First of all, you should tell us which of the two P300 processing pipeline you are using. There are two of them in the sample scenarios : the classical one (records few electrodes) and the xDAWN based one (records more electrodes but reduces dimension with a spatial filter).
I'm using the xDawn scenario, but I adjusted some settings. Because 12 repetitions, with averaging of 1 epoch and 12 votes for the voting classifier did not work for me, I adjusted this to 3 votes and 4 averaged epochs.
I also remember now that I threw away one box because I had no clue what is was doing there. It's about the time based epoching after the decimation box. I use sampling frequency of 256Hz and a buffer of size 32 with a decimation of four. So I don't know why this time based epoching box is there? Also shouldn't it be placed before the decimation box instead?
I also removed some other less important things like the signal display during acquisition...
In the zip-file you can find all the configuration files and the scenarios, together with the recorded files. I placed the directory directly under the "share" directory so that all the directory references are correct.
Please put recorded files somewhere so we can look at them.
See zip-file:
http://www.thewired.be/blog/wp-content/ ... m_Demo.zip
The first two files (p300_dieter_12012010_rec1 and p300_dieter_12012010_rec2, I will call them recX for short now, make sure you only look at the one with date 12012010) are recorded with the online test scenario.
rec3 and rec4 with the acquisition scenario.
rec6 and rec7 again with the online scenario later that day and rec8 with the acquisition scenario.
When I train the spatial filter and the classifier on rec1 and test it on rec2, it works. However, when I test it on rec3 or rec4 it fails. Vice versa, when I train on rec3 then it works on rec4, but not on rec1 and rec2.
You did not include configuration in your previous post. Or did you mean you are using the configuration included in the sample scenario ?
see also scenarios in zip-file
I don't understand what you mean here.
there are two bmp files in the zip-file. the rec1 file shows an P300 recorded with the online test scenario which on average comes a bit earlier that the P300 that you can see in the rec3 file that was recorded with the acquisition scenario.
My stimuli based epoching starts at 100ms and takes 500ms in total.
No it should not... if it was the case, you'd have warning messages about realtime that could not be met !
As far as I remember I did not get that message (I only get one message and it's about the compression for the stream writer). But I will check again tomorrow.
Do you mean you press play, press stop, and then press play again ?
Do you have messages in the console that could help ?
Can you send us the designer log file for a session ?
Could it be that the configuration file(s) generated by the offline training scenario are not written on disk before a delay ? (antivirus problem ?)
Exactly. The EEG file rec6 was recorded when I pressed run in the online test scenario for the first time. I then pressed stop and ran the same thing again, this EEG data is recorded in file rec7. the strange thing is, that when I train the spatial filter and the classifier on rec1 and test it on rec6 it does not work at all, while it does work on rec7. Also, I can press run and then stop immediately after each other and it will work (so I don't have to complete the whole session).
No messages in the console as far as I remember, except the compression warning.
Tomorrow I will do the experiment again and send you the log files.
I had problems with the antivirus when it was updating, because the P300 flashes were irregular, so I shut it down. From now on I think I will switch all those background processes off.
Maybe my own box is causing the problem? I wrote a box to write to shared memory, so that I can send commands to my robot application (I know I should have worked with the VRPN, but I didn't know about it then). I will test this tomorrow too.
my blog
my tweets
my linkedin