A Transmission Design via Reinforcement Learning for Delay-Aware V2V Communications

Abstract

We investigate machine-learning-based cross-layer energy-efficient transmission design for vehicular communication systems. A typical vehicle-to-vehicle (V2V) communication scenario is considered, in which the source intends to deliver two types of messages to the destination to support different safety-related applications. The first are periodically-generated heartbeat messages, and should be transmitted immediately with sufficient reliability. The second type are randomly-appeared sensing messages, and are expected to be transmitted with limited latency. Due to node mobility, accurate instantaneous channel knowledge at the transmitter side is hard to attain in practice. The transmit channel state information (CSIT) often exhibits certain delay. We propose a transmission strategy based on the deep reinforcement learning technique such that the unknown channel variation dynamics can be learned and transmission power and rate can be adaptive chosen according to the message delay status to achieve high energy efficiency. The advantages of our method over several conventional and heuristic approaches are demonstrated through computer simulations.