Enhancing V2X Communications with UAV-mounted Reconfigurable Intelligent Surfaces

TL;DR

This work tackles reliable V2X communications by deploying a UAV-mounted RIS (DRS) to create an additional propagation path. It combines analytic trajectory planning to place the DRS at a midpoint between a V2X pair and a tabular Q-learning strategy to orient the RIS in the azimuth plane, aiming to maximize the instantaneous and cumulative throughput $R^{(n)}$ over a horizon $N$. The key contribution lies in the hybrid solution: a geometric method to optimize the 3D location and a reinforcement-learning-based mechanism to adapt RIS orientation under dynamic vehicle positions, supported by double-Q learning to reduce bias. Simulation results on a highway-like scenario show meaningful throughput gains and reduced path-loss, especially for longer inter-pair distances, underscoring the potential of UAV RIS to extend V2X coverage and reliability in dense traffic. The study also highlights practical limitations and points to future work on multi-pair interference mitigation and deep learning for concurrent servicing of multiple V2X links.

Abstract

This paper addresses the crucial need for reliable wireless communication in vehicular networks, particularly vital for the safety and efficacy of (semi-)autonomous driving amid increasing traffic. We explore the use of Reconfigurable Intelligent Surfaces (RISes) mounted on Drone Relay Stations (DRS) to enhance communication reliability. Our study formulates an optimization problem to pinpoint the optimal location and orientation of the DRS, thereby creating an additional propagation path for vehicle-to-everything (V2X) communications. We introduce a heuristic approach that combines trajectory optimization for DRS positioning and a Q-learning scheme for RIS orientation. Our results not only confirm the convergence of the Q-learning algorithm but also demonstrate significant communication improvements achieved by integrating a DRS into V2X networks.

Figures (4)

• Figure 1: Illustration of the considered scenario with communicating pairs of vehicles moving along a motorway. RIS-equipped UAV is positioned to improve the V2X communications links.
• Figure 2: Path loss of the virtual link for different 2D locations of the DRS and $h_D=500$. The V2X pair is marked by the red circles, and the orientation of the DRS is fixed. Notice the high variability of the path loss values due to the sensitivity of the sinc functions to the relations between the azimuth and elevation angles of the link.
• Figure 3: Average path-loss per cycle during the connectivity duration of V2X pairs, ending as the DRS reaches its optimal location. The use of the Q-learning method clearly improves path loss reduction.
• Figure 4: Average improvement in rate for various allowed inter-v2x pairs distances during testing (i.e., $\varepsilon=0$). As the distance increases the direct link quality reduces and the RIS LOS link is more important.