Research And Implementation Of QoE Prediction Algorithms On IPTV Video

In recent years, video service over network is growing dramatically.In order to obtain users' recognitions, service providers must ensure good QoE of users. QoE comprehensively considers the influence factors from service, user and environment, which directly reflects the recognition degree of service. In order to research the relationship between QoE and the factors, QoE needs to be quantified. The metric widely adopted is Mean Opinion Score (MOS) which is aimed at Voice over IP (VoIP).Similarly, this thesis introduces Mean Opinion Score for Video (MOS_V)as the metric for IPTV video. it's not reasonable to let all the users participate in rating according to MOS V criterion. It brings great challenges to researchers that how to obtain service opinion information for service providers without collecting massive amounts of true scores.QoE prediction of IPTV video issue is proposed on this background.This thesis focuses on researches of IPTV service in two different network environment, local area network and public network. In the early days, the typical objective metrics which consist of Media Delivery Index(MDI) and Program Clock Reference (PCR) are selected based on amounts of surveys. Then the network protocols and packet structure are clearly understood through grasping the packages of network video stream. In the medium term, environment construction, metrics collection,network control, experiment design and so on are finished. At the later stage, pre-process of metric data, modeling and prediction algorithm research are completed. And two optimization algorithms are proposed to improve the prediction ability of QoE effectively during the evolutionary process of modeling.The main content of this thesis is composed of three parts. First,relative technical background is introduced. Subjective assessment metrics of QoE, subjective assessment metrics of IPTV video and data mining algorithms utilized in the experiment are screened out and introduced. Based on the metric criterion, this thesis elaborates the process of experimental environment construction, function implementation of video metrics collection, network environment manual control and experiment workflow design and optimization, respectively.Video metrics data can be quickly and effectively collected through deploying this experiment workflow automatically. After obtaining large enough scale of dataset, the pre-processing methods are applied into it.Meanwhile, adaptive rating modification algorithm and classification width adjustment algorithm are proposed,which solve the difficulties effectively. Then, the data mining algorithms are applied to the processed dataset respectively to predict QoE, and evaluate them to find the best one. At last,this thesis concludes the research work and puts forward the improvement direction of the next work.