Research On Optimization Algorithm For Blind Source Separation Based On LMS And FA

As an important branch in signal processing area,Blind Source Separation(BSS)has been focused by many scholars.Nowadays,blind source separation technology which combines with neural network,mathematical statistics and other disciplines,has been widely applied to biomedical signal processing,image signals,voice signal and other signal processing fields.In the field of instantaneous blind source separation,the traditional blind source separation algorithms is mainly focused on LMS(Least Mean Squares)algorithm,which is represented by Natural Gradient Algorithm(NGA)and EASI(Equivariant Adaptive Separation via Independence)algorithm.In practical engineering,since a part of mixed signals are composed of different kinds of signal,it is necessary to apply swarm intelligence algorithm to blind source separation or modify the traditional algorithms.However,the convergence rate of swam intelligence algorithm is slow.In this thesis,the NGA,EASI algorithm and Firefly Algorithm(FA)are studied and modified.The main work is as follows:1.In view of the performance of traditional blind source separation algorithm is restricted by fixed step size,an adaptive adjustment parameters of blind source separation method is presented.According to the traditional natural gradient algorithm,a new separation indicator is constructed by output signals to reveal the separation degree.Then,step parameters and momentum factors are adaptively updated based on the new separation indicator and chosen reasonably and accurately.In stationary and non-stationary conditions,the modified natural gradient algorithm and EASI algorithm make the received signals separated,and the separation effect is superior to the traditional algorithm.2.It is difficult for the traditional blind source separation algorithm to select activation functions when mixed signals are composed of different types of signal;the firefly algorithm falls into local minimum easily,and its convergence speed is slow.With the regard of the above problem,a kind of blind source separation optimization algorithm based on modified firefly algorithm is introduced.One hand,the modified FA adjusts the step size of the random perturbation term in position updating formula based on the degree of signal separation adaptively.Meanwhile,it takes advantage of the information of the global best solution to guide the search of candidate solutions.And the orthogonal matrix is used to reduce the complexity of the algorithm.In stationary condition,the source signal can be reconstructed from mixed signals which are composed of different kinds of signals.The modified FA solves the problem of the activation function,and the separation effect is better than the firefly algorithm.In this thesis,the LMS algorithm and firefly algorithm are studied and optimized on the basis of previous scholars' research results.And the modified algorithm is verified by mathematical analysis and certification test,which shows the proposed modified strategies correct and valid.It also has practical value and application prospect in a certain extent for blind source separation.