f Improving the performance of F-X prediction filtering at low SNR

As commonly forrnulated, the method of random noise reduction by linear prediction in the f-x domain (Canales, 1984) suffers from three problems. These are (White and Sumrners, 1989): 1) The wavenumber response of the prediction filter does not peak exactly at the wavenumbers of the signal components. 2) The amplitude of the filter response is less than one at the signal component wavenurnbers, leading to attenuation of the signal. 3) Sidelobes in the filter response cause noise at wavenumbers weIl separated from the signal wavenumbers to leak through into the filtered data.