Distributed joint source-channel code for spatial-temporally correlated Markov sources

Abstract

A new distributed joint source-channel code (DJSCC) is proposed for a communication network with spatial-temporally correlated Markov sources. The DJSCC is performed by puncturing the information bits of a systematic linear block code but leaving the parity bits intact, and transmitting the information and parity bits with unequal energy allocations. At the receiver, the spatial data correlation is exploited with a new multi-codeword message passing (MCMP) decoding algorithm. The MCMP decoder performs decoding by exchanging information between codewords from correlated sources, whereas conventional message passing (MP) algorithms exchanges soft information only inside a codeword. The inter-codeword soft information exchange of MCMP yields additional performance gains over the MP algorithm. In recognition that the signals at the receiver are distorted observations of the Markov source and thus can be modeled by a hidden Markov model (HMM), we propose to exploit the temporal data correlation by adding a HMM decoding module to the MCMP decoder. The HMM decoder iteratively exchanges soft information with the MCMP decoder, and this results in significant performance gains over conventional systems.