Time-variable filtering of multichannel signals using multiple windows coherence and the Weyl transform

This paper deals with noise suppression in multichannel measurements, using a certain signal model but without assuming stationarity of the signals involved. This enables application to signals whose spectral characteristics are time-variable. The novelty consists of an algorithm for obtaining an approximation of the Wiener filter for each channel. The filters are computed using the Weyl transform and estimates of the time-frequency coherence function between all channel pairs. Time-frequency coherence functions are estimated using the multiple window method, adapted to peaked spectra. Our method is evaluated on EEG signals from epileptic seizure onsets which are measured at multiple locations on the scalp. The filtered signals give improved time-frequency representations, and also the resulting filters studied in the time-frequency domain reveal otherwise not visible signal features