The Research Of Underdetermined Blind Source Separation Based On Sparse

In the process of obtaining useful information, the useful information, noise of the transmission channel, as well as other source signals are mixed together so that the observed signals are mixed signals. With the complexity of the transmission channel, the original signals could not be directly got from the sensors. So extracting the useful signals from the sensors effectively is a difficult and urgent job, which has caused great attention in the signal processing industry. In this context, blind source separation technology arises at the historic moment,this paper studies the underdetermined mixing system in line with the actual situation. Based on the theory of sparse component analysis, the basic model and classic algorithm of the underdetermined blind source separation are deeply analyzed, especially the two-stage method including mixing matrix estimation algorithm and source signal recovery algorithm. And the algorithm of two-stage to restore the original signal is improved, in order to improve application performance.In view of the large difference of the amplitude between different source signals, a new algorithm called "adjusting the clustering points" was put forward to extract this kind of signals. By analyzing the distribution characteristics of signal sampling points, the points are clustered by by using piecewise method, which estimates the mixing matrix accurately. Experimental results show that the estimation precision of the proposed algorithm is higher than K-means.For the stage of recovering source signal, the source signal estimation algorithm based on adjacent slope is put forward by analyzing the defect of minimal L1-norm method.Source signal recovery phase, based on the analysis of the defect of the minimal L1norm method, paper puts forward the source signal estimation algorithm based on adjacent slope. The sample points which really meet sparse were estimated firstly by using the relation of mixing matrix column vector and the slope of sampling points, then estimating the sample points with low sparse by the changing process of slope. The simulation results prove that this algorithm proposed is better than minimizing L1norm method, and was little influenced by the number of source signals.