On the Fundamental Tradeoffs Between Video Freshness and Video Quality in Real-Time Applications

Freshness and quality are two important metrics in real-time video applications in surveillance networks. However, these two metrics conflict with each other in many cases. To enhance the performance of real-time video services, it is very important, but also challenging, to find a good tradeoff between freshness and quality. In this article, we focus on studying the tradeoff between freshness and quality, where video packets are encoded with scalable video coding (SVC) and adaptive random network coding (ARNC). Moreover, the Age of Information (AoI) is applied to model video freshness, while the video quality is modeled by the number of layers received by users. A utility function is defined to capture the tradeoff between video freshness and video quality. By maximizing the defined utilities, an excellent tradeoff between freshness and quality could be obtained. To solve the formulated optimization problem, the maximization of the utility function is characterized as a Markov decision process (MDP) problem, which is effectively solved by a heuristic algorithm and a deep Q network (DQN)-based algorithm. Simulation results indicate that the applied ARNC technique can achieve higher utility performance than other benchmark transmission techniques, and the DQN-based algorithm outperforms the heuristic algorithm in most cases.

Automated summary

Freshness and quality are important metrics in real-time video applications, but often conflict with each other. This article focuses on studying the tradeoff between freshness and quality using scalable video coding and adaptive random network coding. By maximizing defined utilities, an excellent tradeoff between freshness and quality can be obtained, effectively solved by heuristic and deep Q network algorithms.