Prediction And Optimization Techniques Of Ouality Of Experience For End-to-end Video Streaming In Wireless Networks

With the development of mobile communication technology, video codec technology and video compression technology, the demand of mobile video streaming is on the explosive growth thus highlights the growing importance of performance evaluation. The quality of video streaming directly affects users’perception, and also becomes the important factor for service to attract and keep users. Thus, how to measure and enhance users’feeling about end-to-end video streaming effectively, not only has become a concern of network operators, but also has become a hot topic in academic area.The traditional QoS (Quality of Service) can only measures objective network indicators, such as packet loss rate, bandwidth and so on, but these parameters cannot fully characterize whether the transmission status of wireless networks meets the users’need, while QoE (Quality of Experience) can reflect subjective feelings more accurately from end users’perspective. In this thesis, we investigate the existing QoE evaluation method, and make research on the prediction and optimization of QoE for end-to-end mobile video streaming service.According to the problem of QoE prediction for end-to-end video streaming in wireless networks, we propose a QoE prediction model based on GBDT (Gradient Boosting Decision Tree) algorithm. The proposed QoE prediction model takes cross-layer parameters from the network layer, the application layer, video content features and user equipment into account. Through the simulation test, the performance of our model is verified in prediction accuracy and time complexity, when compared to other no-reference QoE prediction models.According to the problem of QoE optimization for end-to-end video streaming in wireless networks, we propose a QoE enhancement strategy based on encoding bitrate adaptation. The video stream with a set of discrete bitrates is switching dynamically according to two control parameters, the predicted QoE value and the feedback congestion state of end-to-end network, so as to avoid or reduce end-to-end packet loss rate. Our GBDT-QoE prediction model is implemented in the video streaming server and predicts QoE in real time. Simulation results show that our proposed QoE optimization scheme can efficiently improves user-perceived quality, when compared to the scenario without adaptive bitrates.