Spectrum Resourse Management Method of V2X based on Deep Reinforcement Learning

Abstract

Aiming at the problem of spectrum scarcity faced by Vehicle to Everything (V2X) communication, a deep reinforcement learning method is proposed to manage V2X spectrum resources. Firstly, the V2X communication model of a single vehicle to infrastructure link is established. Combined with the constraints such as frequency spectrum subband and transmission power, the optimization problem is constructed to maximize the comprehensive efficiency of V2X communication network. Secondly, considering the non-convexity of the optimization problem, the communication model can be regarded as a Markov decision process. Then, the Dueling-Deep Q Network (Dueling-DQN) algorithm is introduced to obtain the optimal spectrum subband selection and transmission power allocation strategy to maximize the comprehensive efficiency of V2X communication network. Finally, the simulation is carried out on tensorflow software platform to verify the performance of the proposed method. The simulation results show that Dueling-DQN algorithm can obtain higher link performance and V2X communication network efficiency compared with other algorithm.

Keywords

• V2X communication;deep reinforcement learning;spectrum allocation;transmission power distribution;V2X communication network;comprehensive efficiency