UBC Theses and Dissertations
Deconvolution and wavelet estimation in exploration seismology Lines, Laurence Richard
Seismic deconvolution is investigated for signals generated by explosions and mechanical vibrators. In this investigation, the seismic trace is modelled as the convolution of the earth's impulse response with a nonstationary wavelet. By use of several deconvolution techniques, the geologically interesting impulse response is estimated. Wavelet estimation is often an essential part of deconvolution. This study gives the first extensive comparison of the Wiener-Levinson, Wold-Kolmogorov, and homomorphic deconvolution wavelet estimators in exploration seismology. In this discussion, the Wold-Kolmogorov method is shown to be equivalent to Hilbert transform wavelet estimation. wavelet comparisons on noiseless synthetics indicate that homomorphic deconvolution shows considerable promise as a wavelet estimator. However, homomorphic filtering encounters difficulty when additive noise is present. Hence, cepstral stacking is used to reduce the noise problem. This thesis also investigates multichannel Wiener filters which exhibit ideal performance when the wavelets are known. Despite such ideal performance on synthetic data, the advantage of multichannel Wiener deconvolution over single channel methods becomes marginal or nonexistant when wavelet estimates are used. A case history of deconvolving explosion-generated reflection data is shown. Comparisons of deconvolutions with well log synthetics are used in order to provide-an interpretation of the earth's impulse response in the region of interest. Observed differences between synthetic seismograms and deconvolutions are interpreted in terms of the characteristics of seismic signals and velocity logs. A new approach is presented for deconvolving seismograms created by vibrator sources. Using cepstral filtering and spectral division, the wavelet portion of the trace's amplitude spectrum is removed. The undetermined portions of the impulse response's Fourier transform are filled in by autoregressive prediction. Frequency domain prediction can substantially increase the time domain resolution of seismic arrivals. This deconvolution method is particularly well suited to vibroseis data because of the phase characteristics of the vibroseis wavelet and the known bandlimited spectrum of the vibroseis signal. Such an approach obviates restrictive assumptions which are inherent in conventional approaches to vibroseis deconvolution. The usefulness of this new deconvolution method is demonstrated for synthetic and real data.
Item Citations and Data