Particle Flow Filtering Algorithm And Its Application

Particle filtering algorithm is a Monte Carlo method which is based on the sequential importance sampling,its re-sampling step leads to particle impoverishment phenomenon,so the algorithm suffers the problems of low accuracy and high computation complexity.Unlike the particle filter,particle flow filtering algorithm uses the method of particle flow substituted re-sampling to realize Bayesian estimation,improves the estimation accuracy and reduces the computational complexity.Therefore,particle flow filtering algorithm will have important theoretical significance and application value.This dissertation mainly studied about particle flow filtering algorithm and its application.Firstly,the basic theory of particle flow filter and three different ways to implement particle flow filter: quasi irrotational approximation method,parameter approximation method and weak solution form method are introduced,which are theoretical foundation for the further applications of the algorithm.Secondly,the OFDM time-varying channel estimation method based on particle filtering is studied.Traditional OFDM time-varying channel estimation algorithm based on particle filter can achieve well estimation performance,but particle filter's re-sampling will lead to the loss of diversity of particle sample set(particle impoverishment phenomenon),thereby reducing the estimation accuracy of the algorithm.According to the above problem,an OFDM time-varying channel estimation algorithm which applies particle flow filter is proposed.The algorithm uses particle flow to replace re-sampling,through constructing the differential equation to achieve Bayesian estimation and using particle flow to smoothly migrate the prior particles to its posterior distribution in state space.Thus particle updating is achieved and particle impoverishment is avoided.And a parameter approximation method is adopted to realize particle flow filter.Simulation results show that compared with channel estimation algorithm based on traditional particle filter,the proposed algorithm has higher estimation accuracy,lower computational complexity and better robustness to the environment.Thirdly,nonlinear parameter estimation method based on hybrid flow filter is researched.Particle filtering algorithm can estimate strong nonlinear system model well,but it has several problems,such as hard to select the appropriate importance distribution function,particle impoverishment,high computational complexity resulting from the increases of particles and so on.In this dissertation,the system model which consists of linear state equation and nonlinear measurement equation is considered.Then a hybrid particle flow filter algorithm which is combination of Kalman filter and particle flow filter is proposed.This algorithm first uses particleflow filter to obtain a rough estimation result and then uses Kalman filter to get the final estimation result from rough estimation,in which parameter approximation method is adopted to realize particle flow filter.Simulation results show that the proposed method has higher estimation accuracy than the standard particle flow filter algorithm,standard particle filter algorithm and mixed particle filter algorithm,the computational complexity of the proposed algorithm is nearly equals to that of the standard particle flow filter algorithm,lower than that of the standard particle filter and the mixed particle filter.Finally,the single channel signal blind separation algorithm based on particle flow filter is studied.In order to improve the performance of single channel blind source separation algorithm based on particle filter,a single channel blind source separation algorithm based on particle flow filtering algorithm is proposed.The signal separation problem is modeled as a state estimation problem,and particle flow is used to update particles,and then weak solution form method is adopted to realize particle flow filter.Simulation results show that compared with single channel blind source separation algorithm based on particle filter,the proposed algorithm has lower bit error rate and lower computational complexity.