Non-steady-state Nonlinear Signal Processing Theories And Methods In Seismic Data Analysis

This paper is consisted of twelve chapters, the main content include four parts as follow:Part â… : Theory on non-stationary and non-linear signal processing;Part â…¡: Study on separating of overlapped seismic recordings;Part â…¢: wavelet transform and artificial neural networks: its application in discrimination tonuclear explosion and earthquake;Part â…£: Application of adaptive technology in processing of well logging data.In the first part, five theories on digital signal processing are introduced. Among them, Adaptivefiltering, Wavelet transform and Wigner distribution are non-stationary method; Cepstrum andArtificial Neural Networks are nonlinear method. I try hard to make the description formathematical derivation concisely, physical concepts clearly, and make them corresponde to thefollowing applications. For example, when the Adaptive filtering is introduced, special attention ispaid to how to passed to non-stationary filtering from stationary Wiener filtering, and what thephysical sense for each item is. In addition in the introduction of wavelet transform, the conceptof multi-resolution analysis is introduced naturally by means of the Cantor set, this make someengineer to avoiding meet the not-well-known Set theory. The deep studying on method of thenon-stationary and non-linear digital signal processing, conditions for how to combine thecharacteristics of seismic data with the above methods correctly and extract information isdeveloped. Just on this basis, those calculating programs which will be used in the followingchapters are programmed. By the way, some available mathematics tools have been adopted in thestudy.The second part is aimed at separating of overlapped seismic data. Many process problemsfor overlapped records existed in seismology For instance, the multi-times fracture in epicenterfor large earthquake, overlapped explosion in the nuclear hiding, the superposed earthquake phaseof SKS and S wave emitted from interface of the centrosphere and the mantle, and so on. Thisproblem of how to separate the overlapped records is still resolved till now.From the view pointof signal processing, separating of overlapped recordings belongs to the signal reconstruction, butthe reconstruction problem is just in the initial stage, and no full-blown theory for it. Two methodsfor separating overlapped recordings have been proposed in this paper, one is the Adaptivefiltering, the other is the Wavelet Packets. In the previous one, by using the best reference inputand at certain restriction condition (restriction on strength or the initial amplitude), the superposedearthquake phase of SKS and S, the overlapped nuclear explosion of May 11, 1998 in India havebeen separated successfully. In the later one, its better localized characteristics of wavelet in timeand frequency domain synchronously has been used, the delay time of the overlapped explosionshas been determined and the high frequency component is extracted from the data.The third pat involved the discrimination of nuclear and earthquake. Many discriminationparameters have been put forward since l950's. The results show that seismological parameter is avmp importnt metheds in the discriminahng of nuclear and earthquake, but all the existedparambo are limited. Espeially in reed yCars, more efficieni seismoIogy peamAn aredeSiderated to be wOrk out fOr the Iittle and littlc equivalent as welI more and more hiding stop areused in thc toS. In thes papeL a new discrndnation mptor WaVlet Packe ComPOnen RatioMR), is brought forward. By full use of the differnce of their spectrUm vched with hmcbetWeen nuclear and earthqUak, the discrimination efficienCy is rather high for this mcthod. In the.'Synthesis discriminatiOn, the B-P neural network has been used. The ned's inPu include fourmost efficient parameters such as f m,/M.' WPCR, WD and the CepetrUm. Throughdiscriminate tO Iedia nuclcar teSt and the Tawan l999 earthquak, it shows that the...