Orchestrating UAVs for Prioritized Data Harvesting: A Cross-Layer Optimization Perspective (2404.00961v1)

Abstract: This work describes the orchestration of a fleet of rotary-wing Unmanned Aerial Vehicles (UAVs) for harvesting prioritized traffic from random distributions of heterogeneous users with Multiple Input Multiple Output (MIMO) capabilities. In a finite-horizon offline setting, the goal is to optimize the beam-forming design, the 3D UAV positioning and trajectory solution, and the user association/scheduling policy, to maximize the cumulative fleet-wide reward obtained by satisfying the quality-of-service mandates imposed on each user uplink request, subject to an average per-UAV mobility power constraint. With a probabilistic air-to-ground channel model, a multi-user MIMO uplink communication model with prioritized traffic, and a novel 3D mobility model for rotary-wing UAVs, the fleet-wide reward maximization problem is solved via a cross-layer optimization framework: first, K-means clustering is employed to obtain user clusters; then, equipped with a zero-forcing beam-forming design, the positions of the UAVs are optimized via two-stage grid search; next, treating these optimal positions as the graph vertices of a fully-connected mesh, the 3D UAV trajectories (i.e., graph edges) are designed via a learning based competitive swarm optimization algorithm, under an average UAV power consumption constraint, coupled with projected subgradient ascent for dual optimization; consequently, the user association/scheduling strategy is solved via a graphical branch-and-bound method on the underlying multiple traveling salesman problem. Numerical evaluations demonstrate that the proposed solution outperforms static UAV deployments, adaptive Voronoi decomposition techniques, and state-of-the-art iterative fleet control algorithms, with respect to user quality-of-service and per-UAV average power consumption.