The present masters dissertation consists of studies on seismic deconvolution where we look for otimizing the operations of smoothing, of resolution for the estimation of distribution of reflection coefficients, and of recovery of source-pulse. The studied filters are single channel, and the formulations consider the seismic trace as the result of a stationary stochastic process, and we demonstrate the effects of taper windows and of prewhitening in resolution. The applied principle is the minimization of the difference's variance between real and desired outputs, resulting on a system Wiener-Hopf normal equations whose solution is the vector of filter coefficients to be applied in a convolution. Spike deconvolution is designed considering the distribution of reflection coefficients as a white series. The operator compresses the seismic events to impulses, and its inverse is a good approximation to the source-pulse. The application of taper windows and of prewhitening improve the output of this filter. Spike-series filters are designed using the distribution of reflection coefficients. The statistical properties of the reflection coefficients distribution affect the operator and its performance. Taper windows on autocorrelation degrades the output, and improvement is achieved when it is applied to the deconvolutional operator. The Hilbert transform produces good results in the recovery of source-pulse, under the premise of minimum-phase. The inverse of the recovered source-pulse compresses the seismic events to impulses. When the seismic trace contains additive noise, the results obtained with Hilbert transform are better than with spike deconvolution. Smoothing filter suppresses noise in the seismic trace as a function of a prewhitening parameter. The use of smoothed traces improves the spike deconvolution. Double prewhitening generates better results than the simple prewhitening. The matched filter operator is obtained from maximization of a signal/noise ratio function. Deconvolving the output of the matched filter for the estimation of the distribution of reflection coefficients possess better resolution than when using a smoothing filter.
|
