Underdetermined Blind Extraction of Sparse Sources Using Prior Information

The traditional blind source separation (BSS) usually estimates M source signals from N observed mixtures and N ges M. When there are less observed mixtures than source signals, i.e., N < M, BSS becomes a challenging underdetermined problem. So far, most of the techniques for solving the underdetermined BSS problem focus on simultaneous separation of all sparse sources. Motivated by the fact that BSS can extract only a desired source signal by using its prior information, we present a novel method for extracting a specific sparse source by using its prior information in this paper. According to three different cases of characteristics, the mixed signals are divided into multiple segments, which are then processed (such as separated using the traditional BSS) in different ways. The desired estimation is finally extracted by measuring its closeness with a reference signal constructed with prior information. The computer simulation results show the efficiency of the proposed method.