Blind Sparse Source Separation using Cluster Particle Swarm Optimization Technique

C.-C. Liu, T.-Y. Sun, K.-Y. Li, S.-T. Hsieh, and S.-J. Tsai (Taiwan)

Keywords

Blind source separation (BSS), sparse representation, particle swarm optimization (PSO), unknown source number, and underdetermined

Abstract

In this paper, a source number and mixing matrix identifications with Particle Swarm Optimization (PSO) are proposed for blind sparse source separation (BSS) problem which involves more sources than sensors (i.e. under-determined) and the assumption of unknown source number. We regard each particle of PSO as a probable set of mixing vectors, and modify the global item of the conventional velocity updating equation by a cluster center. After particles optimized, the existing base vectors can be extracted from the optimal particle by the proposed adaptive threshold. Then, all source signals could be recovered correctly and precisely. Validation and effectualness of the proposed algorithm are demonstrated by computer simulation examples, and its performance is compared with some existing algorithms.

Important Links:



Go Back