[FieldTrip] Correction baseline : removing the mean value or high-pass filtering ?

Andreas Widmann widmann at uni-leipzig.de
Mon Dec 15 14:36:18 CET 2014

Hi Emmanuelle,

any high-pass filter will at least attenuate the DC offset (the mean) and a linear trend (by definition). A properly designed high-pass filter should have > 80-100 dB DC attenuation and thus, completely remove the DC offset from the data in most cases. As there are exceptions with extreme DC offsets, quite frequently observed for example in BioSemi EEG recordings, the mean is additionally subtracted from the data before high-pass (and band-pass) filtering in Fieldtrip. Have a look at the filter’s frequency response if available.

And yes, your reasoning is correct: a high-pass filter can be used to replace baseline correction in particular applications. This can for example be recommended if you cannot define a clean baseline as for example in many language studies (for detailed discussion see Urbach and Kutas, 2006, doi:10.1016/j.biopsycho.2005.11.012). Baseline correction would remove the mean baseline ERP topography from the data.

Possible (but not necessary) drawback is that you might need higher high-pass cutoff frequencies to remove slow drifts from the data. Take care of possible filter distortions with cutoff frequencies > ~.1 Hz (see Acunzo et al., 2012, pre-print version available here: http://homepages.inf.ed.ac.uk/mvanross/reprints/Acunzo_et_al_in_press_lo_res.pdf and Widmann et al., 2014, pre-print version available here: http://www.uni-leipzig.de/~biocog/eprints/widmann_a2014jneuroscimethods.pdf for discussion).

Hope this helps! Best,

> Am 13.12.2014 um 14:04 schrieb Stephen Politzer-Ahles <spa268 at nyu.edu>:
> Hello Emmanuelle,
> I don't think a high-pass filter can replace demeaning. A high-pass filter removes a linear trend, but not a fixed value. Imagine a case where a trial is a flat line at 100 uV for the whole epoch; demeaning would change this trial to 0 uV for the whole epoch, whereas a high-pass filter would (I think) not change it.
> Detrending, on the other hand, removes a linear trend. There was a discussion very recently on this list discussing detrending vs. high-pass filtering, so you may be interested in checking it out.
> Best,
> Steve
> Stephen Politzer-Ahles
> New York University, Abu Dhabi
> Neuroscience of Language Lab
> http://www.nyu.edu/projects/politzer-ahles/
> Message: 1
> Date: Fri, 12 Dec 2014 15:49:10 +0100
> From: Kristensen Emmanuelle
>         <emmanuelle.kristensen at gipsa-lab.grenoble-inp.fr>
> To: FieldTrip discussion list <fieldtrip at science.ru.nl>
> Subject: [FieldTrip] Correction baseline : removing the mean value or
>         high-pass filtering ?
> Message-ID: <548B0066.3090309 at gipsa-lab.grenoble-inp.fr>
> Content-Type: text/plain; charset="utf-8"; Format="flowed"
> Hi,
> I would need your opinion, please.
> Generally, the baseline correction means to remove a mean value from
> each epoch. I am wondering if a high pass filter, applied on the whole
> recorded signal, can perform the baseline correction.
> What do you think?
> Regards
> Emmanuelle
> --
> Emmanuelle KRISTENSEN
> Doctorante au Gipsa-Lab
> Equipe VIBS
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip

More information about the fieldtrip mailing list