Differential Privacy-Driven Swarm-learning for Enhanced Device Authentication

• Lei Zhang,
• Kai Cheng,
• Lifang Gao,
• Xiantong Liu,
• Chenhu Guo,
• Fanqin Zhou,

Vol. 19, No. 11, November 30, 2025
10.3837/tiis.2025.11.008, Download Paper (Free):

• Differential privacy
• Swarm Learning
• radio frequency fingerprinting
• Device Authentication

Abstract

The proliferation of multimedia Internet of Things (IoT) devices necessitates robust security measures against sophisticated spoofing and forgery attacks. Radio frequency fingerprinting (RFF) provides a strong physical-layer authentication solution by identifying unique, unclonable hardware characteristics in device signals. However, applying machine learning to sensitive RFF data creates significant privacy risks. Collaborative frameworks like swarm learning, while decentralized, are not immune to inference and data poisoning attacks that can expose device data or corrupt the shared model. This paper proposes a differential privacy-driven swarm learning (DPSL) framework to address these dual challenges. Our approach makes two principal contributions: First, it integrates differential privacy by injecting calibrated noise into local model updates, providing formal privacy guarantees against data leakage. Second, it introduces a malicious device evaluation mechanism to detect and neutralize adversarial nodes, thereby ensuring the integrity and robustness of the global model. Experiments on a real-world dataset demonstrate that DPSL achieves robust authentication accuracy exceeding 90%, while effectively mitigating data poisoning attacks from up to 20% malicious nodes. The proposed framework substantially strengthens security and privacy in RFF-based authentication systems, paving the way for more trustworthy and resilient multimedia services.

Statistics

Cite this article