Research On Improved Algorithms For Blind Source Separation Based On PSO And LMS

Blind source separation(Blind Source separation,BSS)is a method to recover the source signal only through the observation signal when the source signal and the transmission channel prior information are unknown.Since the actual mixed signals are mostly aliasing in the non-stationary environment,it is necessary to apply the intelligent algorithm to the field of blind source separation or to improve the traditional algorithm.However,there are some problems in terms of convergence speed and global optimization of intelligent algorithms.The improvement in traditional algorithms is mainly focused on LMS(Least Mean Squares)algorithm,which is represented by natural gradient algorithm and EASI(Equivariant Adaptive Separation via Independence)algorithm.In this paper,we will concentrate on the improved Particle Swarm Optimization(PSO)algorithm and LMS algorithm,including the following:(1)In order to overcome the disadvantages of the traditional nonlinear activation function selection and the slow convergence speed,which is easy to fall into local optimum,an adaptive inertia weight particle swarm blind source separation algorithm is studied.The algorithm adaptively adjusts the inertia weight according to the difference between the fitness of the particle swarm and eliminates the effect of the invalid iteration on the global optimization.The convergence speed and the searching ability of this algorithm is better than the particle swarm algorithm and it solves the problem of the activation function.In the stationary environment,the multi-frequency hopping signals are blindly separated with the algorithm,and the separation effect is superior to the traditional algorithm.(2)Aiming at the problem that the dynamic mixed signal is difficult to be separated in a non-stationary environment,an adaptive inertial weight particle swarm algorithm with genetic hybridization mechanism is studied.The improved particle swarm optimization algorithm can effectively improve the diversity of the particle population and avoid falling into the local optimum.At the same time,the orthogonal matrix is used to reduce the complexity of the algorithm.The dynamic mixing of the mechanical fault signals are successfully separated,and it achieves the purpose of fault detection.(3)The variable step-size blind source separation algorithm with adaptive momentum factor is studied in order to solve the problem that the traditional algorithms have a slow convergence rate and a large steady-state error and a poor signal separation performance in a non-stationary environment.On the basis of natural gradient and EASI algorithm,the PI(Performance Index)constructor is used to separate the signal evaluation index,then the paper chooses different empirical parameters to replace the step length and momentum factor,and adjust the step size and momentum factor adaptively according to the separation.The chaotic signals are blindly separated in a stationary and non-stationary environment respectively.Based on the results of peer research,new improved PSO and LMS blind source separation algorithms are proposed.The improved algorithms are explained in theory and simulation.Each algorithm has its own advantages in separation performance,and it has some reference value for the field of blind source separation.