Predictive Deployment of UAV Base Stations in Wireless Networks: Machine Learning Meets Contract Theory
In this paper, a novel framework that enables a predictive deployment of unmanned aerial vehicles (UAVs) as flying base stations (BSs) to complement ground cellular systems in face of steep surges in wireless traffic is proposed. Considering downlink communications, the goal is to efficiently offload wireless traffic from congested ground cellular networks to flying UAVs. To provide delay-free aerial service, first, a novel machine learning (ML) framework, based on wavelet decomposition and compressive sensing, is proposed to model the cellular traffic pattern. Given the predicted traffic demand, a contract matching problem is formulated to study the optimal allocation of UAVs to hotspot areas for traffic offloading. In order to employ a UAV with enough communication capacity at a reasonable price, a new approach based on contract theory is proposed that enables each BS to determine the payment charged to each employed UAV based on predicted demand while guaranteeing that each UAV truthfully reveals its communication capacity. To optimally allocate each UAV to hotspot areas, a matching approach is applied to jointly maximize the individual utility of each BS and UAV. The prediction error of the proposed ML framework is analytically derived and the results show that the proposed approach yields a performance gain of 10 prediction accuracy, and 28.3 during hotspot events, compared with the support vector machine and the weighted expectation maximization methods, respectively. Simulation results show that the proposed predictive deployment of UAVs can yield a two-fold improvement on average per BS utility and a 60 utility, respectively, compared with a baseline, event-driven allocation of UAVs to ground hotspot areas.
READ FULL TEXT