QRAVDR: A deep Q-learning-based RSU-Assisted Video Data Routing algorithm for VANETs

Abstract

With the rapid development of Internet of Vehicles (IoV) and the increasing demand for video services, video data routing in Vehicular Ad-hoc Networks (VANETs) has become a popular research topic. Challenges such as real-time transmission demands, instability of wireless channels, and high network topology dynamics significantly affect video transmission quality. Although some related studies have used multipath transmission and priority scheduling to improve performance, they usually require accurate models or use a static approach to make decisions, which lack the learning mechanism and the ability to adapt to the dynamic network, resulting in poor video reconstruction quality. To address the above problems, A Deep Q-Learning (DQL)-based RoadSide Unit (RSU)-Assisted Video Data Routing algorithm, named QRAVDR, is proposed for urban VANET environments. The algorithm coordinates the forwarding road segments of different layers of Scalable Video Coding (SVC) video data at the RSUs through DQL, maximizing the video quality at the receiver while minimizing the transmission delay. The Neutrosophic Set Analytic Hierarchy Process method is applied to select the best relay vehicle within the road segments, which guarantees the transmission of keyframes and improves the decoding possibility. Extensive simulation experiments on QRAVDR and other existing algorithms have been conducted using NS-2 employing simulated datasets. The results show that QRAVDR achieves a better overall performance in improving the average frame delivery ratio by about 8.02%, reducing the average end-to-end delay by approximately 9.61%, and improving the average peak signal-to-noise ratio by roughly 7.97%.