未知数量稀疏源盲分离的一种新方法

提出了一种新的用于未知数量稀疏源盲分离的统一方法。为了改善聚类分离的精度，该方法选取混合空间中半径给定的、中心位于原点的超球面以外的所有数据点，然后将这些数据点映射到中心位于原点的单位超球面上以得到集合Cy。由此，原来的聚类变为致密聚类，各聚类互相重叠的现象几乎消失。随后，先通过关于Cy的聚类分离来估计混合矩阵，再根据混合矩阵估计源，其中最佳不相似阚值和相应的聚类数量是自动生成的。计算机仿真结果验证了该方法对具有不同程度稀疏性源的有效性。当源充分稀疏时，重构信噪比大约是300dB。因此，该方法精确、便利。

A new method for blind separation of sparse sources with unknown source number

A new unified method for the blind separation of sparse sources with unknown source number is proposed. In order to improve the accuracy of clustering separation, this paper selects all datapoints outside the hypersphere of a given radius centered at origin, in mixture space, and then projects the data points onto the unit hypersphere centered at origin to obtain the aggregate Cy. After that, the clusters become compact clusters, and the phenomenon for clusters to overlap each other almost disappears. The mixing matrix is firstly estimated by clustering separation of Cy, and then the estimation of sources is made. In this way, the optimal threshold of dissimilarity and the corresponding cluster number are produced automatically. The validity of the method for the sources with varied degrees of sparsity is verified by computer simulation. When the sources are sufficiently sparse, the reconstruction SNR is about 300 dB. So the method is accurate and convenient.