Towards Efficient Network Management and Optimisation for Next-Generation O-RAN

In the rapidly evolving landscape of 5G and beyond networks, intelligent traffic management and resource allocation have become critical challenges that demand sophisticated technological interventions. This doctoral research presents a comprehensive approach to enhancing \ac{RAN} performance through advanced machine learning techniques within the \ac{O-RAN} architectural framework, addressing fundamental challenges in modern cellular network infrastructure. The research fundamentally transforms traditional network management by developing innovative \ac{ML} models and intelligent algorithms that significantly improve network performance, user equipment (UE) throughput, and inter-cell interference management. By introducing a novel xApp deployed in the \ac{RIC}, the study demonstrates a groundbreaking approach to dynamic traffic steering that leverages multiple advanced algorithmic techniques. The methodological framework combines sophisticated clustering, classification, and prediction models, including K-means clustering for identifying UEs with low throughput, \ac{SVM} for precise UE classification, and \ac{LSTM} models for accurate cell throughput prediction. These techniques are strategically integrated to create an intelligent steering mechanism capable of dynamically managing network resources with unprecedented precision. A critical innovation of this research is the development of an intelligent handover algorithm that minimises unnecessary network transitions while ensuring optimal load distribution across cellular infrastructure. By generating precise handover request messages to E2 nodes, the method achieves a more balanced and efficient network resource allocation that addresses longstanding challenges in cellular communication systems. Furthermore, the research extends its contributions to \ac{PMI} optimisation through an innovative \ac{A2C} reinforcement learning model. This approach uniquely addresses inter-cell interference challenges by simultaneously optimising spectral efficiency and interference reduction, presenting a nuanced solution to resource management in increasingly dense network deployments. Rigorous experimental validation conducted within an O-RAN environment demonstrated remarkable outcomes, including significant improvements in user equipment throughput, more uniform network load distribution, enhanced spectral efficiency, and reduced inter-cell interference. These results not only validate the proposed methodologies but also underscore the transformative potential of machine learning in next-generation network design. Beyond its immediate technical contributions, this doctoral research provides a comprehensive framework for integrating advanced ML techniques into radio access networks. It offers a flexible, adaptive approach that anticipates and addresses emerging network challenges, establishing a critical pathway for future network management strategies. The methodological innovations presented here represent more than incremental improvements; they constitute a fundamental reimagining of cellular network optimisation. By demonstrating how artificial intelligence can be strategically applied to solve complex network management challenges, this research opens new horizons for intelligent, responsive, and efficient cellular communication infrastructure. In conclusion, this doctoral work stands as a significant milestone in the evolution of telecommunications technology, bridging theoretical ML approaches with practical network optimisation strategies. It provides researchers and network engineers with a robust, innovative framework for understanding and implementing intelligent traffic steering and resource management in the era of 5G and beyond.