QoE Estimation of DASH-Based Mobile Video Application Using Deep Reinforcement Learning

An increasing number of video content providers have adopted adaptive bitrate (ABR) streaming via the HTTP protocol. The client players usually run an ABR algorithm to determine the optimal quality of video playback in the next few seconds. Faced with unpredictable bandwidth variability, the latest ABR algorithm attempts to achieve the best balance between competing goals of high bitrate, less rebuffering and high smoothness. However, there is no guarantee that optimal resource utilization ensures a high quality of experience (QoE). QoE is also affected by users' preferences for video content. Even for the same movie clip, different users have varied preferences for characters, scenes, plots and other content. In this paper, we propose a Deep-Q Learning Network (DQN) based ABR algorithm to optimize the use of network and client resources in video playback and also improve QoE of users.