| Since the1980s, Separation algorithm research is becoming more and moredeeply. Especially in recent years, there are many excellent frontier algorithms, suchas SCA, the non-negative constraints imposed by means of matrix elements ofnonnegative matrix method and the evolution of the related algorithm. These methodsof blind source separation have made a great contribution to resolve this problem,especially for underdetermined blind source separation model. These new methodshave solved the underdetermined irreversible characteristic that exists in theseparation system.Typical sparse component analysis algorithm is no supervised learningalgorithms with sparse component analysis method for blind source separation, suchas linear clustering, plane clustering. And some others methods convert to TF domainfor image and sound signals. The paper works on the defects of sparse componentanalysis method, improves the algorithm and relaxes the constraints of the objectivefunction.The paper mainly does the following work. Firstly, we make a preliminaryoverview of typical algorithm and the domestic and foreign research course of blindsource separation method. Secondly, we introduce the underdetermined blind sourceseparation model of signal separation in the advanced algorithm and analyze thesolvability conditions of constraint. Finally, a method of time frequency transformdomain is given in view of the problems of signal sparse. We carried on theexperiment analysis. The simulation experiment shows that the improved algorithmrelaxes the source signal sparse requirements. Namely, no matter how much thenumber of active source in auto terms, we can separate source signals successfully, aslong as the number of source signals and the number of sensors to satisfy the relationof N2M2. At the aim of the edge of the Wigner-Will distribution characteristicsand negative problems, we apply the rule of Tikhonov constraint to solve it. The good separation effect was obtained in the experiment. |
PDF Full Text Request