Cross-Domain RF Fingerprinting with FDA-based Representations and Few-Shot Learning
Conference
Sun, Y, Kumar, R, Zhao, T et al. (2026). Cross-Domain RF Fingerprinting with FDA-based Representations and Few-Shot Learning
. IEEE Infocom. Proceedings, 10.1109/INFOCOM59046.2026.11571198
Sun, Y, Kumar, R, Zhao, T et al. (2026). Cross-Domain RF Fingerprinting with FDA-based Representations and Few-Shot Learning
. IEEE Infocom. Proceedings, 10.1109/INFOCOM59046.2026.11571198
In this paper, we propose a cross-domain radio frequency (RF) fingerprinting framework that leverages functional data analysis (FDA) to enable effective few-shot adaptation under domain shift. Specifically, FDA is employed to transform raw data into smooth functional representations, which suppress domain-specific distortions while preserving device discriminative structure critical in low data regimes. By modeling in-phase and quadrature (I/Q) sequences as continuous functions via truncated basis projection, FDA provides denoised and compact representations that are more stable across capture conditions. These functional representations are then processed by a deep learning model to learn domain-invariant embeddings that reflect intrinsic hardware impairments. Built on FDA-based representations, the proposed deep learning model achieves consistently improved cross-domain device identification performance in few-shot settings, outperforming traditional deep learning baselines. The results demonstrate that the FDA plays a key role in improving the accuracy and efficiency of RF fingerprinting under practical deployment shifts.
