[Eeglablist] Single-trial ERSPs

[Simon Ruch]

Dear all

I am interested in single-trial ERSP-data. Since I have found some 
interesting activities in specific time-frequency-bins on the study 
level (I used std_precomp to compute ERSPs), I would now like to look at 
single-trial ERSPs.
My basic question is: how can I compute single-trial ERSPs from the 
*.dattimef-data?
I used the following approach:
1)    Compute the absolute power-values from the complex *.dattimef-data:
ABSPOWER = abs(tf-data).^2
2)    Compute the mean baseline power for each frequency and each trial 
in db:
BASELINE(freq, trial) = 10*log10(mean(ABSPOWER(freq, time < 0, trial)))
3)    Compute baseline corrected power values (in dB) for each 
time-frequency point for each trial:
CORRECTEDPOWER(freq,:,trial)=  10*log10(ABSPOWER(freq,:,trial)) -  
BASELINE(freq,trial)

In order to validate my newly computed single-trial ERSPs, I wanted to 
compare the averaged ERSPs provided by std_precomp (*.datersp-files) 
with my self-made ERSPs.
To do so, I averaged my baseline-corrected power values across trials 
for each subject:
MEANPOWER = mean(CORRECTEDPOWER,3);
The ERSP-data that I produced with this approach for each subject and 
each electrode differed strikingly from the ERSPS computed with 
std_precomp (provided in the *.datersp-files).
Is my approach to compute single-trial ERSPs wrong?
Is my approach to average these single-tiral ERSPs wrong?
I assume that eeglab performs some adjustments in the power-baselines 
(across trials, or even across subjects?) before removing the baseline.


-- 


________________________________________________

*Simon Ruch, Lic. phil.*
University of Bern
Department of Psychology
Division of Experimental Psychology and Neuropsychology
Office D101
Muesmattstrasse 45
CH-3012 Bern
Switzerland
Phone +41 (0)31 631 37 28
Fax +41 (0)31 631 82 12
mailto:simon.ruch at psy.unibe.ch
http://www.apn.psy.unibe.ch/content/team/henke


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120920/685cf10a/attachment.html>