一种小样本情况下的盲源分离算法  

Algorithm for blind separation of sources under less samples

作　　者：理华[1] 侯朝焕[1] 马晓川[1] 杨俊[1] 

机构地区：[1]中国科学院声学研究所

出　　处：《电子测量技术》2008年第6期39-43,共5页Electronic Measurement Technology

摘　　要：本文首次提出了一种小样本情况下的盲源分离算法,这种方法是在常用的自然梯度串行更新算法基础上,通过均匀递减步长并经重复迭代实现。这种算法充分利用了已有的数据,不断变换初始矩阵,达到小样本收敛的结果。这种算法不同于其他自适应步长调整算法的是不需要计算每一步迭代后信号的分离度,不需要在线调整参数,不需要经验参数。仿真结果表明:在采样点较少的情况下,该算法对于已有算法能够更快速的收敛而且算法精度较高。An algorithm for blind separation of sources under less samples is proposed for the first time. This algorithm which reduces step-size equality and repeats iterative is based on natural gradient algorithm. This algorithm can converge quickly under fewer samples by making full use of data and changing initial matrices. The difference of this algorithm from other traditional adaptive algorithm is that the iterative process can be updated without counting separating degree, without adjusting parameter and experience parameter. The simulation results show that this algorithm can converge more quickly and accurately than existing algorithm under fewer samples.

关 键 词：盲源分离 ICA 自然梯度 

分 类 号：TN911[电子电信—通信与信息系统]