Throughput Maximizing and Service Provisioning Strategies for Millimeter Wave Networks

Abstract

Migrating wireless communications to millimeter wave (mmWave) frequencies can satiate the growing demand for higher throughput and delay sensitive service provision. Due to high path loss and poor penetration, mmWave communications are challenged by limited coverage and blockage. Dense access point (AP) deployment and beamforming can enable mmWave networks to increase coverage and combat blockage. Greater AP density gives users diversity in AP selection, and thus a user can probe multiple APs to opportunistically transmit over the \emph{best} channel. Yet, the opportunity to transmit to a previously probed AP may be lost due to inherent network properties (e.g. blockage, deafness, or decentralized scheduling), and the overhead delay and cost of probing more APs may also be detrimental to the search itself.

We analyze the impact of AP diversity, beamforming, and overhead in opportunistic mmWave networks in terms of throughput and service provision capabilities. Decentralized opportunistic solutions are presented within a model that accounts for overhead delay and overhead bit cost. We assume no a priori knowledge of channel conditions, thus a deterministically optimal solution is unattainable in non-trivial scenarios. Stochastically optimal strategies are proposed and genie-aided solutions are presented as tight upper bounds. Ultimately, we present a model which includes the inherent properties of mmWave networks and obtain opportunistically optimal strategies for throughput and service provisioning.

Conditions under which throughput maximization is a sub-martingale are presented, thus the problem is approached as a finite horizon stopping problem with unreliable recall. The optimal opportunistic strategy is a set of a priori computable thresholds. Bounds on average overhead, average delay, and average performance bound of the stopping strategy are presented. The throughput performance of optimal opportunistic strategies and the impact of imperfect network measurements are evaluated via simulations.

Service provision maximization is considered under strict delay or average delay constraints. Both constraints lead to unique non-linear knapsack problems, yet reformulations into linear optimization problems are obtained by expanding the variable space. Optimal probing orders, for specific network conditions, are presented and the intuition of the optimal orders is leveraged for general network conditions. Opportunistic strategies are presented and evaluated via simulations against computationally intense stochastic programming solutions and impractical genie-aided tight upper bounds.