关键词: 盲源分离（BSS）, 学习速率, 分阶段学习, 粒子群优化

Abstract: The purpose of blind source separation (BSS) is to recover the unknown source signals from their linear mixtures without the knowledge of the mixing coefficients. For real-time BSS, the learning rate has an important influence on the algorithm performance. In order to select an appropriate learning rate, this paper proposes an efficient algorithm. According to the dependence between separating signals in current timeslot, the whole signal separation process is divided into two stages: the rapid separation stage and the precise separation stage. A particle swarm optimization method is applied to the rapid separation stage to determine the learning rate, and the learning rate in the precise separation stage is decided by a piecewise function. Simulation experiments demonstrate that significant improvements of  convergence speed and stability are achieved by the proposed algorithm when compared to fixed or other adaptive learning rate methods.

Key words: blind source separation (BSS), learning rate, grading learning, particle swarm optimization