[Eeglablist] Filtering trouble

[Andreas Widmann]

Hi Jakub,

> Is there any way to bypass this? I work on data acquired by Biosemi
> amplifier, having practically no hardware highpass filters, so I need
> a software highpass filter to clean the data, but with a reasonably
> low threshold (0.01)
> And yes, that low (0.01 Hz) filter is a must for me, since higher
> highpass thresholds distort ERPs and are no longer happily accepted by
> reviewers (at least in my field).
Maybe you should describe your data (analysis) in more detail to get 
help in finding a viable solution?

These extreme filters are suited and applicable in analog (and IIR)
filtering where transition and stop band are not clearly separated.
However, this does usually not apply for carefully designed digital FIR
filters.

Just some numbers: With a highpass cutoff (-6 dB) of 0.01 Hz your
maximal transition band width is 0.02 Hz. Using a Hamming windowed sinc
FIR filter you would need a filter length of 42241 points (the eeglab
built in basic FIR filter actually would default to 76801 points).
However, this would mean (in contrast to analog/IIR filters) that you
have *no* signal attenuation down to 0.02 Hz, that is, only signals
cycling slower than once per 50 s are attenuated. What is the standard 
epoch length in your analysis?

The firfilt plugin should be slightly less memory consuming than the 
built in basic FIR filter. Have you already tested?

Best,
Andreas