Abstract

Underdetermined blind source separation (UBSS) is a significant problem which is frequently solved by two-stage approach: estimating mixing matrix and reconstructing source signals. In this paper, an algorithm based on stagewise distributed compressed sensing (SDCS) is proposed for reconstructing sources. In reconstruction stage, the observations firstly are divided into a number of blocks, and then a reconstructing algorithm based on SDCS is applied at each block. Finally, the sources matrix is acquired from these reconstructed blocks. Compared with traditional UBSS methods, the advantages of the proposed method include satisfactory performance and low computational complexity even if the number of sources is larger than 100, as shown in numerical simulations.

Paper Details