Mixing Matrix Estimation Based on Cluster Degree of Time-Frequency Signal for Underdetermined Blind Source Separation

To solve the mixing matrix estimation problem of underdetermined blind separation, through the study of signal linear aggregation feature in time-frequency domain, an estimation method of mixing matrix based on different signal linear aggregation degree in time-frequency domain was proposed in this paper, and focus on the estimation of mixing matrix under the signal linear aggregation degree in weaker conditions. First, the observed signal or the ratio distribution of the corresponding transformation coefficient in time-frequency domain was used to measure the degree of the signal linear aggregation; second, the improved K-means clustering algorithm was applied to estimate the mixing matrix. The proposed method reduces the requirement for signal sparsity and can estimate the mixing matrix accurately. The simulation results show that the proposed method is feasible and effective.