Deep Reinforcement Learning based Adaptive Transmission Control in Vehicular Networks

Abstract

Efficient transmission control is a challenging issue in vehicular networks due to the highly dynamic network environment. In this paper, we propose a Deep reinforcement learning based adaptive Transmission Scheduling Mechanism (DTSM), which is able to adaptively select different transmission control policies based on the current network status and the history data learning. In particular, we first introduce the adaptive transmission scheduling units (ATSU) in both Software-Defined Vehicular Networking (SDVN) controllers and the corresponding base stations. Based on this architecture, we formulate a mathematical model for optimal decision-making in SDVN controllers. Besides, in ATSUs, we proposed a deep Q-learning based transmission control method to dynamically adapt to the time-varying vehicular network scenarios. Simulation results verify that the proposed DTSM solution outperforms the single transmission control method of four existing benchmarks (e.g., TcpVegas, TcpBic, TcpWestwood, TcpVeno) in terms of average throughput and round-trip time.