[Eeglablist] Filtering results

[Luca Finelli]

Hi Jeff

I am not sure whether your message was answered before, but here are my 
comments.

On Thu, 28 Oct 2004, Jeff Johnson wrote:
> We're currently using EEGLAB to do several things to our continuous data, 
> including highpass filtering at a cutoff of .1 Hz.  

I assume you used 'Tools > Filter the data', which calls the EEGLAB 
function eegfilt().

> With different values for filter length 
> (between 1000 and 5000 points on data sampled at 256 Hz),we found that
> the amplitude of the sine wave was attenuated to about 20-30% of its
> original size.
> 1) Is the .1 Hz highpass filter attenuating signals at higher frequencies, 
> such as .5 Hz, and if so, to what degree?  I've become aware of the fact 
> that EEGLAB uses a linear filter, so is there any way to find out the slope 
> of this filter?

You can analyze a filter and its characteristics (frequency response, 
slope, etc.) with the fvtool() function provided in MATLAB.
Now in your case eegfilt() tries to design a high-pass filter with 
pass-band starting at fp=0.1 Hz, and stop-band starting at  85% of fp, 
or 0.085 Hz (15% is the assigned fractional width of transition zones in 
eegfilt()).  
With a 5000 points filter, the magnitude response indicates a -3dB 
attenuation at 0.1 Hz, as desired. Note however that attenuation at 0.085 
Hz is still about -10 dB. Attenuation at 0.5 Hz is virtually 0, as 
expected.
On a test 0.1-Hz sinusoidal, peak amplitude reduction was 43%.

> 2) Is there a systematic way that the filter length should be 
> chosen?  MATLAB runs out of memory for the length of our un-epoched data, 
> so we chose to use 3000 points, which encompasses the .1 Hz wavelength (at 
> 256 Hz srate).

eegfilt() provides a guess for the filter length to use. To find an 
optimal length, you could try different lengths and compare the filter 
magnitude responses. In general, once you reach a desired attenuation at a 
given frequency, there is no need to increase the length further.

> I'm somewhat familiar with analog filters attenuating a signal such as "3db 
> down" at a particular frequency, but does this type of terminology apply to 
> the filtering method that EEGLAB uses?

In my experience the frequency response of MATLAB designed filters shows 
approx. -3 dB attenuation at the specified cutoff frequency, in accordance 
with conventional descriptions.

Hope this helps.

Luca

 ________________________________________________
 Luca A. Finelli, Ph.D.
 
 Computational Neurobiology Laboratory
 The Salk Institute
 
 Swartz Center for Computational Neuroscience
 UCSD

 La Jolla CA 92037 - USA
 ________________________________________________
 
 Eeglablist mailing list Eeglablist at sccn.ucsd.edu
 Eeglablist page: http://www.sccn.ucsd.edu/eeglab/eeglabmail.html
 To unsubscribe, send an empty email to eeglablist-unsub at sccn.ucsd.edu
 To switch to digest mode, send an empty email to eeglablist-digest at sccn.ucsd.edu