基于平行因子分解的频域盲解卷积算法

针对卷积混合盲分离问题,文章提出了一张基于张量平行因子分解的盲分离算法。该算法通过将接收信号的频域相关矩阵叠加成三阶张量,再对此三阶张量进行平行因子分解,最后利用基于K-means聚类的全排列解模糊算法来完成无排列模糊的混合矩阵估计。通过仿真实验,计算分离信号与源信号的相似系数,结果表明提出的算法具有很好的分离效果,而且实现简单,可满足实际应用的要求。

A Frequency Blind Deconvolution Algorithm Based on Parallel Factor Decomposition

To solve the problem of the blind separation of convolutive separation, this paper proposes an algorithm based on tensor parallel factor decomposition （PARAFAC）. Firsdy the frequency domain correlation matrix group of the received signals is stacked in a third-order tensor. Then parallel factor decomposition of this tensor is performed. Finally estimation of the mixing matrix without frequency ambiguity is done using the all-permutations based on K-means algorithm. Simulation results show that the performance of the algorithm is provides good blind separation of convolutive mixture. Also it is relatively simple to implement, which can satisfy the demand of engineering application.