Denoising of seismic signals with oversampled filter banks

Abstract

In several applications such as denoising, when signal expansion is not crucial, oversampled filter banks may outperform critically decimated filter banks. We study the performance of the recently proposed GLPBT (Generalized Lapped Pseudo-Biorthogonal Transform), a class of oversampled filter banks, on noise removal in seismic data, using controlled redundancy. We also investigate heuristics for the choice of an optimal threshold, which appears to depend non-trivially on the noise variance. Tests indicate that carefully designed oversampled filter banks are able to outperform critically sampled ones.