The Application Of The Gaussian Mixture Model And Fuzzy Controller Based On Self-adjusting In ICA Algorithm

Blind Signal Separation (BSS) is brought forward to solve the actual problem ofseparating and identifying some source signals from some detective signals when thesignal transmission channel specialty is unknown. As a traditional algorithm of BSS,Independent Component Analyses (ICA) algorithm is broadly used. The main work ofthis dissertation is around the ICA algorithm. Firstly, some researches anddevelopments of blind signal process (BSP) is presented systemically, and somealgorithms of BSS is concluded.In blind signal processing, the general method assume that the density functionof sources is known in advance or approximated by some parameterized functions.If the assumed density is different from the true density, the source will not beseparated. The probability density function (pdf) of source signal on ICA algorithm ishard ascertained. In according to the trouble, the Gaussian Mixture Model(GMM) isused to approximate the arbitrary density function of source, and estimate theparameters with Expectation Maximization (EM) algorithm.On ICA algorithm, choosing the step of iterative formula is great important. Theconvergence is slow if the step is much smaller, while it is easy to maladjustment ifthe step is much bigger. In the condition of on-line adaptation, it is hard to choose. Inaccording to the trouble, the Fuzzy controller based on self-adjusting factor α isbrought forward to solve the problem and designed.The results simulated by computer show that this algorithm, based on theGaussian Mixture Model (GMM) and the Fuzzy controller based on self-adjustingfactor α , can separate all sources effectively and convergent more quickly thantraditional ICA algorithm.