Data-Driven QoE Modeling for Real-Time Video Streaming and Web Browsing Services in Cellular Networks

Abstract

In recent years, traffic in mobile networks has increased exponentially due to data exchange from specific services, namely video and web browsing services. Video traffic generates the most significant volumes of data, increasing daily due to the emergence of new formats and innovative services, such as live streaming, 360° videos or virtual reality. On the other hand, web traffic generates the second-highest data volume. The increasing demand for network services with high-Quality Service (QoS) requirements sets new challenges for next-generation cellular networks, especially considering the high traffic volume generated and stored in the network by different services that converge on it. This trend has forced network operators to change how they manage their systems from a network-centric to a user-centric approach, making customer experience management a key process in the daily routine of network operators.

This thesis addresses the analysis and modeling of the user Quality of Experience (QoE) of real-time video and web browsing services in cellular networks. For this purpose, the key factors affecting service performance are identified by leveraging information registered in connection traces with data mining techniques. In the case of live video streaming, a parametric model is proposed to estimate the QoE of the encrypted YouTube Live service from packet-level data collected at the interfaces of a wireless network. Unlike previous works, the proposed method is valid for encrypted and adaptive video content. For 360° video service, a study of the impact of the uplink of a mobile network on 360° live video streaming on YouTube is proposed. Unlike previous works, the analysis covers the ingest link, which strongly impacts the latency of live transmissions. Regarding web browsing, an unsupervised web classification scheme is presented to group web pages according to content characteristics that affect the QoE perceived by the end-user.