ESTIMATION OF THE PRIMARY SEISMIC PULSE *

A bstract A seismic trace after application of suitable amplitude recovery may be treated as a stationary time‐series. Such a trace, or a portion of it, is modelled by the expressionwhere j represents trace number on the record, t is time, α j is a time delay, α ( t ) is the seismic wavelet, s(t) is the reflection impulse response of the ground and n j is uncorrelated noise. With the common assumption that s(t) is white, random, and stationary, estimates of the energy spectrum (or auto‐correlation function) of the pulse α( t ) are obtained by statistical analysis of the multitrace record. The time‐domain pulse itself is then reconstituted under the assumption of minimum‐phase. Three techniques for obtaining the phase spectrum have been evaluated: (A) use of the Hilbert transform, (B) Use of the z ‐transform, (C) a fast method based on inverting the least‐squares inverse of the wavelets, i.e. inverting the normal time‐domain deconvolution operator. Problems associated with these three methods are most acute when the z ‐transform of α( t ) has zeroes on or near the unit circle. Such zeroes result from oversampling or from highly resonant wavelets. The behaviour of the three methods when the energy spectra are perturbed by measurement errors is studied. It is concluded that method (A) is the best of the three. Examples of reconstituted pulses are given which illustrate the variability from trace‐to‐trace, from shot‐to‐shot, and from one shot‐point medium to another. There is reasonable agreement between the minimum‐phase pulses obtained by this statistical analysis of operational records and those estimated from measurements close to the source. However, this comparison incorporates a “fudge‐factor” since an allowance for absorption has to be made in order to attenuate the high frequencies present in the pulse measured close to the shot.