Research On Single-Channel Blind Signal Separation Based On Particle Filtering

Single-Channel blind signal separation (SC-BSS) is an important branch of blind signal processing, as the product of the combination of multi-discipline of statistical processing, artificial neural networks and information theory, which has important value of both theoretical and practical. In the number of bind source separation algorithm, the particle filter (PF) which break the limitations from traditional methods, become a new method to solve the problem of BSS.In this paper, we discuss the signal separation problem in nonlinear system under the framework of particle filter. We study and explore the problem of de-noising, separation of mixed-signal, and finally, a novel BSS method based on PF has been proposed. The specific of this paper can be summarized as follows:Firstly, we review the detail of basic principles of the PF frame and process of the algorithm,. For the problems that exist in the particle filter, two improvement strategies which are regular particle filtering (RPF) and particle filtering based on Markov chain Monte Carlo (PF-MCMC) were introduced. Throughout the analysis of simulation results and comparison of some algorithms, we can conclude that consideration of the estimation of the accuracy and computational complexity, PF-MCMC method has the optimal performance.Secondly, we discuss the nonlinear mixed-signal estimation problem under the framework of particle filter. According to the number of mixed-signal, problems are separated into de-noising and separation, the article analyzes the selection of importance of the function and derives the expressions of importance of weight. When the unknown parameters exist, a novel semi-blind separation is proposed. In order to solve this problem in nonlinear systems, the algorithm uses particle filter method, combined the kernel smoothing contraction technique. The simulation results show that, in nonlinear system, the PF-MCMC method can achieve the risk of multi-signal separation effectively.Thirdly, in the field of communication system, we propose a blind signal separation algorithm. Based on joint unknown parameters and mixed-signal estimation algorithm, we can establish the AR model thanks to the characterized of communication system, the issue of blind separation was changed into the joint estimation of the parameter and transformed information symbols. Simulation results show that the proposed algorithm achieve the desired effect of BSS both in Gaussian noise and non-Gaussian noise environment.