[Eeglablist] Complex infomax implemented in runica.m?

Eelke Spaak eelke.spaak at donders.ru.nl
Mon Jul 28 03:13:10 PDT 2014

Dear EEGlab list,

I would like to perform Infomax ICA on a complex time-frequency
representation of my data, as described in Anemüller, Sejnowski, &
Makeig (2003) Neural Networks. Actually I am primarily using
FieldTrip, but since FieldTrip just uses EEGlab's runica.m I thought
this list was the best place to ask the following:

If I input complex data (i.e. the Fourier coefficients for sliding
time windows for a single frequency band) into runica.m, will the
algorithm computed be the correct complex Infomax algorithm? I tried
this and the output looks very reasonable (i.e. well-interpretable
sensor maps, with primarily real and much weaker imaginary weights,
consistent with what I think is expected for MEG data). But it would
be great if this is confirmed by someone who actually knows about

If complex ICA is not implemented in runica.m, maybe someone has
thoughts on whether it would be easy to extend it such that it
implements complex Infomax?

Many thanks in advance!


More information about the eeglablist mailing list