Page 1 of 1

working with p300 speller

Posted: 04 May 2013, 11:53
by saranyamohan
hi,
i am undergraduate student working on p300 speller in bci2000 software,and
we use modular eeg as our amplifier. we followed the procedure given in
wiki and tried conduct a calibration and spelling session after calibration
session we tried doing a offline analysis but failed succeed because of
some error we used a p300 classifier and generated a feature weight which
we stored as parameter file for later use and when we tried to conduct a
spelling session we are getting response even without any connection (i.e
some predefined letters flash) similarly when connected subjects head also
we did not get the targeted letters but we got some predefined random
letters itself can u pls tell us what exactly is the problem and exact
procedure to work with speller.


Thank u

Re: working with p300 speller

Posted: 14 May 2013, 09:44
by mellinger
If you are getting an error in Offline Analysis, it may be due to an incompatibility of the 3.0.5 version with the latest version of Matlab, which may be fixed by updating OfflineAnalysis.m to the latest version from SVN.

With no subject connected, you will get random results because the P3Speller will try to make sense of the random noise it receives. You can use the MinimumEvidence parameter to change this behavior, which is available in the current source code version on SVN.

I cannot understand you want to say when referring to "predifined random letters". In my understanding, something may be _either_ (pre-)defined, _or_ random, but not both.

With a subject connected, you cannot expect reasonable results unless _all_ of the following conditions are met:
* the classifier was trained with the same subject,
* EEG recordings were/are free of artifacts,
* you are recording EEG from central and/or occipital electrodes,
* classifier training data contained an amount of at least 50 p300 responses, i.e. 300 flashes in total,
* the subject was complying in both training and testing session.

Especially, make sure you are aware of how to properly record EEG, which is rather difficult to do without experience, so do not expect to get it right immediately.

Regards,
Juergen