bci2000.org BBS

Hi:

Most papers talked about the data preprocessing like this: get 800 ms segments of data following each intensification for each channel-->concatenate the data segments by channel for each intensification and create a single feature vector corresponding to each stimulus.

I am afraid that there will be some data overlapping in that feature vector. For example, if there is 100 ms intensification every 175 ms (one stimulus), then there will be more than 3 intensifications in the 800 ms segment. That means the 800 ms data segment following one specific intensification is not only elicited by that specific intensification, but also influenced by the following 2 or more intensifications. In fact, I think the data should mainly be decided by the nearest intensification before it. For example, if we define that specific intensification as the 1st one, the data after 175 ms should be mainly decided by 2nd intensification (the intensification after the specific one). The data after 350 ms should be mainly decided by the 3rd intensification (intensification two after that specific one).

I don't know how BCI2000 solves this problem, or maybe it is not a problem at all. I am just curious. Thank you.

Yueqing,

This is not a problem. While you are right that epochs overlap, the only element that is timelocked with the stimulus and thus will remain after averaging several epochs is the response. Using only the 175ms after the stimulus would not work, or at least not very well, since it would disregard most of the important information in the signal (such as the P300 that has a latency of more than 175 ms).

Gerv