[Eeglablist] (no subject)

[Arnaud Delorme]

Dear Karthick,

> I am working on applying methods from nonlinear time series analysis 
> on eeg. Currently, I have a problem with noise removal. I had earlier 
> used the nonlinear filter given with the Tisean package. I would like 
> to know about applying ICA on single channel eeg for noise removal. Is 
> ICA simialr to PCA? Is there a reference I can read to get started?
>
You can see ICA as a generalization of PCA where the constraint about 
axis orthogonality is released so it can better capture the statistics 
of multi-dimensional data. Standard ICA is designed to separate N 
sources from N sensors. If you only have 1 sensor, then you must use 
over-complete ICA. I suggest that you send a message to the ICA mailing 
list (icalist at inf.enst.fr) for more information about this technical issue.

Best,

Arno