The Research Of Underdetermined Blind Source Separation Algorithm Based On The Enhancement Of Sparseness

Blind source separation problem is a traditional and challenging problem in signal processing domain, particularly underdetermined blind source separation has attracted the attention of many scholars At present, the methods to solve the problem of underdetermined blind source separation mainly based on the theory of sparse component analysis generally are divided into two steps: the first step is to estimate the mixing matrix, the second step is to restore and reconstruct the source signal. This paper on the basis of the in-depth research of the achievements of underdetermined blind source separation, aiming at source signals under weak sparseness, studies two kinds of mixing matrix estimation method and the source signals reconstruction algorithm based on compression sensing. In this paper, the main work is as follows.First of all, this paper introduces the basic knowledge of blind source separation model and sparse component analysis theory, expounds several kinds of sparse transformations commonly used, as well as illustrates several traditional mixing matrix estimation algorithms and the basic principle of compression sensing. On the basis of the above contents, this paper also puts forward the measures of the performance of mixing matrix estimation and the recovery of source signal.Secondly, in view of the source signals possessing weak sparseness, this paper studies two algorithms of estimating the mixing matrix. One is based on the angle of outlier detection and the fuzzy c-means clustering, the other one is time-frequency monophyletic point combined with a number of sources of the mixing matrix estimation algorithm. By removing isolated time-frequency points and extracting monophyletic points to enhance the signals’ sparseness Further and more, this paper introduces the reconstruction algorithm of source signals based on compression sensing and K-SVD dictionary according to the consistency of underdetermined blind source separation model and the compressed sensing model.Finally, simulation experiments are applied to verify these proposed methods using speech signals. The experimental results show that the all three algorithms proposed in this paper simply operate whose matrix estimation error is small and the precision of source signals’ recovery is high.