Optimal Resource Allocation in NR-V2X: A Deep Reinforcement Learning approach for Aperiodic Traffic in Vehicular Networks

In the dynamic environment of New Radio Vehicle-to-Everything (NR-V2X) communications, ensuring high reliability and efficiency is crucial. The Sensing-Based Semi-Persistent Scheduling (SB-SPS) approach, defined by the 3rd Generation Partnership Project (3GPP) for V2X communication, is known to be limited in scenarios involving aperiodic traffic and variations in packet sizes. This underscores the necessity for more flexible and adaptive scheduling approaches to manage network resources effectively. We propose an enhanced resource allocation scheme utilizing a Deep Reinforcement Learning (DRL) network that dynamically adjusts key communication parameters—Modulation and Coding Scheme (MCS), Resource Reservation Interval (RRI), and Keep-Alive probability (P<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">keep</inf>). DRL’s ability to learn and adapt to unpredictable and intricate conditions makes it effective in optimizing resource reservation under diverse traffic scenarios. Our approach aims to improve the Packet Reception Ratio (PRR) and reduce Data Age (DA), specifically addressing the challenges posed by aperiodic traffic and variable packet sizes. By leveraging simulations conducted in various traffic scenarios, we validate the superiority of our adaptive method over the traditional SPS approach.