Clustering evoked potential signals using subspace methods

This work proposes a clustering technique to analyze evoked potential signals. The proposed method uses an orthogonal subspace model to enhance the single-trial signals of a session and simultaneously a subspace measure to group the trials into clusters. The ensemble averages of the signals of the different clusters are compared with ensemble averages of visually selected trials which are free of any artifact. Preliminary results consider recordings from an occipital channel where evoked response P100 wave is most pronounced.