The Application Of Particle Filter Resampling Algorithms In Blind Equalization

Because of the unique application in the non-linear and non-Gaussian field, more and more experts have put great emphasis on Particle Filter. Particle filter is an algorithm that is based on the Monte Carlo analysis and the Recursive Bayesian Estimation. Like other predictive filters, the state space can be recursively got from the measure space with the system model by using the Particle Filter. It uses the particles to describe the state space. And the discrete random estimations composed by particles and weights approximate the actual posterior state distribution. According to the algorithm, it can be updated by the iteration of the algorithm.However the particle filter has its own drawbacks. For example, the intrinsic degenerate phenomenon of the particle, the loss of the particle diversity, the selection of the importance equation. Generally speaking, we adopt the method of resampling to tackle with the degenerate phenomenon of particles. As a result, the algorithm of resampling plays an important role in the improvement of the efficiency of particle filter and the application of the field.The paper summarizes the status in quo and the significance of the research at first. And then, it introduces the fundamental theories of particle filter and the principle of the particle filter algorithm. Subsequently, it presents several particle filter resampling algorithms, such as Multinomial Resampling, Stratified Resampling, Systematic Resampling and Residual Resampling.The paper proposes the priori probability density function to substitute the optimal importance density function to resolve the degenerate problem. It also introduces a novelparticle filter resampling algorithm——descending dichotomy. From the simulation results,we can easily figure out that the new algorithm has better average performance than the others.After a brief introduction of the blind equalization, the paper proposes to apply the particle filter algorithm to the blind equalization. The paper suggests to take the advantage of the characteristic of the particle filter to replace the real parameters of the channel with the average one. This suggestion eliminates the need to sample the channel posterior distribution during the process of channel identification, reduces the complexity of the algorithm and meanwhile it helps to finish the channel identification during the process of the equalization.At last, the paper presents some simulation results and analysis of the application of the novel resampling algorithm in blind equalization. It is very obvious from the simulation results that the novel resampling algorithm has faster convergence and channel identification speed. And also, the average effect of the equalization is better than the other three resampling algorithms that are specified in the paper.