Denoising becomes a nontrivial task when the noise and signal overlap in multiple domains, such as time, frequency, and velocity. Fortunately, signal and noise waveforms, in general, tend to remain morphologically different and such differences can be used to separate body-wave signals from other waveforms such as ground roll and cultural noise. The key in denoising using a near-source wavelet is to find a wavelet that is a close approximation of the true source signature and remains uncontaminated by the Green’s function in any significant manner. An inverse filter designed using such a wavelet selectively compresses the body waves that can be extracted using median and low-pass filters. The overall phenomenon is explained with a synthetic example. The idea is also tested on a land data set generated using a large weight-drop source in which the wavelet recorded approximately 3 m from the source location fulfills the criteria set in our method. The results suggest that the incremental effort of recording an extra trace close to the source location during acquisition may provide previously unavailable denoising opportunities during processing, although the trace itself may be redundant for imaging.
Read full abstract