基于密度的空间聚类与霍夫变换相结合的欠定盲源分离混合矩阵估计

Mixing matrix estimation using density based spatial clustering combined with Hough Transform in underdetermined blind source separation

为解决欠定盲源分离中混合矩阵估计问题,提出了一种基于密度的空间聚类与霍夫变换相结合的混合矩阵估计算法。该算法首先通过基于相角的单源时频点处理增强信号的稀疏性,然后针对K-means算法需预先设置聚类个数的问题,采用基于密度的空间聚类算法对单源点进行自动分类以估计源信号个数,进而估计得到混合矩阵。为提高估计混合矩阵的精度,采用霍夫变换方法修正聚类中心。基于密度的空间聚类算法的运用也克服了霍夫变换峰值簇拥问题。实验结果表明,基于密度的空间聚类与霍夫交换相结合的方法能在源信号数量未知情况下准确估计混合矩阵,且估计精度高于K-means算法和基于密度的空间聚类算法。

A method using the combination of density based spatial clustering and Hough transform is proposed to solve the mixing matrix estimation problem in underdetermined blind source separation. The method firstly conducts the single source point processing in the time-frequency domain based on phase angle to improve the signal sparsity,and then,to overcome the K-means algorithm＇ s restriction of giving the cluster number in advance, uses the density based spatial clustering algorithm to automatically classify single resource points to estimate the source signal number, and then estimates the mixing matrix. To imorove the mixing matrix estimation accuracy,the Hough transform is used to modify the clustering center. Moreover,the use of the density based spatial clustering algorithm can solve the peak clustering of Hough transform. The simulation results show that the proposed method using the density based spatial clustering and Hough transform can accurately estimate mixing matrixer when the source signal number is unknown, and its estimation accuracy is higher than the K-means method and the density based spatial clustering algorithm .