Study On Particle Filtering And Its Improved Algorithm In MC-CDMA System

In recent years, particle filter is the focus in fields of nonlinear non-Gaussian signal processing research. It is a kind of Monte Carlo method based on the Bayesian theory. By using a large number of particles and its weights, the posterior probability density distribution is approximated and the optimal estimation accuracy can be achieved when the quantity of particles is large enough. At present particle filter is widely used in target tracking, parameter estimation, system identification, navigation, positioning and other fields.Multicarrier CDMA system is one of the most promising solution in future high-speed wireless communication systems. It combines the advantages of OFDM and CDMA systems with high spectrum efficiency, strong resistance to frequency selective fading and easy to implement. In this paper, Rayleigh fading channel is simulated using first order AR model, the Alpha stable distribution is adopted as non-Gaussian noise model, so that the research of particle filter and its improved algorithms can be achieved based on multi-carrier CDMA system and its application.Particle filter has the advantages of high estimation precision and wide range of application. However, it has the problems of particle degeneracy, loss of diversity and calculation complexity. Presently two main approaches are adopted to deal with the above shortcomings:First is optimizing re-sampling technique, and the second is to choose the appropriate proposal distribution. The existing improved algorithms and strategies are summarized in this paper and then novel improved particle filter algorithms are proposed according to the two aspects.For the improvement of re-sampling techniques, firstly auxiliary particle filter, regularized particle filter and Gaussian particle filter are introduced as the improved algorithms which are all proposed to ameliorate particle degeneracy and increase particle diversity, meanwhile four common re-sampling algorithms:multinomial re-sampling, stratified re-sampling, residual re-sampling and systematic re-sampling are presented. Computational complexity and estimation accuracy of the improved algorithms and re-sampling algorithms above are compared by simulation. Drawbacks of the existing re-sampling algorithms are analysed detailedly, and then the adaptive re-sampling particle filter algorithm are proposed. Two parameters are introduced in improved re-sampling algorithm, so that computational complexity and estimation precision can be adjusted by changing the parameters. Through the simulation, calculation methods of the two parameters are given under different signal to noise ratio, consequently the parameters are adjusted adaptively when noise energy is different, in this way the adaptive re-sampling particl filter algorithm can be achieved optimal estimation accuracy. The Improved algorithm to some extent avoids the drawbacks in common re-sampling process, thereby the estimation accuracy is increased, and meanwhile, computational quantity is reduced so it is more appropriate to real time applications.For the improvements of proposal distribution, this paper summarizes the existing improved algorithms, then introduces unscented particle filter which is the most representative algorithm of improved proposal distribution. The observation noise will be approximated as Gaussian in unscented particle filter, and the mean and variance of the proposal distribution are estimated using unscented kalman filter. Because of considering the latest observations, the estimated values are closer to the posterior probability distribution, thereby estimation accuracy is increased, but the computational complexity is increased simultaneously. To solve this problem, maximum likelihood particle filter is proposed. Samely the observation noise is looked on as Gaussian distribution, and the likelihood function is constructed, while the particles are selected by using maximum likelihood method. Compared with the unscented particle filter, maximum likelihood particle filter realized to improve the estimation accuracy while reducing the computational quantity.Swarm intelligence algorithms are novel methods which are inspired in principle of biological evolution and are used to solve complex optimization problems, having the advantages of wide range of applications, high optimizing efficiency and do not need specific information, etc. Recently swarm intelligence algorithms are concerned widely by researchers in different research fields. As the member of swarm intelligence algorithm, genetic algorithm, particle swarm optimization and artificial bee colony algorithm are introduced in the paper. Combined with the advantages of particle swarm optimization and artificial bee colony algorithm, the improved artificial bee colony particle filter is proposed. Simulation results show that the improved intelligence algorithm has a faster convergence rate and is applied to optimize particle filter proposal distribution can improve estimation accuracy. However, more complicated computation is needed, that is the improved direction of swarm intelligence algorithms.