A Time-Variant Wiener Filter for Dispersed Waveforms

A time-variant Wiener filter based on the regionally dispersive characteristics of long-period surface waves is presented. The dispersion curves are determined with maximum entropy spectral analysis. The filter performance is tested with synthetic chirp waveforms and with real seismic data. The filter enhances the estimation of signals and, in particular, the surface wave magnitude measurability, for waveforms with signal-to-noise ratios down to dB RMS. For lower signal-to-noise ratios the signal estimates are less reliable, and magnitude estimates may be biased. Compared with bandpass filtering, noise rejection improvement ranges from 3 to 9 dB. The filters ability to separate multiple signals is limited by filter impulse response side lobe interference. The filter performance is furthermore limited by dispersion of noise, by non- stationary noise, by the filters inherent ability to generate false signals from broadband noise components, by the signal bandwidth, by the regional dispersion curve variation, by narrowband filter response characteristics, and by the reliability and resolution of state-of-the-art spectral analysis methods.