and my apologies if this is a
> common topic.
> 
> Erin McMullen Jonaitis
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 3257 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20080314/dfdc9dc0/attachment.bin>