Slicing Resource Allocation in Vehicular Ad-hoc Networks

To meet the diverse QoS requirements of vehicular ad hoc networks, this paper studies the characteristics of Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) links within vehicular ad hoc networks. These services are mapped to two types of slices defined in 5G networks: V2V link services require low latency, corresponding to the uRLLC slice; V2I link services require high data transmission rates, corresponding to the eMBB slice. To address the issue of action coordination across multiple V2V links, this paper employs the MADQN resource allocation algorithm in reinforcement learning and the Markov decision process to model the problem, performing distributed resource allocation. Simulation results show that compared with MADDPG and DQN algorithms, this algorithm has good convergence and lower average loss, while significantly improving the transmission rate of V2V links in the uRLLC slice and the overall capacity of V2I links in the eMBB slice.