Using p300Speller with Unicorn

[cmontilla]

Hi
I am using P300 Speller with Unicorn. I have applied this BCI2000 reference:

https://www.bci2000.org/mediawiki/index ... UnicornADC

However, when I use P300 Classifier GUI the classifier's results are not very good if I apply cross-validation.
Can someone tell me if this is due to the device settings?

Thanks.

[mellinger]

HI,

you could try to activate the Common Average Reference (CAR) in the P300Classifier GUI.

Also, the P300Classifier GUI shows you how classification accuracy improves with the number of epochs averaged. It is quite normal that you need to average over, say, 15 epochs to get an accuracy of 100%.

-Jürgen

[cmontilla]

Hello,

Thank you very much for your reply.

I understand what happens when the CAR filter is applied.
However, it is not entirely clear to me when it is "necessary" to use it. I have tested what you mentioned and it reaches 100%.
I will try adjusting the numer of sequences so that I use the minimum amount possible.

Thanks again.

[mellinger]

The CAR filter removes signal content common to all electrodes, so it is most useful to get rid of electromagnetic interference originating from sources not too close to the participant, and affecting all electrodes in (nearly) the same way, e.g. a fluorescent lamp.

Another possible situation is an artifact that originates at the ground or reference electrode. Such an artifact would also enter into all signals in the same way, and would be removed by a CAR filter.