Research On Blind Source Separation Algorithm Based On Fusion Momentum

After hybrid system, the source signal become mixed signal. Under the conditionof both source signal and the mixing system are unknown, only according to the knownmixed signal to separate the mixed signal to get the source signal, this technique iscalled blind source separation. Where the output signals are estimated the estimate ofthe original source signals. Although the output signals and the source signals havedifferences in the amplitude and sequence, it has no significance on the results. Blindsource separation is used widely; its application field involves the brain functionimaging field, economic metrology field, image feature extraction and neural network.After a brief introduction on the significance, the development history andresearch status at home and abroad of blind source separation technique, this papermainly studies the problem of adaptive blind source separation based on the linearinstantaneous mixture model. Especially discuss the least-mean-square algorithm andthe recursive-least-squares algorithm. Aimed at the disadvantages of the two algorithms,it put forward a new algorithm. The new algorithm is more effective. The specificcontent as follows:First, this paper introduces the basic theory of the blind source separationalgorithm, including the linear instantaneous mixing model, the detachability of mixedsignal and the uncertainty of blind source separation. In addition, it complement thebasic knowledge are used in the algorithms, including random vector and independence,gradient optimization method, evaluation index of the performance for blind sourceseparation and negative entropy theory, the centering and whitening of the pretreatment.Secondly, it introduces the definition, derivation and properties of principalcomponents for the nonlinear principal component analysis simply and describes theproblem based on the nonlinear principal component analysis. Then two kindsalgorithms of the nonlinear principal component analysis are proposed: theleast-mean-square algorithm and the recursive-least-squares algorithm. The two algorithms are compared through simulation experiments and performance analysis.Thirdly, in order to overcome the shortcomings of the traditional least-mean-squarealgorithm, it gives the least-mean-square algorithm based on fixed momentum factoraccording to the momentum technique in the neural field. So the algorithm is improvedgreatly. Because the momentum factor is fixed, a big problem still exists in thealgorithm. The problem is that the algorithm will diverge when the momentum factor istoo large, but the convergence speed is too slow if the momentum factor too large. Inorder to solve this problem, it put forward to the algorithm based on the optimizemomentum factor. The central idea is to optimize the momentum factor and get theoptimal solution of the momentum factor. So, the overall performance of the algorithmis improved. After the momentum factor is optimized, it alleviated the conflict betweenconvergence speed and steady state. Simulation results show that the performance ofnew algorithm is better than the traditional least-mean-square algorithm.Finally, this paper gives a summary about the research in blind source separationtechnology and made different level prospect for it.