[FieldTrip] change in default behavior spectral analysis functions

[jan-mathijs schoffelen]

Dear all,

I would like to report a slight change in the default behavior of the spectral analysis routines (ft_freqanalysis, and the low level specest_XXX functions). I do not expect that this change has  a major effect on your analysis results, but I want you to be aware of the change.
The issue at hand is the way in which the low-level spectral analysis routines preprocess the data just prior to taking the fft, and it relates to the following FAQs: 

http://fieldtrip.fcdonders.nl/faq/why_does_my_tfr_look_strange
http://fieldtrip.fcdonders.nl/faq/why_does_my_tfr_look_strange_part_ii

As of revision 3752 of ft_freqanalysis (the version information of the function used to generate your data can be obtained from freq.cfg.version) we introduced a default preprocessing step, which entailed the fitting and subtraction of a first order polynome to the time-domain data, prior to fourier transformation. This step was intended to address both abovementioned FAQs, and removes the DC-component and the linear trend. However, after hot and lengthy discussions in the FieldTrip team we decided to change the default behavior into just removing the DC-component. This change will be effective as of revision 5106.

Once again, I don't expect this change to have whopping effects on your results, e.g. all of a sudden losing significance or so. If it does, it warrants attention because something funny might be going on in the lower frequency components of your data.

If you want to emulate the default behavior of ft_freqanalysis as it existed prior to revision 3752, you need to specify cfg.polyremoval = -1;
If you want to emulate the default behavior of ft_freqanalysis as it existed between revision 3752 and revision 5106, you need to specify cfg.polyremoval = 1;

The new default is now cfg.polyremoval = 0;



Happy computing,

Jan-Mathijs


 
Jan-Mathijs Schoffelen, MD PhD 

Donders Institute for Brain, Cognition and Behaviour, 
Centre for Cognitive Neuroimaging,
Radboud University Nijmegen, The Netherlands

Max Planck Institute for Psycholinguistics,
Nijmegen, The Netherlands

J.Schoffelen at donders.ru.nl
Telephone: +31-24-3614793