Tools
Tools
Naval Dos Santos, Orlando (2025) Selective prioritisation of real-time IP packets to improve quality of service in 5G wireless sensor networks. Doctoral thesis, London Metropolitian University.
The successful deployment of 5G and future-generation networks depends critically on the provision of Ultra-Reliable Low-Latency Communication (URLLC), which is vital to meeting the stringent requirements of emerging real-time applications. However, the static, rule-based scheduling algorithms traditionally used for Quality of Service (QoS) management are ill-equipped to handle the dynamic, stochastic nature of modern wireless environments. This thesis addresses this critical gap by designing, implementing, and empirically validating a novel, multi-stage Artificial Intelligence (AI) framework that enables a transition from passive network analysis to active, autonomous control. The central research question investigates whether a hierarchical suite of Machine Learning (ML), Deep Learning (DL), and Deep Reinforcement Learning (DRL) techniques can dynamically manage real-time traffic to significantly enhance QoS in 5G Wireless Sensor Networks (WSNs).
The research methodology unfolds in three logical stages, validated within a live network testbed. First, supervised learning techniques – including regression and classification models – are used to quantitatively evaluate and “fingerprint” traditional schedulers, establishing a robust performance baseline and confirming that predictability is a critical QoS metric. Second, the investigation proceeds to predictive forecasting, where a comparative analysis reveals that a tuned RF model leveraging rich instantaneous metrics outperforms complex sequential DL architectures for predicting next-step network delays.
The culmination of this research work is the development of an autonomous scheduling agent. By framing the network scheduling as a Markov Decision Process (MDP), a novel Transformer-based RL agent is trained online. The findings demonstrate that this agent can learn an optimal policy in real time, achieving 100% scheduling success with ultralow latency and closing the loop from analysis to control.
Collectively, this research contributes a comprehensive, validated AI-driven framework for intelligent QoS management. It provides a practical roadmap for evolving from static network configurations to adaptive, self-learning systems. The results confirm that the proposed hierarchy of AI techniques can successfully navigate the complex trade-offs in 5G WSNs, significantly outperforming traditional methods and paving the way for truly autonomous network control.
 |
View Item |