Dependence Balance Based Outer Bounds for Gaussian Networks with Cooperation and Feedback

TL;DR

This paper introduces new outer bounds on the capacity regions of Gaussian networks with feedback and cooperation, improving upon existing bounds and accurately reflecting the impact of noise variances in various channel models.

Contribution

The paper develops dependence balance based outer bounds for Gaussian networks with feedback and cooperation, providing tighter bounds than the cut-set bound across multiple models.

Findings

Outer bounds improve upon the cut-set bound for all non-zero feedback noise variances.

In the limit of large noise variances, bounds converge to no-feedback capacity.

Outer bounds are strictly tighter than cut-set bounds for all non-zero cooperation noise variances.

Abstract

We obtain new outer bounds on the capacity regions of the two-user multiple access channel with generalized feedback (MAC-GF) and the two-user interference channel with generalized feedback (IC-GF). These outer bounds are based on the idea of dependence balance which was proposed by Hekstra and Willems [1]. To illustrate the usefulness of our outer bounds, we investigate three different channel models. We first consider a Gaussian MAC with noisy feedback (MAC-NF), where transmitter $k$, $k=1,2$, receives a feedback $Y_{F_{k}}$, which is the channel output $Y$ corrupted with additive white Gaussian noise $Z_{k}$. As the feedback noise variances become large, one would expect the feedback to become useless, which is not reflected by the cut-set bound. We demonstrate that our outer bound improves upon the cut-set bound for all non-zero values of the feedback noise variances. Moreover, in the limit as $\sigma_{Z_{k}}^{2}\to \infty$, $k=1,2$, our outer bound collapses to the capacity region of the Gaussian MAC without feedback. Secondly, we investigate a Gaussian MAC with user-cooperation (MAC-UC), where each transmitter receives an additive white Gaussian noise corrupted version of the channel input of the other transmitter [2]. For this channel model, the cut-set bound is sensitive to the cooperation noises, but not sensitive enough. For all non-zero values of cooperation noise variances, our outer bound strictly improves upon the cut-set outer bound. Thirdly, we investigate a Gaussian IC with user-cooperation (IC-UC). For this channel model, the cut-set bound is again sensitive to cooperation noise variances but not sensitive enough. We demonstrate that our outer bound strictly improves upon the cut-set bound for all non-zero values of cooperation noise variances.