PERFORMANCE OF THE MATCHED-PHASE NOISE-FILTER WITH ESTIMATED NOISE SPECTRA

We demonstrate properties of a new type of noise filter, the matched-phase filter, for low signal-to-noise data. It requires that the shape of the noise amplitude spectrum be approximately known. The method assumes no a priori information about the phases of the noise spectrum, nor anything about the signal, except that its spectrum be uncorrelated with that of the noise. We demonstrate algorithm performance in cases where one's knowledge of the noise spectrum is and is not precise. We also demonstrate that when the shape of the noise spectrum is known exactly, the matched-phase filter can reveal signals at a signal-to-noise ratio (SNR) of less than -100 dB and increase the SNR to order 0 dB. (C) 1995 Academic Press, Inc.