Big Data-Driven AI-based Wireless Network Personalization

Abstract

There is a consensus that trends in emerging wireless technologies, such as explosive data requirements and proliferating services and applications, are creating serious issues for the management of experience. This is due to the fact that, unlike Quality-of-Service (QoS), for which technical metrics are available such as packet error rate, it is difficult to analytically model and measure user experience in wireless networks. Besides, current communication networks use design methodologies that prevent the realization of maximum network efficiency. For all these reasons, more agile, intelligent, and flexible networks are required. Such networks should be capable of micro-managing resources in a way that meets each user's service quality expectations while using a minimum amount of resources. This micro-management of network resources has ushered in the concept of wireless network personalization. Personalized networks optimize two correlated and contradicting objectives in real time: satisfaction and resource utilization. In this thesis, the utilization of Artificial Intelligence (AI), big data analytics, and real-time non-intrusive feedback is proposed in order to enable the personalization of wireless networks. Based on each user's actual QoS requirements and context, a Multi-Objective Optimization (MOO) formulation enables the network to micro-manage and optimize the provided QoS and satisfaction levels simultaneously. In addition, to enable feedback tracking and measurement, satisfaction is modeled based on the Zone of Tolerance concept. Furthermore, a big data-driven AI-based personalization framework is proposed to integrate personalization into wireless networks. Moreover, a MOO problem is formulated to model the personalized resource allocation problem in wireless networks. Then, to validate the formulated problem, a proof-of-concept prototype is built using the wireless personalization framework in which Evolutionary Multi-Objective Optimization (EMOO) is utilized to find the optimum Pareto front solutions. Finally, since privacy is a crucial concern, a privacy-preserving framework, which is based on the concept of Differential Privacy, is proposed for personalized wireless networks.