A statistical wavelet processing system

Abstract

The increasing use in recent years of ‘wavelet processing’ for seismic reflection data marks a shift in emphasis from processing for a predominantly structural interpretation to processing for a stratigraphic interpretation, which requires a high resolution wavelet of more stable phase characteristics. A wavelet processing system must begin with a ‘model’ or concept of the formation of the reflection wavelets, and such models have tended to become more complex as the sophistication of wavelet processing has increased. The advanced system described here is based on a model of the recorded wavelet as the convolution of a number of component wavelets, due to the source, the earth, and the recording system. After attenuation of noise by frequency-wave number (F-K) domain processing, a pre-stack desig-nature process makes use of the whole shot record for maximum statistics in estimating the effective source wavelet, and corrects that part of the total wavelet to a broad-band, zero phase characteristic. Interrelated post-stack processes then analyse the time variant wavelet components, mostly due to absorption and transmission effects, for correction also to a zero phase characteristic. The input reflection wavelet is assumed to be minimum phase, though this may require preliminary correction processes for certain sources. Problems with this minimum phase assumption in more conventional deconvolution processes are shown to be due to their method of estimating the phase spectrum, rather than with the validity of the assumption itself. This system of wavelet processing results in a final section on which the reflection wavelets have a stable broad-band zero phase characteristic, which not only gives the interpreter maximum resolution, but is also necessary for the application of subsequent inversion processing for lithologic interpretation.