P300classifier could not derive weights.

[csp586]

I'm a beginner of BCI2000.
At the moment, I tried P300 speller with g.USBamp.
I recorded several EEG data by following the P300 speller tutorial and 4 recorded files of "THE", "QUICK","BROWN" and "FOX".
But despite lot trials, during the offline analysis I only got <0.001 or 0.0001 r^2 values.
So whenever I applied these files to p300Classifier, it always gave me this error message,

"Unable to generate usable weights, try different parameters or dataset."

Could you give any advice or tips to overcome this problem?
Thank you very much.

[gschalk]

Hi,

It looks like the quality of your recorded EEG is very poor. Did you check it for artifacts as described in the tutorial on the wiki? Did you run OfflineAnalysis to see whether your responses look similar, although small, to those in the tutorials?

I suggest to consult with somebody who has EEG experience.

Gerv

[csp586]

Thank you very much for your quick reply.
I have checked the signal and impedance already. I think the EEG signal was ok without much artifact.
We have one channel p300 speller g.Tec provided with g.USBamp simulink. Interestingly, in the exactly same condition, the software shows sometimes 90% classification results.
Is it possible the firmware upgrade of g.USBamp causes some problem? Quite recently, we upgrade the firmware for g.USBamp. I just wonder it cause some compatibility problem.
Today, we tried BCI2000 using g.MOBILab+ and then we can get classifier's weight by p300classifier. But the classificaiton results just show 70% for the maximum.
We also think CRT monitor may help us, but when we see the g.Tec's software working, the monitor doesn't look main reason.
We are completely stuck at the moment. We really want to make BCI2000 system work with our EEG system. Could you give me any advice?
Thank you.

[gschalk]

Hi,

What are you using to configure the online system? Manual configuration, P300GUI, or P300Classifier? Have you read the P300 tutorial on the wiki or in the BCI2000 book?

Gerv

[csp586]

Thank you for your reply.
I'm using P300Classifier.
I tried the experiment after reading the P300 tutorial on the wiki.
Thus you mean we don't need to consider different setting of stimulus duration or other thing? I will try it again.
Thank you very much, Gerv.

[gschalk]

How many channels do you have? You mention something about one channel.

Gerv

[csp586]

I'm using 8 channels, Fz, Cz, Pz, P3, P4, Oz, PO7, PO8.
Many thanks for your help.

[csp586]

Dear Gerv

Finally, we found out the problem of our amplifier. Strangely, our amplifier could not obtain proper EEG with 256Hz sampling frequency.
We could get P300 with higher sampling frequency.
Now we can go further step.
Thank you very much for your help.

C.