[FieldTrip] spectral_Granger_causality

[Schoffelen, J.M. (Jan Mathijs)]

Hi Edoardo,

Begin forwarded message:

From: Edoardo Pinzuti <edoardo.pinzuti at gmail.com<mailto:edoardo.pinzuti at gmail.com>>
Subject: spectral_Granger_causality
Date: 30 May 2019 at 12:16:08 CEST
To: j.schoffelen at donders.ru.nl<mailto:j.schoffelen at donders.ru.nl>

Dear M.J Schoffelen,

I am a phd student of the Brain Imaging Center of Frankfurt am Main. I usually take your work on spectral Granger Causality, as  reference.
I have few short questions on nonparametric GC that I hope you can answer.

1) The nonparamteric GC needs the full spectrum decomposition 0 Hz up to Nyquist frequency, is what you usually do ?


Yes.


2) I set the Fieldtrip function to granger.conditional='yes'  , the output is source-->target conditional on all other sources of the network?

I wouldn’t use this functionality, unless you really understand what you are doing. The code is poorly documented, and in general behaves poorly (numerically).

3) I saw that you do the spectral analysis on channels and then project them to source space (multiply with spatial filter) does it make a difference to do the spectral analysis directly on virtual channel data?

I don’t know where you saw this, but indeed the order of doing linear transformations of data in general should not make a difference for the result. So: doing a spatial filtering step, followed by an FFT should yield the same result as doing an FFT followed by a spatial filtering step (provided the same spatial filter is used in both situations).


4) Padding. Would you suggest to pad the data? I have 0.250 ms task length, now I pad the data to 0.400 ms with a spectral resolution of 2.5 Hz and being interested on beta frequencies I use Hanning taper.

For the non-parametric spectral factorization to behave well, you need to have sufficiently smooth spectra (as well as a sufficiently high number of frequency bins). I would recommend using somewhat longer time windows (if possible), use zero padding (cfg.pad in ft_freqanalysis, e.g. of 4), and use the ‘dpss’ + ’tapsmofrq’.

Good luck,
Jan-Mathijs



thanks a lot for your help
Best,
Edoardo



_______________________________________________
fieldtrip mailing list
https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
https://doi.org/10.1371/journal.pcbi.1002202

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20190603/c48061cf/attachment-0001.html>
-------------- next part --------------
_______________________________________________
fieldtrip mailing list
https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
https://doi.org/10.1371/journal.pcbi.1002202