A Practical Method for Multi-source Deblending Using Spatio-temporal Compressive Sensing

This paper presents a new method for simultaneous source deblending developed in the context of compressive sensing. The deblending solution is formulated as an inverse problem which is solved in local overlapping spatio-temporal widows extracted from the blended data. To constrain the solution, the unknown sources are assumed to have a reduced rank with minimal nuclear norm. This will promote the sparsity of seismic data without relying on the use of a given data decomposition method such as a Fourier or curvelet transform. In our case, the decomposition is data-driven which arguably would lead to better data modeling and therefore a better source separation. The proposed method is generic and can be applied to all configurations of simultaneous shooting. Test results on triple-source data show a good deblending quality which preserves the frequency content of the data after separation. The proposed method is robust to acquisition noise such as swell allowing it to be flexibly applied at early stages of a typical marine seismic processing sequence.