Research On Joint Iterative Receiver Technology In Communication System Based On Probabilistic Graphical Model

BICM (Bit-interleaved Coded Modulation) can effectively improve the spectralefficiency for wireless data transmissions. LDPC (Low-density-parity-check) is apowerful code with low-latency and parallel computation. As a key technology in thefuture wirless communication system, LDPC-BICM takes advantage of them and getscoding gain. Nonetheless, it is necessary to estimate channel information andeliminate intersymbol interference for LDPC coded BICM-ID system with unknownchannel information or in ISI channel. How to get better performance with cost aslow as possible by exchanging extrinsic information between demodulator anddecoder is a mainly concerned problem.The iterative receiver such as iterative decoding, joint channel estimation anddecoding, joint detection and decoding are analyzed by probabilistic graphical modelin this dissertation for LDPC-BICM-ID in different channels and with different noises.The main contents are summarized as follows.1. A low-complexity adaptively belief propagation and difference-Mapalgorithm is proposed. The computational complexity can be reduced by opening thecheck node in factor graph. To make up the performance loss, one method is toadaptively adjust the normalized coefficient of check node message; the other is topresent a new selective belief propagation difference-map message update rule byborrowing the difference-map strategy for variable node messages oscillation. Thesimulation results show that the proposed algorithm has lower complexity in lowEb/N0and better performance in high Eb/N0compared to LLR-BP algorithm.2. The reason for no iteration gains in LDPC-BICM-Iterative decoding system isanalyzed. A novel iterative scheme is proposed by learning from the reweightedmethod in LDPC factor graph, and an uniformly reweighted decoding algorithm and adifferent mapping min-sum iterative receive algorithm are proposed. Different fromconventional iterative scheme, the proposed one adds modulator and multipliers indemodulator, and inputs reweighted exponential priori information into decoder. In result, the average mutual information at the output of demodulator is increased initeration and iteration gains are gotten. The theoretical proof is also given. The impactof weight on performance is dicussed and explained by variational theory, and theoptimal weight is forcasted by EXIT chart. Simulation results show that the twoalgorithms can get iteration gains in AWGN and Rayleigh fading channel.3. When flat fading channel information is unknown in LDPC-BICM-ID system,the parameteric message passing algorithm's robustness is worse and convetinaliterative scheme with particle filter has higher compliexity. A nonparametric messagepassing algorithm with MCMC (Markov chain Monte Carlo, MCMC) particle filter isproposed based on factor graph. To improve estimation accuracy and reducecomplexity, Markov Chain Monte Carlo (MCMC) max-sum message update rule isderived to be consistent with the max-sum update rule in demodulator and decoder.And on this basis, a low complexity particls message schedule is proposed byemploying mode and list of particle sequentially in iterations. The messages updatedin iteration between the demodulator and decoder are decreased in a selected updateschedule.A global sparse message update schedule is designed to reduce complexityby less channel estimation and provide more accurately extrinsic information.Estimation accuracy is improved in result. The simulation results illustrate that thejoint channel estimation and decoding algorithm has enhanced robustness, goodperformance and lower compliexity.4. Joint detection and decoding is concerned for LDPC coded SISO(single-Inputsingle-Output)/MIMO(Multiple-Input Multiple-Output)system in ISI channel. At first,two reweighted detection algorithms based on edge appearance probability and factorappearance probability are proposed to improve performance according to Markovrandom field theory. And then a joint detection and decoding algorithm based onprobabilistic graphical model is proposed by unifying LDPC decoder and detector intoreweighted framework and designing a global and decisional message update schedule.Simulation results illustrate performance improvement and complexity reduction.