[Eeglablist] decomposing unaveraged data trials

Scott Makeig smakeig at ucsd.edu
Sun Oct 23 15:27:58 PDT 2005

Zongzei -

I looked at this paper - we did omit a key word from the Methods 

"... Extended-ICA decomposition was performed on [CONCATENATED] 
31-channel, 1-sec data epochs from 300 to 700  target stimulus trials 
for each of 28 normal control and 22 clinical subjects ..."

We performed one decomposition of the concatenated data (a signale 
matrix of size (channels,epochs*latencies)). Attempting to separately 
decompose single epochs  would not allow clean decompositions (and would 
also present a difficult component clustering problem) -- since each 
epoch does not contain enough data at allow the large number of 
component processes to 'express their independence' (as I like to put it).

Scott Makeig

zonglei zhen wrote:

Hello everyone:
Now I am reading a paper about single trial ICA analysis (Tzyy-Ping Jung et
al,Analysis and Visualization of Single-Trial Event-Related
Potentials,HBM01), but I can not learn the concrete procedure from the
can someone explain to me the details?
In our exp,30 subjects,520 single-trial per subject,32 channel. Do I need
do 520 ICA analysis for every subject?


More information about the eeglablist mailing list