Performance Studies of a Coding Scheme for a Slepian-Wolf Source Network

We investigate the simplest Slepian-Wolf source network, consisting of an information source correlated with a second source used as a side-information for the decoder but not the encoder. The information source and side-information source are modeled as binary random sequences } { }, { i i Y Y X X = = , respectively, for which the pairs ) , ( i i Y X are independentidentically distributed (i.i.d). In 1973, Slepian and Wolf characterized the optimum rate at which X can be encoded so that the decoder can recover X with the aid of Y . It is only in recent years, however, that practical coding schemes have been proposed which encode at a rate close to the SlepianWolf rate. One such coding scheme, proposed by Ramchandran et al.[2], uses a binary linear block channel code for encoding X while the decoder is furnished with the coset for each source block. We propose an alternate coding scheme, in which we also use a linear block channel code based encoder, but instead use a decoder which minimizes distortion as measure by the per-letter expected Hamming distance between source blocks and decoded source blocks. We provide the results of computer calculations indicating the distortion improvement over the Ramchandran et al.[2] coding scheme that can be expected with our coding scheme. The comparisons are done for several source models and for several linear block codes.