Protograph LDPC-Based Joint Source-Channel Coding for Markov Sources

In this paper, we investigate the protograph LDPC-based joint source-channel coding (JSCC) system for transmitting Markov sources and propose a re-design approach for source codes in this system. To sufficiently utilize the memory structure inherent within the source sequence, firstly, we propose a novel joint source-channel (JSC) decoding technique with a three-stage serially concatenated framework. Secondly, the source protograph extrinsic information transfer (PEXIT) algorithm is modified to evaluate the source decoding threshold, which can analyze the error-floor performance for this system. Finally, the conventional source codes are optimized to improve the performance on the error-floor region. Both the bit-error-rate (BER) results and the modified source PEXIT algorithm show considerable performance gains over the systems in which the temporal correlations are not utilized. Also, the optimized source codes can acquire better source decoding thresholds than traditional codes, which perform better the error-floor performance in the JSCC system.