摘要
分析了超定盲源分离中的自然梯度算法最终不能稳定收敛的原因,针对解决这一问题的方法中存在的不足进行了分析和研究。采用了一种基于分离矩阵的步长自适应在线盲源分离算法,较好地实现了收敛速度与稳态误差的最优结合。同时,在信号随机减少或增加时改进算法也能够达到较好的分离效果,仿真结果验证了改进算法的收敛稳定性与分离有效性。
The paper analyzes first the reason that natural gradient algorithm of overdetermined BSS can not stably converge eventually. Then,the problems of exiting methods are analyzed and studied. Finally,the on-line BSS algorithm with adaptive step length based on separating matrix is presented to realize the optimum combination between convergence speed and steady-state error. At the same time,the algorithm can achieve a better separating result when the signal is randomly reduced or increased. The simulation result verifies the convergence stability and the separating effectiveness of the two improved algorithms.
出处
《无线电工程》
2012年第6期28-31,共4页
Radio Engineering
关键词
分离矩阵
变步长
自适应
盲源分离
separating matrix
variable step length
adaptive
blind source separation
