Model-Predictive Control in Communication Networks

Abstract

This dissertation consists of 8 papers, separated into 3 groups. The first 3 papers show, how model-predictive control can be applied to queueing networks and contain a detailed proof of throughput optimality. Additionally, numerous network examples are discussed, and a connection between the stability properties of assembly queues and random walks on quotient spaces is established. The next two papers develop algorithms, with which robust forecasts of delay can be obtained in queueing networks. To that end, a notion of robustness is proposed, and the network control policy is designed to meet this goal. For the last 3 papers, focus is shifted towards Age-of-Information. Two main contributions are the derivation of the distribution of the Age-of-Information values in networks with clocked working cycles and an algorithm for the exact numerical evaluation of the Age-of-Information state-space in a similar set-up.