Deep Q-learning Enabled Wireless Resource Allocation for 5G Network Based Vehicle-to-vehicle Communications

Vehicle-to-vehicle (V2V) communication, a facilitator of intelligent transportation system (ITS), requires effective cooperation among vehicles in vehicular environment. The major challenge of V2V communication lies in reliable transmission and cooperation under latency conditions. This paper investigates the wireless resource allocation of 5G network based V2V communication. Network slicing is a key technology proposed in 5G network, which is used as a feature of the system model in our paper. V2V links and vehicle-to-infrastructure (V2I) links should access to different network slices. In the wireless resource allocation, it is hard to formulate the reliable transmission and latency constraints into the optimization problems. To address this problem, we use deep Q-learning to handle the resource allocation. In our framework, each V2V link is regarded as an agent. Through proper reward design and training mechanism, V2V agents successfully learn to select the sub-channel and transmission power based on the local observations. The simulation results show each V2V link can effectively satisfy the successful transmission of the message while maximizing the capacity of all V2V links.