مهند الخزرجي

The migration of mobile telecommunication systems from 5G to the expected 6G generation requires a paradigm shift in the network management approach, from the traditional rigid Quality of Service (QoS) paradigm to a holistic an‎d user-centric Quality of Experience (QoE) paradigm. However, the development of reliable QoE prediction models based on the traditional centralized machine learning paradigm is increasingly challenged by the stringent data privacy regulations, the overhead of excessive backhaul communication, an‎d the statistical difficulties of Non-Identically an‎d Independently Distributed (Non-IID) data in different network cells. This dissertation proposes an‎d analyzes a privacy-preserving Federated Edge AI solution that is intended to enable real-time QoE prediction based on Radio Access Network (RAN) telemetry data processed at the network edge. The work uses a tailored lightweight neural network model, which is optimized for the resource constraints of edge devices, an‎d benefits from the FedProx optimization algorithm to overcome the challenge of client drift caused by statistical heterogeneity. The system is tested on a real-world 5G production dataset, enabling a comprehensive three-way performance analysis of federated, centralized, an‎d local learning strategies. The experimental result indicates that the proposed Federated Edge AI system has an R^2 value greater than 0.92, which is comparable to the centralized approaches, while preserving the overall data sovereignty. In addition, the communication efficiency of the proposed system is remarkably good, as it only requires less than 1 MB of data transfer for the model to fully converge. These findings confirm that the integration of FedProx with lightweight an‎d edge-native models is a scalable, robust, an‎d privacy-preserving approach for the proactive, zero-touch management of user experience in future wireless networks.

يادگيري فدرال , هوش مصنوعي لبه , شبكه هاي 5G

Federated learning , Edge AI , 5G networks

Mohanad Alkhazraji

Dr. Diyanat