A New Implement Method for Maximizing Autocorrelation in Blind Signal Separation

Separating signal sources by maximizing the autocorrelation (also called the time delay correlation) of the interesting signal is a group of important blind source separation methods (BSS). Relying on the fact that the interesting practical sources vary more smoothly than noise, they made inspiring success in analysis of brain mapping data. However, by well investigating these algorithms, we find that they made an unnoticed assumption in maximizing the autocorrelation of signal source. Unfortunately, this assumption, which believes that the time delayed covariance matrix is symmetrical, is not well met in general. In the present study, a new method to maximize autocorrelation is proposed. It does not rely on such assumption, and shows a better performance than traditional algorithm in the compare experiment. It is also proved that, the traditional algorithms and our method are equal to each other if the assumption is met.