DRL-Driven Localization With UAV in Near-Field Communications

In this article, we propose a deep reinforcement learning (DRL)-based multipoint localization scheme (MLS) to efficiently localize Internet of Things (IoT) devices using a single autonomous aerial vehicle (AAV) equipped with a large-scale multiantenna configuration in near-field communication (NFC). By utilizing the spherical wave-based near-field steering vector, the multiantenna array on the AAV captures both the Angle of Arrival (AoA) and received signal strength indicator (RSSI) measurements from IoT devices to estimate their locations relative to the position of the AAV. This approach eliminates the need for multiple hovering points required by a single-antenna AAV (SA-AAV) or the deployment of multiple SA-AAVs. To enhance localization accuracy, key hovering points for the multiantenna AAV (MA-AAV) are strategically selected, with weights assigned based on signal strength to prioritize stronger and more reliable signals. Furthermore, DRL dynamically adjusts the position of the MA-AAV to optimize the tradeoff between localization accuracy and energy consumption. Extensive simulations conducted across rural, urban, and dense urban scenarios demonstrate that the proposed DRL-based MLS significantly improves localization accuracy while reducing the energy consumption of the AAV.