RLFN-VRA: Reinforcement Learning-based Flexible Numerology V2V Resource Allocation for 5G NR V2X Networks

With the advancement of 5G technology, the second generation of New Radio Vehicle-to-Everything (NR V2X) based on 5G NR has been developed. NR V2X incorporates the high transmission capabilities of 5G into vehicular networks, substantially enhancing a vehicle's ability to perceive its environment and make decisions. To address the varying demands for delay and reliability in different transmission tasks within this new generation of cellular vehicular networking, NR Cellular Vehicle-to-Vehicle (C-V2V), a Reinforcement Learningbased Flexible Numerology V2V Resource Allocation (RLFNVRA) algorithm is introduced. The algorithm utilizes NR flexible physical layer parameters, adopts deep reinforcement learning to dynamically configure the subcarrier spacing, and selects spectrum resources in a distributed manner at the vehicle end. The transmission evaluation of messages with different delay requirements, reliability requirements, and traffic densities shows that the algorithm has strong convergence ability and can meet the different requirements of transmission tasks on delay and reliability. Compared with the traditional method, this algorithm can reduce the probability of message loss by at least 5%.