Statistical mechanical aspects of joint source-channel coding
arXiv:cond-mat/0303033 · doi:10.1209/epl/i2003-00466-6
Abstract
An MN-Gallager Code over Galois fields, , based on the Dynamical Block Posterior probabilities (DBP) for messages with a given set of autocorrelations is presented with the following main results: (a) for a binary symmetric channel the threshold, , is extrapolated for infinite messages using the scaling relation for the median convergence time, ; (b) a degradation in the threshold is observed as the correlations are enhanced; (c) for a given set of autocorrelations the performance is enhanced as is increased; (d) the efficiency of the DBP joint source-channel coding is slightly better than the standard gzip compression method; (e) for a given entropy, the performance of the DBP algorithm is a function of the decay of the correlation function over large distances.
6 pages