Mixing Matrix Estimation in Blind Source Separation Based on CFSFDP

Abstract

Mixing matrix estimation is one of the key techniques in underdetermined blind source separation. To obtain an accurate estimation of mixing matrix, an effective estimation algorithm based on clustering by fast search and find of density peaks (CFSFDP) is proposed in this paper. First, the received signals are transformed into a Time-Frequency (TF) domain where each component have liner clustering characteristics. Then, the remaining angles are clustered by CFSFDP to classify the points belonging to the same channel. Finally, the estimation of the mixing matrix can be calculated by the results of clustering, and the number of sources can also be found. The results show that the proposed algorithm can effectively estimate the mixing matrix with high accuracy.