No abstract

Abstract. Quality of Service (QoS) optimization are not sufficient to ensure users needs. That's why, operators are investigating a new concept called Quality of Experience (QoE), to evaluate the real quality perceived by users. This concept becomes more and more important, but still hard to estimate. This estimation can be influenced by a lot of factors called : Quality of Experience Influence Factors (QoE IFs). In this work, we survey and review existing approaches to classify QoE IFs. Then, we present a new modular and extensible classification architecture. Finally, regarding the proposed classification, we evaluate some QoE estimation approaches to highlight the fact that categories do not affect in the same the user perception.

No abstract

No abstract

In mobile networks, crowdsourcing in Quality of Experience (QoE) assessment phase involves collecting data from the user terminals or dedicated collection devices. A mobile operator or a research group may provide applications that can be run in different mobility test modes such as walk or drive tests. Crowdsourcing using users’ terminals (e.g., a smartphone) is a cheap approach for operators or researchers for addressing large scale area and may help to improve the allocated resources of a given service and/or the network provisioning in some segments. In this work, we first collect a dataset for three popular Internet services: on-demand video streaming, web browsing and file downloading at the user terminal level. We consider two user terminals from two different vendors and many mobility test modes. The dataset contains more than 220,000 measures from one of the major French mobile operators in the Île-de-France region. The measurements are effectuated for six months in 2021. Then, we implement different models from the literature for estimating the QoE in terms of user’s Mean Opinion Score (MOS) for every service using features at radio or application levels. After that, we provide an in-depth analysis of the collected dataset for detecting the root cause of poor performance. We show that the radio provisioning issues is not the only cause of detected anomalies. Finally, we discuss the prediction quality of HD video streaming service (i.e., launch time, the bitrate and the MOS) based only on the physical indicators. Our analysis is applied on both plain-text and encrypted traffic within different mobility modes.