Research On Techniques Of Blind Separation Of Co-Frequency Digital Signals Based On Particle Filtering

With the growing of communication needs and the use of massive complex communication technologies, the electromagnetic space becomes complex and crowed, co-frequency signals emerged in the third party no-cooperative reception, which brings some difficulties in signal analysis and information extraction. Blind separation algorithm based on particle filtering is an effective method for dealing with these problems. This thesis mainly talks about the single channel blind separation of digital mixed signals based on particle filtering algorithm, aimed at making some improvements of the traditional methods. Its mainly contribution involves:1. Firstly, the mathematical description of the traditional filtering problem was given in the point view of Bayesian signal processing. Secondly, the base model of two digital mixed signals was established, and the mathematical description of the single channel blind separation of digital mixed signals was given in the point view of Bayesian estimation. All of this established the theoretical foundation of the paper.2. To deal with the low estimation accuracy and rate of convergence in fixed parameters estimation of mixed digital signals using particle filtering, an improved method was proposed. By modeling the sampling distribution of the parameters as BETA distribution, the efficiency of the parameters’ sampling distribution was enhanced, and the parameter estimation accuracy and separation performance were improved. To testing the performance of the proposed algorithm, the Cramer Rao bound for the parameters’ estimation was also derived.3. In order to improve the ability of the traditional separation algorithm to adapt to the time-varying channel, a new algorithm was proposed. By modeling the time-varying parameters as the first order AR model, and choosing the prior distribution of the parameters as the sampling distribution, the traditional blind separation algorithm was successfully extended to time-varying condition. The posterior Cramer Rao bound was also derived to testing the parameters’ estimation performance of the proposed algorithm.4. To reduce computation complexity of particle filtering based single channel blind separation of digital communication signals, a novel low complexity algorithm was proposed. Firstly, Due to the particles’ searching space increased exponentially, the high computation load was analyzed. A sampling schemes were proposed, namely partial sampling, by noting that the numerical computation of the symbol’s sampling distribution is similar with the Viterbi decoding algorithm, and inspired by the M-algorithm, we only keep several branches of the whole searching space. Finally, the relation of the computation complexity of the partial sampling algorithm between smoothing length is polynomial. For example, when the smoothing length is 4, the partial s ampling’s searching space is nearly one twentieth of the traditional method. The theoretical proof was also provided.5. To improve the weak point of the partial sampling method which has a big performance lost under high SNR condition, a new sampling method named hybrid sampling was proposed. Firstly, the reason why the partial sampling method has a big performance lost under high SNR condition was analyzed. Secondly, combined the advantages of the traditional sampling method and the partial sampling method, the smoothing interval was divided into two parts, each of them select a searching method between the traditional sampling method and the partial sampling method according to the correlation of the symbol and the observed signal. The hybrid sampling method could make a good comprise between the computation complexity and the performance. Theoretical proof of the correctness of the proposed method was also given.