Universal Coding for Correlated Sources with Memory

|Universal coding problem for the system of Slepian and Wolf (the SW-system) has rst been investigated by Csisz ar and Korner. They considered the correlated memoryless sources, and established a universally attainable error exponent as a function of rate pair (R1;R2) that is positive whenever (R1; R2) is an inner point of the admissible rate region of the SW-system, which is speci ed depending on what source statistic is given. However, when the sources have their memory, the universal coding problem for the SW-system remains open, in spite of its importance. In this paper we shall deal with such a universal coding problem for the SW-system. Especially, when the sources are ergodic Markov sources, we show that there exists a sequence of universal code such that the probability of decoding error vanishes whenever (R1;R2) is an inner point of the admissible rate region. Keywords|Slepian-Wolf coding system, universal coding, correlated Markov source, information spectrum