The Nonlinear Blind Source Separation And Its Application Technology Research

Recently, blind source separation has become a new method widely used in thefield of medical signal processing, seismic signal processing, communication signalprocessing and so on. The mixture models of blind source separation include linearinstantaneous model, convolution model and nonlinear model, and nonlinear blindsource separation is more close to real environment. Nonlinear blind source separationhas attracted many researchers’ attention and developed quickly in recent years.Nonlinear blind source separation has a widely application in real environment, forexample, the channel or device has nonlinearity characteristic, when the signals throughthe channel or received by the receiver, will produce nonlinearity distortion. In this case,if we still use linear blind source separation algorithm, will produce a completely wrongresults.This dissertation is divided into five chapters while the main work focuses on thesecond chapter, the third chapter, and the fourth chapter. The following gives anoverview of every chapter.Chapter one is the prodrome of the whole dissertation. The research background,current status and the significance of blind source separation are introduced, and thestructure of the dissertation is given.Chapter two is the basic theory of blind source separation. First, we introduce themixture model of the blind source separation, such as linear instantaneous mixturemodel, convolution mixture model and nonlinear mixture model. The IndependentComponent Analysis, cost function and the nonlinear mixture model are especiallyintroduced and we discuss the separability of the nonlinear blind source separation. InCDMA system, the independence of the users meets the basic requirements of blindsource separation, so we can use blind source separation to solve the multi-userdetection problem. In this chapter, we propose an algorithm for multi-user detection,and verify the algorithm by the bit error rate and the estimation of the spreading code.Chapter three introduces a new method for the estimation of the score function.The cost function of MMI (Minimizing Mutual Information, MMI) can use either in linear/nonlinear blind source separation. In this chapter, the nonlinear blind sourceseparation algorithm is optimized by MMI based on specific nonlinear mixtures modelknown as post-nonlinear-linear (PNL-L). We propose a new method for the estimationof score function which can solve the problem of hard nonlinear mixture model likepost-nonlinear-linear (PNL-L).The proposed algorithm can estimate the score functionof sub-gaussian or super-gaussian by changing the parameters.Chapter four introduces a new nonlinear blind source separation algorithm,geometric post nonlinear blind source separation. The traditional nonlinear blind sourceseparation algorithm can’t be divided into two independent phase, the whole separationprocess is carried out at the same time. The gpICA algorithm is based on the geometricalgorithm to fulfill the linearization. The algorithm is divided into two phases, thelinearization and the linear blind source separation. In this algorithm, when thelinearization is finished, we can use any linear blind source separation algorithm toachieve the separation. We proposed a new algorithm based on the original algorithm.The new one is to separate the amplitude of the signals into few pieces unequally, butevery piece has the same amount of samples. And the new algorithm can compensatethe error of the original algorithm.Chapter five summarizes the work of the whole dissertation and introduces thefuture work.