Adaptive approaches to signal enhancement and deconvolution (with particular reference to reflection seismology)

Deconvolution and signal enhancement are important aspects of digital<br/>signal processing. Many techniques have been developed to achieve these<br/>twin aims, the vast majority however were designed to deal with stationary<br/>signals. However, many practically occurring signals are significantly<br/>non-stationary and these techniques are rendered at least partially<br/>ineffective.<br/><br/>This thesis is devoted to the study of adaptive techniques, these form a<br/>class of methods which are specifically designed to give the flexibility<br/>to deal with non-stationarity. The thesis demonstrates the value of<br/>adaptive approaches to problems of deconvolution and signal enhancement,<br/>particularly in reflection seismology. Adaptive processes are divided<br/>into two classes - modelled and empirical. The power of both these<br/>approaches is demonstrated by concentrating primarily on one algorithm of<br/>each class. In the case of the modelled approach the technique chosen is<br/>a recent approach to deconvolution based on the methods of optimal control.<br/>The method is redeveloped in discrete-time, the theory is extended to<br/>include the important problem of noise reduction in deconvolution, and for<br/>the first time, the method is applied to physical problems. The principal<br/>application is to the deconvolution of seismic data incorporating both<br/>stationary and non-stationary models. A second application is to the<br/>deconvolution of data derived from a velocity meter.<br/><br/>The empirical approach to adaptive processing is illustrated by the so-called<br/>LMS (least mean-square) algorithm. The theory of this method is<br/>rationalised and extended both for broadband inputs, particularly for the<br/>important area of non-stationary random processes, and for narrowband inputs.<br/>Two new configurations of the LMS algorithm are introduced for signal<br/>enhancement. One, dubbed the generalised comb filter is designed for the<br/>enhancement of signals which may be considered to consist of a series of<br/>slowly time-varying wavelets of unknown form, recurring at roughly constant<br/>intervals and embedded in random noise with unknown properties. The theory<br/>of this method is developed and the technique is applied to the enhancement<br/>of voiced speech and to the enhancement of seismic signals. This seismic<br/>enhancement has two forms - one for highly reverberant single-channel<br/>seismic data, and the other for enhancing multi-channel data. The second<br/>novel configuration of the LMS is in the form of a sparse adaptive filter,<br/>that is one with relatively few coefficients in relation to its length,<br/>with the objective of signal enhancement by cancellation of multiple<br/>interfering sinusoids. This technique is also applied to the problem of<br/>speech enhancement.

Name dc:type.qualificationname

Ph.D.

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of Southampton

Year dc:date.issued

1983

Author dc:creator

• Clarkson, Peter Martin

Advisor dc:contributor.advisor

• Hammond, J.K.