Robust Optimization in Non-Linear Regression for Speech and Video Quality Prediction in Mobile Multimedia Networks

Abstract

Quality of service (QoS) and quality of experience (QoE) of contemporary mobile communication networks are crucial, complex and correlated. QoS describes network performance while QoE depicts perceptual quality at the user side. A set of key performance indicators (KPIs) describes in details QoS and QoE. Our research is focused specially on mobile speech and video telephony services that are widely provided by commercial UMTS mobile networks. A key point of cellular network planning and optimization is building voice and video quality prediction models. Prediction models have been developed using measurements data collected from live-world UMTS multimedia networks via drive-test measurement campaign. In this paper, we predict quality of mobile services using regression estimates inspired by the paradigm of robust optimization. The robust estimates suggest a weaker dependence than the one suggested by linear regression estimates between the QoS and QoE parameters and connect the strength of the dependence with the accuracy of the data used to compute the estimates.