基于单源点检测的欠定混合矩阵的聚类分析

Clustering analysis of underdetermined mixing matrix based on single-source-point detection

为提高欠定盲源分离的混合矩阵估计精度,提出一种单源点(SSP)检测与具有噪声的基于密度空间聚类(DBSCAN)相结合的新方法。首先,把时域的观测信号变换成时频域的稀疏信号,采用SSP检测以增强稀疏信号的线性聚类特性,并通过镜像映射把线聚类转变成数据的致密聚类;然后,利用DBSCAN搜索高密度的点并不断连接近邻数据以形成聚类簇,从而自动寻找聚类的数量和相应的聚类中心。由于每个聚类中心对应于欠定混合矩阵的一个列向量,因此采用所提出的聚类分析方法可实现对混合矩阵的估计。对音频信号的仿真结果证明,该方法能有效地提高混合矩阵的估计精度。

To improve the accuracy of mixing matrix estimation for underdetermined blind source separation ( UBSS), a new method combining single-source-point ( SSP) detection and density based spatial clustering of applications with noise ( DBSCAN) is proposed. Firstly, the observed signals in the time domain are transformed into sparse signals in the time-frequency domain, SSP detection is used to enhance the linear clustering characteristics of the sparse signals, and the linear clustering is translated into compact clustering by mirroring mapping. And then, the DBSCAN is used to search for high-density points and continuously connect the neighbor data to form clusters, it can also automatically find the number of clusters and the corresponding cluster centers. Since each cluster center corresponds to one column vector of the underdetermined mixing matrix, the proposed clustering analysis method can be used to estimate the mixing matrix. The simulation results of the audio signals show that the proposed method can effectively improve the estimation accuracy of the underdetermined mixing matrix.