An Architecture for 3D UAV Positioning over 5GmmWave Networks

This paper presents an architecture for three-dimensional outdoor positioning of Unmanned Aerial Vehicles (UAVs) over a 5G Millimeter Wave (mmWave) network at 28 GHz, evaluated through a ray-tracing simulation of the University of Brasília School of Engineering Sciences and Technology (UnB-FCTE) campus generated with Sionna RT. Each Access Point (AP) performs local Direction of Arrival (DoA) estimation from a single Physical Resource Block (PRB) pilot snapshot and forwards only the B-bit quantized angles over the backhaul. The central processing unit then estimates the UAV position via the Extra Trees Regressor (ETR), trained offline on the ray-traced database. The architecture is benchmarked against two classical positioning baselines, angular Multilateration (MT) and an Extended Kalman Filter (EKF) with a 3D Gauss-Markov process model, on 400 Monte Carlo trajectories. Experiments sweep the four DoA algorithms: Delay and Sum (Bartlett), Minimum Variance Distortionless Response (Capon), MUltiple SIgnal Classification (MUSIC) and Estimation of Signal Parameters via Rotational Invariance Techniques (ESPRIT). The evaluation also sweeps the number of active APs from 2 to 6 (each equipped with a Uniform Planar Array (UPA)), as well as the angular quantizer bit-depth from 4 to 16 bits. The proposed regressor reaches competitive accuracy with at least 4 active APs and lower per-slot complexity than the EKF baseline, while 8-bit quantization recovers the unquantized performance and a single PRB suffices for reliable pilot-based DoA estimation.