Dynamic Reinforcement Learning-Driven Digital Twin for Optimised Multimedia Traffic Management in B5G SDN Core Networks

The rapid growth of multimedia applications and the rising expectations for an enhanced Quality of Experience (QoE) among users have emphasised the need for innovative approaches to ensure efficient delivery of video content in 5G and Beyond 5G (B5G) networks. Machine Learning (ML) techniques are increasingly being explored to address these challenges by enabling intelligent traffic management and QoE optimisation. Within this landscape, Software-Defined Networking (SDN) plays a pivotal role as a facilitator of dynamic resource allocation and QoE-centric network management. This paper introduces AIMTWIN, a reinforcement learning (RL)-driven Digital Twin framework designed to optimise multimedia traffic management in B5G SDN environments. AIMTWIN integrates real-time telemetry from physical networks with a dynamic virtual model to provide adaptive and efficient traffic routing. By prioritising static paths for QoS-critical multimedia flows and dynamically managing background traffic, the framework delivers superior network performance and user satisfaction. Experimental results on small and medium scale topologies highlight AIMTWIN’s ability to achieve consistently Excellent QoE compared to state-of-the-art routing methods, positioning it as a scalable and robust solution for next-generation networks.