On Constant Gaps for the Two-way Gaussian Interference Channel

TL;DR

This paper introduces the two-way Gaussian interference channel, derives new outer bounds on its symmetric sum-rate under different adaptation scenarios, and shows that simple non-adaptive schemes are nearly optimal within a constant gap.

Contribution

It provides the first outer bounds for the two-way Gaussian interference channel with complex gains and demonstrates the near-optimality of non-adaptive schemes across all regimes.

Findings

Outer bounds under full and partial adaptation are derived.

Non-adaptive schemes achieve rates within a constant gap of these bounds.

Simple Han and Kobayashi schemes are sufficient for near-optimal performance.

Abstract

We introduce the two-way Gaussian interference channel in which there are four nodes with four independent messages: two-messages to be transmitted over a Gaussian interference channel in the $\rightarrow$ direction, simultaneously with two-messages to be transmitted over an interference channel (in-band, full-duplex) in the $\leftarrow$ direction. In such a two-way network, all nodes are transmitters and receivers of messages, allowing them to adapt current channel inputs to previously received channel outputs. We propose two new outer bounds on the symmetric sum-rate for the two-way Gaussian interference channel with complex channel gains: one under full adaptation (all 4 nodes are permitted to adapt inputs to previous outputs), and one under partial adaptation (only 2 nodes are permitted to adapt, the other 2 are restricted). We show that simple non-adaptive schemes such as the Han and Kobayashi scheme, where inputs are functions of messages only and not past outputs, utilized in each direction are sufficient to achieve within a constant gap of these fully or partially adaptive outer bounds for all channel regimes.