GB1478842A - Suppression of multiply reflected seismic signals - Google Patents

Abstract

1478842 Seismic prospecting SEISMOGRAPH SERVICE (ENGLAND) Ltd 24 April 1974 [24 April 1973] 19353/73 Heading G1G Multiple reflection signals in digitalized seismic recordings are attentuated by taking a first estimate of the multiple reflection signals, by e.g. operating on the input traces by a Wiener prediction filter, determining time and amplitude corrections to optimize the match between the estimate and the actual recording, applying these corrections to the estimate at portions thereof at energy minima, i.e. where the squared modulus of the envelope of the multiple estimate is a minimum, and subtracting the result from the value of the input seismic recording trace. In Fig. 8, the input trace S 1 in digital form, see Fig. 6a (not shown) is fed into Wiener prediction filter 8A, 8B. Element 8A solves a series of Wiener normal equations, wherein the solution gives a prediction operator f t which is essentially a time advanced version of the input, such that when the operator f t is convolved in element 8B with the input discretely sampled trace signal S 1 the output gives a time advanced signal which is an estimate of the multiple reflection signals. As this estimate is time advanced by a time α, element 8C introduces a delay equal to this amount. To account for time varying errors in the multiple reflection signal estimate at different points in the time scale, element 8D segments the estimate signal so that corrections may be made-for each segment. Element 8D, accordingly determines where the squared modulus of the envelope of the estimate signal is a minimum by taking the estimate signal as the real part of this signal, convolving the Hilbert operator: with the real part, to give the quadrature component of the analytic signal defining the envelope, see Fig. 7 (not shown). Elements 8E calculates the required time and amplitude corrections on the segments of the estimate signal, which are correlated with the corresponding segments of the input signal S 1 . The time shift required for each estimated signal segment is inferred from the time shift required to maximize the correlation. This maximum value of crosscorrelation #m#, is divided by the zero-lag autocorrelation function ### of the relevant estimated signal segment to yield the amplitude correction factor K such that: when no primary reflection signal is present in the segment, which, it is stated, is an event of low probability. Element 8F performs the amplitude and time corrections to the estimated signal, which corrected signal is subtracted from the input signal at 8E to produce an output signal S 2 in which the multiply reflected signals are attenuated, Fig. 6D (not shown). In other embodiments, the initial estimates of the multiple reflection signals may be synthesized by retro-correlation or by a Backus type filter. The new estimate of multiple reflection signals maybe obtained for each trace of a multi-trace recording from information derived from each of the individual traces or a single estimate may be derived from one or more traces and is then applied to correct a number of input traces which may not by themselves yield sufficiently reliable estimates.