Non-Asymptotic Performance of Symmetric Slepian-Wolf Coding
Shriram Sarvotham, Dror Baron, Richard G. Baraniuk · 2005
Abstract — Like most results in information theory, the classical Slepian-Wolf theorem is asymptotic. We characterize how quickly the limits put forth by information theory can be reached. Our contribution is two-fold. First, we investigate the non-asymptotic regime for two binary sources that generate sequences with a symmetric joint distribution. Second, we provide tight converse and achievable bounds and also show how to construct codes for any rate pair within the feasible rate region. The key result is that the feasible rate region for the non-asymptotic regime is obtained by translating the corresponding rate region for the asymptotic regime. R y 1