Title

On Behalf of the User – Towards Advanced and Applicable QoE Modeling in Networked Settings

Organizers

Markus Fiedler and Jörgen Gustafsson

Motivation and objectives

We are witnessing an ever-growing number of digital services and applications to be provided via networked systems. These services and applications are becoming increasingly complex, both in how they interface with the user in how they are implemented. The competition between service providers is growing, which entails the needs to provision in a cost-effective manner while keeping the quality up to mark. Furthermore, industries are becoming digitalized, where the services are frequently time critical, and high quality and reliability are essential properties. Thus, providing a good QoE is an important ingredient of successful service and application provisioning. QoE is user-oriented per definition and has traditionally been evaluated in regulated laboratory settings, with clear tasks at hand (such as watching a ten-second video and provide a rating afterwards). However, it is not feasible to have users assessing many of today’s complex services in this way in a real-life setting, where neither user actions nor system behaviours can be fully controlled. Instead, it becomes necessary to derive estimations of QoE from indicators found in the systems themselves, and from the ways systems, services and applications are used and how they interact. As many services are provided over-the-top, network providers in particular are facing the challenge of being held responsible for quality that they can monitor and control merely in indirect ways, e.g. through link dimensioning, instead of having proper handles and interfaces at their disposal. Thus, they need indicators and models that allow for revealing QoE issues on behalf of the user, while pointing at potential underlying issues on system and network level, and enabling optimization of those. The design of the QoE models must match the needs of a system and network operator, both for consumer and industrial perspective.

During the recent years, QoE research has brought about a set of high-quality models, which rely on rich and complex input signals, and are strongly context-dependent. However, the emergence of new (versions of) services and applications threatens the applicability of those models. A service or network provider may appreciate readily applicable and future-proof models that are based on available data (from any telling measurement point in the system) and knowledge (e.g. from earlier models). Such models might use sources of information in the vicinity of “QoE Mainland”, for instance from usage of services, and employ new analysis and modeling approaches, where machine learning is a key technology. Last but not least, successful cases of “QoE By Design’’, showing how QoE knowledge and models have enabled superior service and application design should be communicated to the QoE Community.