[Eeglablist] question about filtering

[Andreas Widmann]

Hi Karolina,

> Am I right to assume that to get a 1Hz high-pass filter I should enter a 2Hz edge?
> 
> EEG = pop_eegfiltnew(EEG, [],2,[],true,[],1);
Yes,  with default filter order this will give you a 1 Hz cutoff. 2 Hz passband edge is, however, rather extreme; so this is not necessarily what I would recommend for your purpose. You can have a 1 Hz cutoff highpass combined with a narrower transition band if you manually set filter order. For example with 1 Hz transition band width, your stopband edge would be at 0.5 Hz and the passband edge at 1.5 Hz. You compute filter order for this filter as

df = 1; % requested transition band width
fs = 500; % sampling frequency
order = 3.3 / (df / fs) % round to next even integer, here 1650

EEG = pop_eegfiltnew(EEG, [],1.5,1650,true,[],1);

will then give you a 1 Hz highpass with 1.5 Hz passband edge. See here if you want to read more on the underlying concepts:
http://home.uni-leipzig.de/biocog/eprints/widmann_a2015jneuroscimeth250_34.pdf

> Also, only theoretically, is it possible to high-pass filter data twice? Would that distort the data and disable ICA?
> 
> Let's say we have a dataset filtered with a cutoff frequency of 0.1Hz but then we want to run ICA at 1Hz because it's the recommended threshold... Is it possible to take the 0.1 dataset and apply a more robust filter to it?
Technically this is possible. The 0-0.2 Hz band will then be attenuated more strongly. Given the 1.5 Hz passband edge filter I suggested, the differences will be minimal as this band is completely in the stopband of the new filter. Given your 2 Hz passband edge filter there will be small differences which I would not expect to be problematic. However, in general, one should also consider that the convoluted interval is longer (the sum of both filter orders) possibly spreading potential artifacts unnecessarily wide. As computation time for the higher highpass cutoff is considerably shorter due to lower filter order, personally I usually prefer to have to two parallel (on separate copies of the data) rather than consecutive filtering steps.

Hope this helps,
Andreas

> Am 14.06.2018 um 17:41 schrieb Karolina Broś <k.bros at uw.edu.pl>:
> 
> Dear all, 
> 
> I am preprocessing my data and I want to run ICA for artifact rejection on 1Hz-filtered data and then apply weights to 0.1Hz-filtered data... 
> 
> From what I understand, the EEGLAB eegfiltnew function uses passband edges which do not exactly correspond to cutoff frequencies.
> 
> Am I right to assume that to get a 1Hz high-pass filter I should enter a 2Hz edge?
> 
> EEG = pop_eegfiltnew(EEG, [],2,[],true,[],1);
> 
> and to get 0.1 I should enter 0.2Hz?
> 
> EEG = pop_eegfiltnew(EEG, [],0.2,[],true,[],1);
> 
> 
> Also, only theoretically, is it possible to high-pass filter data twice? Would that distort the data and disable ICA?
> 
> Let's say we have a dataset filtered with a cutoff frequency of 0.1Hz but then we want to run ICA at 1Hz because it's the recommended threshold... Is it possible to take the 0.1 dataset and apply a more robust filter to it?
> 
> I would really appreciate your feedback!
> 
> Karolina
> 
> 
> 
> Institute of Applied Linguistics
> University of Warsaw
> ul. Dobra 55
> 00-312 Warszawa
> www.ils.uw.edu.pl
> 
> www.karolinabros.eu
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to eeglablist-request at sccn.ucsd.edu