Dynamic cross-spectral analysis of event-related EEG using ensemble averages

A method of investigating the dynamic cross-spectral properties of event-related EEG is reported. A sequence of multivariate autoregressive models is fitted to segments of the EEG within which the data exhibit local stationarity. For estimation a procedure involving ensemble averages is presented. Time courses of autospectra and coherence are derived from the multivariate models. The method is applied to event-related EEG recorded from primary and supplementary motor areas during externally triggered finger movements.