OpenViBE forum

Hello,

Just a couple quick questions.
1) In the BCI scenario included with OpenViBE, what effect does the Simple DSP with "log(x+1)" have on the signal? In other words, how is taking the logarithm of the signal advantageous to the operator? Also, are there similar functions that can be used to further clean the signal/make it easier to process and analyze?
2) My understanding of the "feature aggregator" module is that it takes a signal in the form of a matrix, then produces corresponding vectors that can be utilized by the Classifier/Trainer boxes. Is this correct?
3) How can an Xdawn-utilizing scenario be implemented to create a different but efficient BCI software, that could be optimized for use with prostheses?

Thank you

Hello,

asrinivasan wrote:1) In the BCI scenario included with OpenViBE, what effect does the Simple DSP with "log(x+1)" have on the signal? In other words, how is taking the logarithm of the signal advantageous to the operator? Also, are there similar functions that can be used to further clean the signal/make it easier to process and analyze?

the logarithm can reduce the effect of outliers on some classifiers like the LDA, some also say it makes it more gaussian which is beneficial for the LDA, but I do'nt know exactly why this is.
For motor-imagery you could use CSP, maybe it will be included in the next release. See this topic.

asrinivasan wrote:2) My understanding of the "feature aggregator" module is that it takes a signal in the form of a matrix, then produces corresponding vectors that can be utilized by the Classifier/Trainer boxes. Is this correct?

This is also how I see it. For the motor-imagery it is simple, the signal average box already transforms the signal matrix to an aggregate vector, so the feature aggregator just transforms it to the right type, from signal to a feature vector type.

asrinivasan wrote:3) How can an Xdawn-utilizing scenario be implemented to create a different but efficient BCI software, that could be optimized for use with prostheses?

The xDAWN algorithm is mostly used for evoked potentials and thus useful in the P300 paradgim, probably less useful for motor imagery. To control a prostheses with the P300 will be very slow, but if you use it in a goal oriented fashion it should be doable, i.e. put the intelligence in the prostheses and use the BCI for the high level control, attach to each character a certain command. Controlling a protheses with motor-imagery is probably not a good idea as it does not have enough degrees of freedom or states.

Best regards,
Dieter Devlaminck

Dear Dieter,

yes indeed, the CSP Spatial Filter box that you submitted will definitely be included in the upcoming release... Thank you for this contribution.

So asrinivasan, if you want to give it a try, it is already in the trunk of the SVN repository.

Hope this helps,
Yann