Research On Blind Source Separation Algorithm Of Convolutive Mixture

In the objective environment, signal which we received through the sensor not onlycontains the original information, but also include the noise and other signals. Thus, whenthe channel and the other a priori knowledge of the source is unknown, it is tricky problemthat we estimated source signals just only from the observation signals. We call suchproblem for the blind source separation (Blind Source Separation, BSS) problem. With thedevelopment of blind source separation technology, it has have a wide range of applicationsin communication systems, voice separation, biomedical, image processing and many otherfields.According to the source signal hybrid approach, blind source separation problem canbe devided into linear mixed, convolution mixing and nonlinear mixed categories. Sincelinear mixed is simple, it has emerged many excellent algorithm. But in practice, the signalwill occur time delay during transmission, the convolution hybrid model is more practicalsignificance than the instantaneous mixture, so this article focuses on the convolution blindsource separation algorithm research.For linear mixed model, a blind separation algorithm based on the peak value and theimproved particle swarm is proposed. Improved particle swarm algorithm instead of thetraditional algorithm optimize objective junction which based on kurtosis maximization.The blind source separation simulation results for four conference speech signal verify thevalidity of algorithm. However, the algorithm processes the signal type single, and therecan be only one Gaussian signal in source. This paper presents an improved algorithmwhich based on a nonlinear function and simplified particle swarm. Nonlinear function wasused as objective function based on sourse signal types, then the simple particle swarmoptimization was uesd to optimize it. Simulation results show that the algorithm canachieve the efficient blind source separation separation for various types of source signaland containing two gaussian signals. Compared with other algorithms, the proposedalgorithm has faster convergence speed and higher separation accuracy.A new blind source separation algorithm based on peak value and particle swarmoptimization was proposed for problems convolution mixed blind source separation.reference base was applied to stratifie for extracting the source signal firstly. It can be achieve the eliminate source signals by through decorrelation operation, and it ultimatelyachieve an orderly extraction of the source signal. Computer simulation results show thatthe algorithm can be realized on BPSK, PAM and random signals convolution of mixedblind source separation.A new blind source separation algorithm based on fourth-order cross cumulant andparticle swarm optimization was proposed for problems convolution mixed blind sourceseparation. Fourth-order cross cumulant was used in the algorithm as objective function andtraditional gradient algorithm was replaced by particle swarm optimization algorithm foroptimizing the objective function. the algorithm could achieve source signal fromconvolution mixed signal. the shortcoming of being easy for gradient algorithm to fall intolocal extremum was avoided. Simulation result show that the algorithm can achieve theblind source separation for convolution mixed signal.