Signal extraction via multitaper spectra of nonstationary data

Some methods for estimating time-frequency distributions and extracting common parts of nonstationary signals using multitaper, or multiple-window, methods are described. These include estimates of the generalized dual frequency spectrum, and a high-resolution spectrogram. The spectrograms have a much higher time resolution than standard spectrograms, and the corresponding time-frequency distributions have a much lower bias than conventional time-frequency distributions. The filter depends explicitly on the signal coherence between different frequencies and rejects stationary noise. These methods are illustrated with a set of bivariate seismic data.