Edge Caching Based on Deep Reinforcement Learning in Vehicular Networks

Abstract

As vehicles are connected to the Internet, various services such as infotainment and automated driving can be provided. However, these services require a large amount of data download. When downloading content which has the large size, the content delivery latency can become too long to meet the constraints. To solve this problem, methods for caching the content close to the vehicles are being studied. Macro base station (MBS) and road side unit (RSU) provide storage spaces at a close distance from the vehicles and they can reduce the time required to deliver the requested content. In this paper, we propose a caching strategy in RSUs, aiming to maximize the amount of content delivered from RSUsin order to reduce the delivery latency. Besides, since RSUs are densely deployed in urban areas, RSUs can cache more content by reducing duplicate content among them. Deep deterministic policy gradient (DDPG) is adopted to decide how to cache content in RSUs. Experiments show that the proposed method not only maximizes the amount of content downloaded from RSUs, but also decreases the update cost.