Signal Processing Parameters / Trend Control

[Chris]

Dear BCI2000 - Group !

As I am new to BCI2000-use and -development, please let me introduce myself: My name is Chris Veigl, I am studying informatics and prepare my diploma at Fortec-Institute (Technical Univesity Vienna). My work concentrates on the hard- and software design of an affordable bio/neurofeedback system that has BCI-capabilities (my software-project is called "brainBay" and does visual and audio-feedback for EEG and other biosignal data.)
The EEG-Amplifier I use is the open source project "modularEEG" which is featured by the openEEG community (http://openeeg.sf.net)

Basically, I want to try to evaluate the great results of the BCI2000-project with low-cost hardware like the modular EEG (2-6 channels, 10 bit resolution, 256 Hz sampling rate, 1uVpp-noise, DRL - circuit, rs232-interface, price for manufacturing ca. 250 USD)

After reading the design- and implementation- documents, I started to modify the code of the Random-Number EEG-Source to establish a connection to a selectable COM-Port and to parse the data coming from modularEEG in the process-function. The implementation was quite straight-forward. After some problems with the horizontal cursor-movement i read the forum-entries to fix the x-trend-control, mean and gain-settings and set up the RJB-task.
Unfortunately, my first results were only slightly above random (hopefully )

Do you think that it could be possible to get control over the cursor when using only 2 physical channels at C3 and C4 referenced to Cz ?
What setting would you recommend ?
Is trend-control essential and how is a good strategy to start with it ?

Currently I use AR-Signal Processing, 10 and 12 Hz bands in the MUD-matrix equally weighted, and no beta frequencies
How are your expieriences with the MUD in the RJB-task ?
What combinations of bands do usually work well for few channels ?

thanks for any help,
best regards,
Chris.

[gschalk]

Chris,

As mentioned previously, you should be getting decent results with only one or two channels. I recommend that, if you only have two channels, you start with referencing C3 and C4 to, say Cz. When you reference C3 and C4 to the ear, you will get a lot of spatial noise that we typically get rid of with what we call a Laplacian Montage. You should read the related articles of Dennis McFarland et al. to this topic. They should be very instructive.

In terms of the TrendControl, I would set YTrendControl to 1.

In general, it is unlikely that you will achieve decent results without analyzing and understanding the signals, and then adapting the online configuration. This is why, unfortunately, BCI research is still quite hard to do

I suggest to start with a screening procedure (using the screen.exe program) where you instruct the subject to move his/her hands (for the up target), both feet (for down targets), left hand (for the left target) and right hand (for the right target). You should also repeat this procedure with simple imaginations of the same tasks.

Then, you need to compare brain signals associated with these tasks to rest or to each other. This will highlight the locations/frequencies that change with these tasks. Again, without doing this, the chance for success are rather small. You can use the Tutorial from the BCI2000 Workshop and the Mario analysis program that comes with BCI2000 as a start. Once you find consistent differences, you can start giving feedback, instructing the subject to use the same task that produced the differences that you found in the analyses.

I hope this helps,
The Gerv