An Enhanced Joint Source-Channel Decoder

Tandem coding and decoding has been demonstrated to yield arbitrarily low error rates provided a sufficiently large block length is used. When applied to practical systems that are inherently limited to a finite complexity and therefore to finite block lengths, such a strategy may be largely suboptimal. Indeed, a tandem decoding strategy ignores two types of information: the source memory and the residual redundancy of the source coder. Moreover, conventional source decoders, within a tandem decoding strategy, are designed to perform the inverse operation of the source coder and may severely decrease performance if errors are still present at their input. One viable alternative, that has been demonstrated to yield gains, is the design of a joint source-channel decoding scheme that would take the additional sources of redundancies — the source memory and the source coder’s residual redundancy — into account. In this context, we propose a novel, iterative joint source-channel decoding algorithm. The proposed scheme is derived from a Bayesian network representation of the coding chain and incorporates three types of information: the source memory; the residual redundancy of the source coder; and finally the redundancy introduced by the channel coder. Specifically, we modify an existing algorithm by first deriving a new, equivalent Bayesian network representation of the coding chain. Next, we derive a fully consistent methodology, within the framework of Bayesian networks, for performing the iterations. The proposed algorithm is shown to yield significant gains along with a drastic reduction in computational complexity when compared with the existing one. Finally, we outline additional possible improvements on the proposed algorithm. They include methods for further reductions in computational complexity at no cost in performance in one case, and at a slight cost in performance in the other.