[Eeglablist] continuous data - why do ICA on continuous data

[Philip Michael Zeman]

Arnaud,

Do you believe the reason 'better' results are obtained by doing ICA on
continuous data rather than by doing ICA on concatenated epochs (that were
previously segmented) is because of the discontinuities in the derivative of
the signal at the concatenation boundries?



----- Original Message ----- 
From: "Arnaud Delorme" <arno at salk.edu>
To: "Philip Michael Zeman" <pzeman at alumni.uvic.ca>
Cc: "Kline, Keith A" <Keith.Kline at uth.tmc.edu>; <Eeglablist at sccn.ucsd.edu>
Sent: Wednesday, April 06, 2005 6:23 PM
Subject: Re: [Eeglablist] continuous data


> Philip Michael Zeman wrote:
>
> >You can do ICA on the unsegmented data. One of the functions called by
> >pop_runica concatenates all 'trials' or 'segmented epochs' before doing
ICA.
> >You should get the same components whether you are concatentating
previously
> >segmented data or you are running it on continuous data.
> >
> >Viewing the ICA sources (EEG.icaact) is a difference story.  If your data
is
> >really noisy it will be harder to see anything in the continuous data
than
> >if you segment your data and then ensemble average it.  (This can be done
> >after running ICA on the continuous data if you wish.)
> >
> At this point of our research and comparing between ICA applied to
> concatenated data epochs or applied to the continuous data, it seems
> that running ICA on the raw data returns cleaner (and more dipolar) ICA
> components (this is an observation made by my colleague Julie Onton who
> compared the two). You should high pass the raw data above 0.5 Hz before
> running ICA though.
>
> Arno
>
>
>