Tight Exponential Strong Converse for Source Coding Problem with Encoded Side Information

TL;DR

This paper establishes a tight strong converse exponent for the source coding problem with encoded side information, improving upon previous bounds and providing new insights into the role of the soft Markov constraint.

Contribution

The paper derives a tight strong converse exponent for the source coding problem with encoded side information, clarifying the role of the soft Markov constraint.

Findings

The derived exponent reduces to the known tight case when side information is absent.

The soft Markov constraint is shown to be a positive part of the exponent, not just a penalty.

Numerical evidence suggests the soft Markov constraint is strictly positive.

Abstract

The source coding problem with encoded side information is considered. A lower bound on the strong converse exponent has been derived by Oohama, but its tightness has not been clarified. In this paper, we derive a tight strong converse exponent. For the special case such that the side-information does not exists, we demonstrate that our tight exponent of the WAK problem reduces to the known tight expression of that special case while Oohama's lower bound is strictly loose. The converse part is proved by a judicious use of the change-of-measure argument, which was introduced by Gu-Effros and further developed by Tyagi-Watanabe. Interestingly, the soft Markov constraint, which was introduced by Oohama as a proof technique, is naturally incorporated into the characterization of the exponent. A technical innovation of this paper is recognizing that the soft Markov constraint is a part of the exponent, rather than a penalty term that should be vanished. In fact, via numerical experiment, we provide evidence that the soft Markov constraint is strictly positive. The achievability part is derived by a careful analysis of the type argument; however, unlike the conventional analysis for the achievable rate region, we need to derive the soft Markov constraint in the analysis of the correct probability. Furthermore, we present an application of our derivation of strong converse exponent to the privacy amplification.