未知源信号个数盲信号分离的半参数统计方法

Blind source separation with an unknown number of sources -a semiparametric statistical approach

将半参数统计模型引入源信号个数未知的盲分离中,给出了源信号个数(其值n不大于观测信号的个数m)未知,混合矩阵列满秩时,盲分离半参数统计模型的估计函数,得到了由此估计函数给出的半参数统计学习算法.由于分离模型的任一分离点都是该算法的平衡点,比已有算法有更好收敛稳定性.通过计算机仿真验证了半参数统计学习算法的有效性和良好的收敛稳定性.

The problem of blind source separation (BSS) with an unknown number of sources is considered. Firstly, the semiparametric statistical approach is introduced into the BSS, and an estimating function for the semiparametric statistical approach in BSS is proposed, from which a learning rule is obtained. Each separating point is the equilibrium of the learning rule. Finally, the computer simulation shows that the learning rule obtained from the semiparametric statistical approach has better performance in convergence than the one based on the natural gradient.