Abstract
Separation of signal and noise is a challenging process in seismic processing, particularly for seismic data acquired in a desert environment, where the near surface is very complex and the recorded data suffers from signal scattering and ambient noise contamination. Classic transforms, such as the Fourier or the wavelet transform, have a number of limitations in representing complex seismic wavefields, because of the nonstationarity of seismic data, as they operate on a one-dimensional scale. Advanced algorithms are needed to address these challenges. In recent years, multiscale methods, such as the seislet transform, have started to be utilized by the industry to address these challenges. The seislet transform is an effective sparse multiscale wavelet-like transform specifically tailored for optimal representation of seismic data. The transform computes multiscale orthogonal basis functions, which are aligned along varying slopes or dips of seismic events in the input data.
To effectively separate signal and noise, one needs to further estimate any signal component that has leaked into the noise estimate and restore it back to the separated signal. This is achieved by an innovative and recently developed signal-and-noise orthogonalization technique. In this approach, we first estimate the signal component using one of the transforms. We then apply the orthogonalization scheme to retrieve the leaked signal energy and restore it back to the initial signal estimate.
We have applied the seislet transform and signal and noise orthogonalization to several field datasets with the aim of improving the signal-to-noise ratio, increasing the frequency bandwidth and enhancing the image quality. (C) 2019 Elsevier B.V. All rights reserved.