Mixed Matrix Estimation Based on Sparse Structure Subspace and Density Peak Clustering

In response to issues such as the inability to automatically determine the number of clusters, low estimation accuracy, and poor robustness in underdetermined blind source separation(UBSS) methods for matrix estimation, this paper proposes a matrix estimation algorithm based on Density Peaks Clustering(DPC) and Structural Sparse Subspace Clustering(SSSC) clustering. The method preprocesses the time-domain mixed signals to enhance signal sparsity. Subsequently, it employs the SSSC algorithm to determine the number of source signals and clustering results. Finally, the DPC algorithm is utilized to determine the centroids of each cluster, thus obtaining the estimated mixed matrix. Simulation experiments on mixed speech signals validate that the proposed algorithm exhibits strong robustness and higher estimation accuracy under varying degrees of noise and different mixing scenarios.