卷积混合盲源分离中1种改进的排序模糊消除算法

A Improved Permutation Alignment Algorithm in Convolutive Blind Source Separation

频域方法可以有效地解决卷积混合盲源分离问题。针对频域方法中存在排序模糊,基于分离信号相邻频点功率谱密度的相关性较高的原理,提出1种改进的排序模糊消除算法。相比于原算法,扩展了参考频点的取值范围,同时还采用了1种置信度量方法,能够获得更准确的排序估计。仿真实验表明所提算法有效地消除了排序模糊,并且能够纠正某一频点排序的突发错误,从而降低排序错误传播的发生概率,提高卷积混合盲源分离算法的鲁棒性。

The convolutive blind source separation problem can be solved efficiently in frequencydomain. To solve the permutation ambiguity problem in frequency-domain, this paper presents a improved permutation alignment algorithm which exploits the power spectral density correlation between adjacent frequency bins of separated signals. Contrast to conventional algorithm, we extend the reference frequency bin to frequency range and consider a confidence measure, so that accurate permutation estimation can be acquired. Experimental results have verified the proposed algorithm could solve permutation ambiguity problem effectively. Furthermore, our algorithm corrected burst errors in some frequency bins, hence to minimize the spreading of the misalignment, improve the robustness of convolutive blind source separation algorithm.