Video User QoE Perception And Prediction

Video streaming has become the main application of the Internet.In order to provide better network services for video users,operators need to understand and guarantee the userfs viewing experience.The user's viewing experience,QoE(Quality of Experience),is the user's subjective feeling of video service quality.If there is no cooperation with the end users,the operator cannot directly aware the user's subjective QoE.Even for objective QoE parameters of the application layer,such as video frame rate,bitrate,initial delay and stall,are difficult for operators to obtain.Therefore,operators often monitor network traffic to estimate the QoE of video users.However,this solution is currently facing new challenges due to the use of HTTP Adaptive Streaming(HAS)and the increasing popularity of encrypted video streams.In the HAS mechanism,the Adaptive BitRate algorithms(ABR)can adapt to network changes and select the video chunks that match the network states for transmission,which will obscure the user QoE information brought by network traffic fluctuations.This will make it impossible for the operators to judge whether the user's QoE is improved because the result of ABR adaptation or the improvement of network conditions.In addition,more and more HAS video streamings are beginning to adopt encrypted transmission services,which makes operators unable to continuly use deep parsing methods for network traffic analysis.This paper focuses on the problem of fuzzy traffic characteristics caused by the ABR algorithn on video client.At the same time,it is also noticed that the ABR algorithm uses the video chunk as the basic unit for adaptive rate adjustment.Therefore,different from the traditional methods use 'package' as the granularity,this paper proposes video traffic analyzing and mining at the granularity of video 'chunk',and then evaluates the user QoE.Specifically,the main work of this paper includes the following two parts:The first part is how to reconstruct a video chunk series from encrypted HAS video traffic.Firstly,based on the network measurement method,the transmission characteristics of the HAS encrypted video streaming are analyzed.Then,based on the analysis result of the video traffic transmission mode,a chunk series reconstruction algorithm for the HAS video streaming is proposed.The experimental results show that the root mean square error of the video chunk series reconstructed from the network traffic by our algorithm is not higher than 0.1325 which has good fitting accuracy.Since the player's states directly affect the user QoE and it can be characterized by the application layers objective QoE parameters,the second part of the work is to infer the player's states based on the characteristics of the video chunk series.Specifically,the thesis first proposes a new application layer objective QoE metric,buffer comprehensive state,to characterize the player's states.Then,using the two-way videtraffic between the HAS client and the server,the video chunk series is reconstructed,and the time series data mining technology is used to establish a prediction model between the network traffic characteristics and the new application layer objective QoE parameter for real-time estimation.The video playback state is predicted and measured in real time according to the established model,thereby achieving cross-layer preception and prediction of the video user's QoE.The experimental results show that our method can obtain better real-time prediction results.