Blind Source Separation Based On Improved Particle Swarm Optimization Of Population Classification And Particle Concentration

Signal processing technology has become more and more important in the information era, with the deepening of the level of information and the increasing complexity of communication environment, signal processing difficulties are increased, the difficulties faced by the signal processing also increases. In practical situations, the signals what we concerned are superimposed with noise, and quite a large of them are known little about the source signal and the transmission signal, which makes blind source separation technique showed irreplaceable advantages. After20years of research and development, the blind source separation technology has achieved remarkable results both in theory and applications.The blind source separation technology combined with practical issues win the attention of researchers with its unique advantages. A lot of algorithm comes out with theirs specialty and insufficient. Based on the particle swarm optimization algorithm, the blind source separation technology are widely used with less parameters and easy to realized. But the defects of local optimum and slow convergence affect the performance.To improve the defects, this paper presents two kinds of improved particle swarm optimization algorithm and applies it to the separation. One is based on the dynamic factors and population classification mainly told us that we can use social model and cognitive model to realize the evolution of model by adjusting the learning factor, for the good and poor particles in each iteration; another algorithm is based on particle concentration and dynamic factor that according to the defects caused by the general particle swarm algorithm did not make full use of fitness differences, we use the local optimal value of weighted average about all particles instead of the local optimal about signal particle. What’s more, to avoid the algorithm being trapped in local optimal solution, the local optimization and global optimization get balance by controlling the learning factors. This paper gives a detailed principle and separation process about the algorithms that are used in blind source separation. The simulation results verify the feasibility and validity of this algorithm compare the performance about the signal to noise ratio,the similarity coefficient and the crosstalk error with the traditional algorithm that reflect the advantage of the improved algorithm accuracy and convergence speed advantage.