Skip to content

Efficient and Robust Radio Frequency Fingerprint Identification via Introspective Multi-Teacher Distillation

2026 · IEEE Transactions on Cognitive Communications and Networking · Vol 12, pp. 12650-12662 · 0 citations · 54 references
Computer Science

Abstract

Radio frequency fingerprint identification (RFFI) is a key physical-layer authentication technique for cognitive and secure wireless networks. However, the increasing sophistication of physical-layer impairments and adversarial attacks poses significant challenges to the robustness and computational efficiency of RFFI models, especially for resource-constrained and lightweight wireless devices. To address these challenges, we propose an Introspective Multi-Teacher Distillation (IMTD) framework, which jointly exploits clean and adversarial teacher models through a collaborative multi-loss optimization strategy. Here, “introspective” denotes uncertainty-aware student self-assessment: the student dynamically controls the contribution of teacher-guided distillation according to its prediction confidence. By introspectively transferring complementary knowledge from heterogeneous teachers, IMTD effectively enhances model robustness against joint physical and adversarial perturbations while significantly reducing model complexity and inference cost. Experiments on a laboratory hardware–software co-design platform show that IMTD outperforms the compared defense methods across the evaluated perturbation intensities and signal-to-noise ratios. The results support lightweight closed-set RFFI on the tested platform; scalability to heterogeneous hardware, open-set transmitters, and specific edge-device deployment remains to be validated.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.