3D Deployment with Machine Learning and System Performance Analysis of UAV-Enabled Networks

Exploring the base station (BS) placement in both horizontal and vertical directions is beneficial but challenging for unmanned aerial vehicle (UAV)-enabled wireless network. In this paper, we propose a three dimensional (3D) deployment approach for UAVs and analyze the system performance of finite UAV-enabled networks in which UAVs are equipped with BS. By modeling UAVs as a deep reinforcement learning (DRL) agent, we propose a novel framework to deploy UAVs in 3D space to maximize the network utility. Then utilizing tools from stochastic geometry, we model the locations of UAVs as binomial point process (BPP) and derive exact expressions of coverage probability for directional antennas and omnidirectional antennas equipped UAVs. The expressions are functions of UAVs' altitudes and sector angles. The analysis is meaningful for setting UAVs' altitude and sector angle of directional antennas. Simulation results show that 3D deployment of UAVs achieves a remarkable system performance and the analysis provides useful performance trends.