Deep Reinforcement Learning based Edge-Enabled Vehicle to Everything Service Placement for 5G Millimeter Wave.

Vehicle to Everything (V2X) communication and services are the foundation for realizing the future of Intelligent Transportation Systems (ITSs), as the potential number of users and device connections in the future is huge, in the billions, attracting many stakeholders. However, many such services have strict performance requirements, especially in terms of latency and bandwidth, and the introduction of the fifth-generation millimeter wave (5G mmWave) multi-access edge computing (MEC) can further reduce latency while increasing bandwidth. However, due to the characteristics of millimeter-wave, factors such as rain and obstacles can degrade system performance, and the available computing resources at edge nodes are limited, so it is worthwhile to study the problem of optimal V2X service placement on suitable computing nodes to meet service quality. To this end, this paper designs a model for the dynamic environment of 5G millimeter-wave communication for the problem of optimal V2X service placement (DVSP). Previous work has not considered the placement of V2X services in adverse conditions such as rain and obstacles when millimeter-wave communication is introduced. To address this problem, a deep reinforcement learning-based V2X service placement algorithm (DQN-VSPA) is developed. Simulation results show that the DVSP model successfully guarantees and maintains the QoS requirements of all different V2X services. In addition, the proposed DQN-VSPA algorithm has the lowest latency compared to the heuristic algorithm.