On Second Order Rate Regions for the Static Scalar Gaussian Broadcast Channel

TL;DR

This paper derives second order achievable and converse rate regions for the static Gaussian broadcast channel in the finite blocklength regime, highlighting the importance of superposition and rate splitting for optimal performance.

Contribution

It introduces the largest second order achievable region for the Gaussian broadcast channel, emphasizing the role of superposition, rate splitting, and optimal coding order under finite blocklength constraints.

Findings

Superposition and rate splitting are essential for optimal second order regions.

The optimal coding order differs from the infinite blocklength capacity case.

Finite blocklength penalties affect the performance of superposition coding.

Abstract

This paper considers the single antenna, static Gaussian broadcast channel in the finite blocklength regime. Second order achievable and converse rate regions are presented. Both a global reliability requirement and per-user reliability requirements are considered. The two-user case is analyzed in detail, and generalizations to the $K$-user case are also discussed. The largest second order achievable region presented here requires both superposition and rate splitting in the code construction, as opposed to the (infinite blocklength, first order) capacity region which does not require rate splitting. Indeed, the finite blocklength penalty causes superposition alone to under-perform other coding techniques in some parts of the region. In the two-user case with per-user reliability requirements, the capacity achieving superposition coding order (with the codeword of the user with the smallest SNR as cloud center) does not necessarily gives the largest second order region. Instead, the message of the user with the smallest point-to-point second order capacity should be encoded in the cloud center in order to obtain the largest second order region for the proposed scheme.