Research On The Memory Effect-based QoE Evaluation For HTTP Adaptive Streaming Service

With the rapid development of Information and Communication Technology,the mobile streaming media service has gained popularity.Among various wireless multimedia solutions,HTTP Adaptive Streaming(HAS)is developing rapidly and is expected to gain further popularity within the next few years.HAS can dynamically adapt the video to capabilities of the end user device and conditions of current network so as to give an influent playback and result in a higher QoE generally.Quality of Experience(QoE)is an end to end concept,which describes the overall acceptability of an application or service,as perceived subjectively by the end-user.HAS technology aims at improving the video user QoE,so the evaluation of HAS service has gained much concern by the academics and industry.The QoE of HAS is mainly influenced by factors in three layers,i.e.the network layer,application layer and user layer.The existing HAS QoE evaluation methods only take parts of influence factors into account.Moreover they are usually stateless,for they only taking current system and environment conditions into account without considering the impact of past conditions and experiences on the subject's quality judgment,which generally leads to a certain deviation from the user's actual experience quality.Memory effect is a psychological influence factor of past experience,by adding the factor into the QoE evaluation can improve the accuracy of user experience evaluation.On the basis of fully investigating the QoE evaluation methods of HAS,we propose a QoE model for short-form HAS video,which takes the network layer factors,application layer factors and user layer factors into account and can make personalized evaluation for different types of users.In order to collect user data,we built a streaming media system based on Apple's HTTP Live Streaming framework.Our subjective test system can ensure the playback of cross-platform and users can give their evaluation information after finishing viewing.The user data collected was then analyzed by multivariate linear return analysis method so as to solve the model parameters.Then we conducted a series of subjective experiments to verify the accuracy of the model and finally we analyzed the application of the model.