Universal Wyner-Ziv coding of discrete memoryless sources with known side information statistics

Abstract

We consider a universal variant of the Wyner-Ziv problem of lossy source coding with decoder side information, where the marginal distribution of the side information is known, while the conditional distribution of the source sequence given the side information is only assumed to lie in a given parametric family. Our approach combines minimum-distance channel estimation with a recent scheme of Jalali, Verdu and Weissman based on discrete universal denoising with auxiliary information. Our scheme is universal under mild regularity conditions. We also give a concrete example of a family of sources satisfying the regularity conditions.