Dynamic Trust-Based Secure Node Selection in 5G/6G Federated Learning

As 5G/6G networks expand, mobile federated learning (FL) has emerged as a key method for distributed intelligent computing. Current FL systems rely on fixed, trusted third parties (TTPs) for security, such as key management and gradient aggregation. This creates single points of failure and risks collusion between servers and TTPs, potentially exposing user data. The risk increases when FL servers run on the $5 \mathrm{G} / 6 \mathrm{G}$ core network or external servers. The core network handles both data and device authentication, which can lead to collusion with application servers. This setup harms both training fairness and user privacy. We propose a decentralized TTP selection algorithm based on dynamic trust evaluation. Before each training round, devices collaborate to select a temporary trust threshold (TTP) by examining past interactions, trust scores, and available resources. This method spreads security tasks across devices and reduces control by the core network or external servers. The algorithm considers energy usage and adheres to 3GPP device-to-device (D2D) communication standards. It ensures secure TTP selection with low computing and communication costs, improving both fairness and efficiency. Experimental results demonstrate that our approach outperforms existing methods in accurately identifying malicious devices, thereby reducing the malicious device participation rate (MDPR) and enhancing the overall security and robustness of federated learning.