Federated learning-driven digital twin framework for adaptive resource management in 6G edge networks

Abstract

The anticipated transition to sixth-generation (6G) wireless systems is set to redefine how network resources are managed in environments characterized by vast device heterogeneity, stringent latency requirements, and increased autonomy at the network edge. As centralized control paradigms struggle to keep pace with these demands, there is a growing need for adaptive, decentralized solutions that can make intelligent decisions in real time. In this study, we propose a new architecture that integrates federated learning (FL) with digital twin (DT) technologies to improve the responsiveness and efficiency of resource management in edge-enabled 6G networks. Our approach enables edge nodes to collaboratively train machine learning models without the need to share raw data, thereby preserving privacy and reducing communication overhead. These local models contribute to a central digital twin—a virtual replica of the network environment—that continuously evolves to reflect real-time operational states and predict system behavior. Within this framework, the digital twin enables dynamic optimization across multiple domains, including spectrum distribution, computation task offloading, and energy balancing, by leveraging insights generated from the distributed FL models. Simulation results across varied 6G scenarios reveal that the proposed system offers considerable improvements in network performance metrics, such as reduced latency, higher resource utilization, and enhanced scalability under high device density. The hybridization of FL and DT within a unified architecture demonstrates a viable pathway toward autonomous, self-optimizing network infrastructures, aligning with the envisioned capabilities and challenges of future telecommunications ecosystems.