Blind source separation of EEG data using matched filters

In this paper a new method for blind source separation of electroencephalographic (EEG) signals is proposed. In contrast to most blind source separation algorithms, our method does not employ higher order statistics. It is shown that in case of EEG signals matched filters can successfully be applied to achieve an excellent signal separation. The proposed algorithm is verified on simulated as well as on real EEG data. Due to the strong prior on the sources introduced by the matched filters, at first glance the application of our algorithm appears to be restricted to EEG data. However, by using different matched filters our method is extendible to any kind of non-white signal. Since the computations are based on second order statistics and the matched filters can be arranged in a parallel structure our algorithm is extremely fast and therefore suitable for on-line applications.