The Diversity-Multiplexing-Delay Tradeoff in MIMO Multihop Networks with ARQ

TL;DR

This paper investigates the fundamental tradeoffs between reliability, data rate, and delay in MIMO multihop networks using ARQ protocols, proposing novel designs and analyzing performance in both high and finite SNR regimes.

Contribution

It introduces a new Variable Block-Length ARQ protocol that optimally balances the diversity-multiplexing-delay tradeoff in multihop networks and characterizes finite SNR performance with delay constraints.

Findings

Optimal ARQ protocol equalizes link performance across the network.

Performance is limited by the weakest three-node sub-network.

Queueing delays influence the design of ARQ to balance error probability and delay.

Abstract

We study the tradeoff between reliability, data rate, and delay for half-duplex MIMO multihop networks that utilize the automatic-retransmission-request (ARQ) protocol both in the asymptotic high signal-to-noise ratio (SNR) regime and in the finite SNR regime. We propose novel ARQ protocol designs that optimize these tradeoffs. We first derive the diversity-multiplexing-delay tradeoff (DMDT) in the high SNR regime, where the delay is caused only by retransmissions. This asymptotic DMDT shows that the performance of an N node network is limited by the weakest three-node sub-network, and the performance of a three-node sub-network is determined by its weakest link, and, hence, the optimal ARQ protocol needs to equalize the performance on each link by allocating ARQ window sizes optimally. This equalization is captured through a novel Variable Block-Length (VBL) ARQ protocol that we propose, which achieves the optimal DMDT. We then consider the DMDT in the finite SNR regime, where the delay is caused by both the ARQ retransmissions and queueing. We characterize the finite SNR DMDT of the fixed ARQ protocol, when an end-to-end delay constraint is imposed, by deriving the probability of message error using an approach that couples the information outage analysis with the queueing network analysis. The exponent of the probability of deadline violation demonstrates that the system performance is again limited by the weakest three-node sub-network. The queueing delay changes the consideration for optimal ARQ design: more retransmissions reduce decoding error by lowering the information outage probability, but may also increase message drop rate due to delay deadline violations. Hence, the optimal ARQ should balance link performance while avoiding significant delay.