关键词: 欠定盲源分离, 盲信号分离, 稀疏表示

Abstract: Solving underdetermined blind source separation (BSS) via sparse representation has become a hot spot recently. Usually mixing matrix is estimated at first, and then the source signals are estimated. It is important to implement the later estimation. In this paper, the estimation of the source signals is discussed in detail. Usually, linear programming or shortest-path technique is used. Since they are all with complex computation, they are very hard to be operated. To improve the algorithm, we propose a new algorithm by setting suitable initial value direction projection by direction projection. A steepest descent algorithm is associated to estimate the source signals. The new algorithm is easy to be implemented and with fast convergent speed. The algorithm satisfies the demand of convergent speed in BSS.

Key words: underdetermined blind source separation, Blind source separation, sparse representation