Matrioshka orthogonal matching pursuit for blended seismic source separation
Abstract
Simultaneous shooting becomes attractive in seismic data acquisition, especially when a survey has to be acquired in harsh meteorological environment or under strict environmental regulations. Despite the evident time-saving advantage, the simultaneous-source method has a considerable draw-back: the sources interfere with each other creating cross-talk in the data, which leads to significant increase of the processing complexity. Whether a preliminary deblending step (i.e. separation of signals originating from different sources) is necessary or not, remains an open question. In any case, a lot of processing sequences for simultaneous-source data start with deblending. In this paper we propose such method based on identifying coherent features in the data and classifying them according to their source of origin. Following the principle of matrioshka dolls, nested Orthogonal Matching Pursuits (OMP) are used for signal decomposition. Our parametric dictionaries are adapted to seismic data and are progressively constructed during decomposition. The method shows encouraging results on synthetic 2D datasets and is scalable to large datasets of industrial size.