RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication
Artículo científico (JournalArticle)
Prescriptive Resource Management for LTE V2X Mode 4 Sidelink via Multi-Task Learning: A Multi-Objective Optimization Approach
[EN] This paper presents a new prescriptive radio resource management approach for sidelink communications in LTE V2X mode 4, improving reliability in high-vehicle-density scenarios. Given that mode 4 operates in a decentralized manner, i.e., vehicles sense and select resources autonomously, its use is essential in scenarios where there is no cellular network coverage. However, the semi-persistent resource sensing and selection mechanisms defined by Third Generation Partnership Project (3GPP) significantly degrade performance in congested scenarios, where interference and collisions increase as more vehicles attempt to transmit V2X packets. Thus, we propose a prescriptive agent that intelligently selects the optimal parameters for sidelink communication based on different scenarios. This agent makes optimization decisions based on prior learning from a Deep Neural Network (DNN) with an Multi-task learning (MTL) architecture. In turn, this MTL-DNN model has been trained using a dataset generated from an analytical model that complies with 3GPP specifications for LTE V2X communications. The results obtained show that the MTL-DNN model achieves a correlation of 99.64% with the analytical model used. Furthermore, because it is a multi-task model, it also predicts system errors that lead to different Packet Delivery Ratio (PDR) levels. In addition, the experimental results show that, by dynamically adjusting configuration parameters with the prescriptive agent, the measured PDR levels are higher than those obtained with the standard fixed configurations defined by 3GPP. Specifically, at a traffic density of beta = 0.15 veh/m, the proposed agent maintains a reliable communication level (PDR >= 0.9 ) at distances greater than 240 m, where the performance of the standard configuration typically deteriorates., This work was supported by the "Development of an Experimental Test Environment for Performance Evaluation of Wireless Networks for the Communications Laboratory of the Universidad Tecnologica Israel" under Project UISRAEL-2026-PRY-0012, and the "Time Domain Channel Sounding for Dynamic Scenarios Using Low-Cost Equipment" under Project FIEC-003-2024.