[Eeglablist] filtering data

[Tjcollins3 at aol.com]

Kevin,
It appears to me that your ringing problem is a classic effect of having a 
rectangular window (aka "boxcar" function) for the EEG data prior to doing an FT 
on it.  I'd suggest applying a partial raised cosine taper window to the data 
prior to sending it into the FT.  (I'm presuming that the EEGLAB FT filter 
function does not apply any windows or perform any other preprocessing to the 
data prior to the FT.  If there is a window function applied, then there is no 
need to apply another window.)

The partial raised cosine taper window function is as follows:

        0.5[1-cos (px/d)]        for 0 £ x £ d
u (x) = 1                              for d £ x £ 1-d
        0.5[1-cos (p(1-x)/d)]          for 1-d £ x £ 1  where 0 £ d £ 0.5

I found d = 0.1 to be useful (i.e., taper the first 10% and the last 10% of 
the signal while leaving the remainder unaffected).  I wrote a short MAT 
function which you can use that is set up for d = 0.1.  (I have the SP Toolbox and 
still needed to write this function.  It's cheap and well worth the investment.)

In case the special characters in the above equation get screwed up, I'm 
attaching a short narrative in MS Word with the reference cited.

One final note, AOL automatically zips multiple attachments, so if your 
institution filters zip file attachments from emails please advise and I can send 
each separately to you.

Hope this is helpful.
Thomas J. Collins III


In a message dated 2/10/05 11:20:30 AM Central Standard Time, arno at salk.edu 
writes:


> Subj:Re: [Eeglablist] filtering data
> Date:2/10/05 11:20:30 AM Central Standard Time
> From:    arno at salk.edu (Arnaud Delorme)
> Sender:    eeglablist-bounces at sccn.ucsd.edu
> To:    guisek1 at mail.montclair.edu (Kevin Guise)
> CC:    eeglablist at sccn.ucsd.edu
> 
> 
> 
> 
> 
> >> I am trying to band pass filter my eeg data between 1 and 30 Hz.  I 
>> noticed that EEGlab uses a fourier transoform and what appears to be an ideal 
>> band pass filter. I have some experience using such a method to filter digital 
>> images for stimuli, and found that I get 'ringing' in the image using ideal 
>> filter functions.  Might this be the case while filtering one dimensional 
>> eeg data?  Might it be advisable to use a butterworth function? Basically what 
>> I am wondering is this: does eeglab use the fourier filter because I do not 
>> 
> Yes 
> it does. I programmed this filter because it was a better alternative than 
> no 
> filtering at all for users who do not have the signal processing toolbox. I 
> observed that with this type of filter the signal drop-off was not as high 
> as 
> with standard FIR filtering. I do not know about ringing. The best 
> alternative 
> to it might be to buy the signal processing toolbox?
> Arno
> 


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20050210/19ff782e/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: prcos.ZIP
Type: application/zip
Size: 3009 bytes
Desc: not available
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20050210/19ff782e/attachment.zip>