Enabling QoS-QoE Learning And Prediction For Real- Time Communication Based On WebRTC Video Communication In Wireless Networks

Research of wireless video real-time communication’s (RTC) quality of experience (QoE) is insights for both theory and practice. First, with the development of Internet and electronic portable devices it is becoming a killer application on mobile devices. There are specialized video RTC tools for mobile handheld devices, such as Google WebRTC, Apple Facetime, and Microsoft Skype, etc. Second, though some recent work have been conducted on measuring quality of service (QoS), it has been realized that graceful QoS does not necessarily represent QoE for end users, in particular in wireless environment, as will be shown in this paper. It is QoE, however, that ultimately determines the user-perceived service quality. Finally, research of QoE is popular in both academia and industry, and research of RTC’s QoE is still not enough, especially in some recently applications, related work is very limited.Currently, QoE research of WebRTC is still in the primary stage. Web real-time communication (WebRTC) is increasingly popular with its many advantages including install-less, open source, cross-platform, etc. To the best of our knowledge, there is no systematic study of WebRTC QoE. In this paper, we are the first to study WebRTC QoE on WiFi networks. Specifically, this paper studies the following problems:(1) This paper proposed a new QoE metric, named QoEPolar. The traditional video quality metrics, e.g., Structural SIMilarity index (SSIM) and Mean Opinion Score (MOS) cannot accurately reflect video freezing, which is found to be the QoE metric that users care the most. Besides, they cannot be obtained easily in real-time. Thus, we propose to use the time interval between two consecutively played video frames (named QoEPolar) to evaluate real-time video playback QoE. We find it can reflect video’s playback continuity and picture quality. It is also a general metric which can be easily obtained in most of real-time video systems.(2) We systematically measured and evaluated the correlation between video QoE and various wireless network quality metrics. In this paper, we design and conduct systematic and extensive measurements in an indoor WiFi environment. During the experiments, we collect two types of network QoS metrics including wireless signal/ link quality metrics and network data transfer quality metrics. And we evaluated the correlation between video QoE and various wireless network quality metrics with mathematical statistics methods.(3) We proposed two video QoE models by machine learning approach. The first one is QoE mapping model, it can be used by a user to estimate her video QoE before she initializes a video call. And the second one is QoE prediction model, it can be used for the system to adjust service strategy in real time during a video call. Experimental results based on the measurement dataset show the models are accurate. Their F1 scores are above 70%.