A Technical Study On The Communication Signal Demodulation Based On Pf

Particle filter (PF) demodulator is a new application in the signal processing domain with particle filter algorithm. The goal of research is to estimate the magnitude, carrier frequency and phase of the communication signal these three parameters by paticle filter algorithm, and achieve demodulation of the signal at last. In the theory of numerical methods, the representative point theory sampling mehod has faster convergence rate of error and better representation than Monte Carlo sampling method. When it is applied into the paticle filter, it will improve the filter's performance in small sample number situation. So it has reduced the computation load of the algrithm. The main research work in this thesis is presented as follows:First, review the research condition of communication signal demodulation, the historical development, domestic and foreign research result of the particle filter algorithm, but also the historical perspective and application direction of the representative point theory in numerical methods. And list the model of communication system, from the view point of digital communication signal, separately summarize the demodulation principle and steps of ASK signal, FSK signal, PSK signal and QAM signal. Analys the synchronization problem in detail, include carrier synchronization and symbol synchronization.The demodulation problem of communication signal in fact is the parameter estimation problem. So the PF method of parameter estimation can be applied into the demodulation problem. And propose the PF demodulation algorithm. First, from the mathematical theory foundation (including Monte Carlo sampling and recuision Bayes estimation), and take the sequential important sampling (SIS) algorithm and the sequential important resampling (SIR) algorithm as the foundation in the algorithm, derive basic step of the algorithm gradually. The key of this chapter is to apply the PF algorithm into the modulation of digital signal. Take the signal's magnitude, carrier frequency and phase these three parameter vector as the estimated state. Through the PF algorithm and calculate posteriori probability density variance change situation of estimated state each iteration time to check symbol jump time. Finally, realize the paramer estimation and signal demodulation. At last, confirm the validity of this method through simulation.The standard PF algorithm uses Monte Carlo sampling method, it exists the shortcoming of slow error convergence rate which causes the big computation load, bad time problem. Therefore apply the representative point sampling method of numerical theory method into PF to eliminate this problem. First, analyse representative point sampling have the big superiority in error convergence rate compare to Monte Carlo sampling. And explain the definition and gain methods of F-deviation representative point and MSE-deviation representative point. Then apply these good representative points sampling methods into PF algorithm to replace the Monte Carlo smpling method. They are mainly applied into the initial sampling of prior probability distribution, but also the representative sampling of Gussian kernel desity of regularized resampling in resample step. And list the detailed steps of the algorithm. At last, confirm the performance improvement of standard PF algorithm in small sample number situation through simulation. And it means the algorithm's computation load can be reduced.The improvement PF algorithm can be applied into the communication signal demodulation, therefore propose the PF demodulation algorithm based on representative point sampling method. Representative initial sample from the prior probability distribution of the signal's magnitude, carrier frequency, phase these three parameter vector, and representative sample from the Gaussian kernel density in the regularized resample algorithm. And list the steps of the signal demodulation algorithm in detail. At last, compared with the standard PF algorithm, confirm that the improved PF demodulator also has better performance in a small sample number through simulation. From another view point, it has reduced the computation load.