Research On Blind Source Separation Algorithm Based On Adaptive Combination Of Momentum And Dual System

Blind signal processing has become a research hotspot of signal processing in recentyears, blind signal refers to the source signal and the signal transmission channel isunknown or poorly understood. Blind source separation process is to recover the sourcesignal from blind signal according to some statistical properties of the source signal (suchas independence). It has become significance research subject in many fields, for instance,biomedicine, medical picture, voice signal processing, graphics image recognition andcommunication system, physical geography, earthquake detection, etc. So it has wideapplication prospect.Firstly, the paper reviewed the research background, development history, currentsituation of the development and the application of the actual blind source separationproblem of the and elaborated on the basis of blind source separation theory including themathematical model of blind source separation problem, basic assumption, themathematical theory and the cost function optimization criterion of the algorithm and theperformance evaluation criterion, etc.Secondly, introducing several kinds of the batch algorithm and the adaptive algorithmand simulation experiment was carried out.We focused on studying the natural gradientalgorithm, analyzed its defects and its optimization algorithm was given. For its lowconvergence speed, reference the momentum thoughts of neural networks, addingmomentum to natural gradient algorithm during the separation matrix iteration whichmade the separation performance of the algorithm have been improved to a certain extent.However, the convergence of momentum natural gradient algorithm was very sensitive tothe size of the fixed momentum factor. To solve it, we introduced the momentum that canvary adaptively with the iterations of separation algorithm. Experiments showed that thenew algorithm well eliminated the question that the selection of the fixing initial of thealgorithm, but convergence performance of the algorithm still can not break out therestriction of the step length parameter.Thirdly, in order to further mitigate and eliminate this restriction, the author designed a program that thorough combinatorial factor adaptively combined two different step sizesvariable momentum separation system on the basis of combination of theoretical. Thecombinational algorithm combines the respective advantages of the two separation system.The new algorithm can adaptively select the appropriate step to iterative update,simulation results showed that the new algorithm achieved a good separation effect.Finally, the author summarizes the main contents of this thesis and points out thebottle-neck and the future prospect of the development of blind source separationtechnology.