Theoretic Shaping Bounds for Single Letter Constraints and Mismatched Decoding

TL;DR

This paper develops theoretical bounds on the achievable transmission rates for shaping schemes with single-letter constraints, considering both matched and mismatched decoding, and introduces novel bounds for the latter case.

Contribution

It introduces new theoretical bounds on the capacity of shaping schemes under mismatched decoding, using a random code model and novel analytical techniques.

Findings

Derived bounds for matched decoding using a modified AEP theorem.

Presented two new lower bounds on capacity under mismatched decoding.

Showed performance differences between typical and non-typical codewords in large codebooks.

Abstract

Shaping gain is attained in schemes where a shaped subcode is chosen from a larger codebook by a codeword selection process. This includes the popular method of Trellis Shaping (TS), originally proposed by Forney for average power reduction. The decoding process of such schemes is mismatched, since it is aware of only the large codebook. This study models such schemes by a random code construction and derives achievable bounds on the transmission rate under matched and mismatched decoding. For matched decoding the bound is obtained using a modified asymptotic equipartition property (AEP) theorem derived to suit this particular code construction. For mismatched decoding, relying on the large codebook performance is generally wrong, since the performance of the non-typical codewords within the large codebook may differ substantially from the typical ones. Hence, we present two novel lower bounds on the capacity under mismatched decoding. The first is based upon Gallager's random exponent, whereas the second on a modified version of the joint-typicality decoder.