An Accelerated Algorithm for Adaptive Multiple Subtraction Using L1 Estimator
Lei Liu, Wenkai Lu · 2015
Summary Multiples predicted by surface-ralated multiple elimination (SRME) and many other methods have to convolve with a matching filter for the purpose of being correctly subtracted from original data. The blind separation of convolved mixtures (BSCM) method uses a L1 estimator to estimate the matching filter. The matching filter calculated by BSCM is proved performing better than least square matching filter. However, the computational complexity of BSCM is much larger than least square methods. This shortcoming limited its use when the size of the matching filter increases or the matching filter is designed a 3D filter. In this abstract, we proposed an accelerated method which is called FBSCM. This algorithm uses the iterative shrinkage-threshold strategy to decrease the complexity of the iteration procedure in BSCM. The support region of the matching filter is detected to decrease the dimension of the L1 estimator. The combination of these two strategies accelerated the BSCM method about 10-20 times of the computing time. The examples are given to show that the performance of FBSCM compared with least square matching filter and BSCM.