The Approximate Optimality of Simple Schedules for Half-Duplex Multi-Relay Networks

TL;DR

This paper proves that for any half-duplex multi-relay network with independent noises, an optimal relay scheduling policy with at most N+1 active states exists, simplifying the network's operation.

Contribution

It extends the conjecture of simple relay schedules being optimal from Gaussian diamond networks to all memoryless half-duplex networks with independent noises.

Findings

Optimal schedules have at most N+1 active states.

Structural properties of the optimal solution are characterized.

The proof uses submodular function minimization and linear programming techniques.

Abstract

In ISIT'12 Brahma, \"{O}zg\"{u}r and Fragouli conjectured that in a half-duplex diamond relay network (a Gaussian noise network without a direct source-destination link and with $N$ non-interfering relays) an approximately optimal relay scheduling (achieving the cut-set upper bound to within a constant gap uniformly over all channel gains) exists with at most $N+1$ active states (only $N+1$ out of the $2^N$ possible relay listen-transmit configurations have a strictly positive probability). Such relay scheduling policies are said to be simple. In ITW'13 we conjectured that simple relay policies are optimal for any half-duplex Gaussian multi-relay network, that is, simple schedules are not a consequence of the diamond network's sparse topology. In this paper we formally prove the conjecture beyond Gaussian networks. In particular, for any memoryless half-duplex $N$-relay network with independent noises and for which independent inputs are approximately optimal in the cut-set upper bound, an optimal schedule exists with at most $N+1$ active states. The key step of our proof is to write the minimum of a submodular function by means of its Lov\'{a}sz extension and use the greedy algorithm for submodular polyhedra to highlight structural properties of the optimal solution. This, together with the saddle-point property of min-max problems and the existence of optimal basic feasible solutions in linear programs, proves the claim.