Spectrum Sharing and Consensus Performance of Vehicular Networks based on Deep Multi-User Reinforcement Learning.

The idea of an agent is useful for describing circumstances in which it is difficult or perhaps impossible for a single entity to gain all the necessary knowledge about the state of a system. The multiagent system is known to be useful in designing distributed solutions. Control action, data, or even both are distributed. In this research, we investigate the performance of vehicular networks based on deep multi-user reinforcement learning, where numerous V2V links use the already occupied V2I frequency spectrum, in terms of spectrum sharing and consensus. Goal is a multiuser strategy for reaching the spectrum that enhances network distributed behavior without any contact or message communication. Due to the large number of vehicles and to overcome the problem, we developed consensus and spectrum sharing algorithms based on deep multi-user reinforcement learning.