FedTrans6G: Federated Transformer Framework for Privacy-Preserving Resource Management in 6G-Enabled Consumer IoT Ecosystems

Abstract

The rise of 6G technologies and the proliferation of consumer IoT devices introduce critical challenges in achieving scalable, low-latency, and privacy-preserving intelligence at the network edge. Traditional federated learning (FL) frameworks, while decentralized, often fall short in addressing heterogeneity, communication efficiency, and data privacy under 6G constraints. To overcome these limitations, we propose FedTrans6G, a novel Federated Transformer Framework designed for secure and adaptive resource management in 6G-enabled IoT ecosystems. FedTrans6G integrates lightweight transformer-based models with hierarchical federated learning, enhanced by differential privacy and homomorphic encryption to ensure end-to-end confidentiality. The framework features an adaptive resource allocation mechanism that leverages transformer attention scores for real-time optimization across edge, fog, and cloud layers. We validate FedTrans6G through extensive simulations using real-world and synthetic IoT datasets. Empirical results show that FedTrans6G outperforms state-of-the-art baselines in accuracy (+6.2%), latency (−43.8%), and energy efficiency (−30%), while significantly reducing privacy leakage. Ablation studies further confirm the effectiveness of each architectural component. The proposed system demonstrates practical viability for next-generation, privacy-aware, and resource-efficient edge intelligence. FedTrans6G offers a paradigm shift for 6G IoT, bridging the gap between intelligent model design and federated privacy guarantees—paving the way for secure, scalable, and sustainable edge computing infrastructures.