Efficient Mode Selection and Vehicle Pairing for Underlay V2X Networks

Device-to-device communications (D2D) have been been a key driver in supporting vehicle-to-everything (V2X) cellular communication for vehicular networks. Underlay D2D communications in which devices reuse cellular spectrum in an opportunistic manner, face several challenges such as efficient mode selection and link stability in the presence of interference. In this paper we address the problem of mode selection and vehicle pairing for a V2X network underlaying a cellular multi-tier network. To solve the problem dynamically and independently from the network, while maximizing the spectral efficiency of the radio resources, we apply a reinforcement learning approach. The scheme we propose is fully decentralized and relies on a simplified yet efficient state space. The performance of the solution is evaluated through the simulation of a two-tier network operating in an urban environment, with realistic modeling of vehicular mobility. The performance of our approach is compared to two other benchmark approaches in terms of spectral efficiency and average user data rate. Our simulation results show that our approach can improve the spectral efficiency and average data rate for 90% of the vehicles in the network, while also ensuring around 12% of improvement in terms of overall network spectral efficiency.