Research On Particle Filtering And Its Application In Communication

Communication systems are usually modeled as linear Gaussian dynamic ones incommunication signal processing. But actually the transmit power amplifier does not completelywork in the linear region and there exists large amount of impact noise created by the natural andartificial electromagnetic interference outside in the wireless communication environment, thus thesignal processing problem in wireless communication is actually a state estimation problem ofnon-linear and non-Gaussian systems. Since particle filtering(PF) gains significant advantages innon-linear and non-Gaussian models, the research on it is both meaningful theoretically andvaluable practically.This thesis mainly studys about the application of PF in the multiuser detection and blindseparation of chaotic signal.Firstly, starting from Bayesian estimation and Monte Carlo method, the basic principles ofparticle filtering algorithms are introduced, and the complete algorithm flow is given.Secondly, the time-varying multiuser detection based on PF algorithm is researched.Conventional multiuser detection methods usually assume that the amount of active users in thewhole system is a constant number which is commonly chosen to be equal to the maximal numberof users the system can contain. Conventional multiuser detection methods perform well under thisassumption. But the number of active users and their parameters are often time-varying in practicalmulti-access mobile communication systems, consequently the performance of the conventionalmultiuser detection methods is seriously deteriorated. Thus, it is necessary to find out a dynamicestimation method to achieve joint estimation of the number of active users, the parameters and dataof the users. In view of this problem, in this paper, the multiuser dynamic model is formed utilizingrandom set theory. When the number of active users is time varying in CDMA systems, themultiuser detection method based on Markov Mente Carlo PF is researched and the joint estimationof user's state and data is achieved. When both the number of active users and their amplitudes aretime varying in CDMA systems, the time-varying multiuser detector based on Rao-Blackwellisedparticle filter(RBPF) algorithm is proposed not only to trace the number of active users and thechange of their amplitudes but also to estimate the users' transmitted data. BER, system capacityand near-far effect of the proposed algorithm are given. Simulation results show that the proposedalgorithm performs better than conventional OMD algorithm and basic PF algorithm .Finally, blind separation based on PF for chaotic signal in dynamic environment is studied. Theconventional separation methods perform well when the number of source signals is constant. But in practice the number of source signals is time varying. Under this condition, conventionalmethods can hardly achieve blind separation. In this paper, random set theory is used to deal withthe case of time-varying of source signals number, PF tracks the number of active source signals,and then chaotic signals are blind separated. Simulation results show that the proposed algorithmcan achieve the effectively separation of chaotic signal under the condition that the number ofsource signals is time-varying.