Multi-Modal Feature Fusion and Gaussian Mixture Model based Robust Open-Set Radio Frequency Fingerprint Identification

Radio frequency fingerprint identification (RFFI) is a promising authentication technology based on hardware-specific radio frequency (RF) front-end characteristics. The device-specific hardware features can be extracted at the receiver by analyzing the received signal and used for authentication. However, distortions introduced by different antennas severely degrade the extraction of RF fingerprint features in multiantenna receiving scenarios. To address this challenge, a multimodal feature extraction network and an attention mechanismdriven feature extraction module are proposed to achieve indepth signal characterization. In addition, for open-set and multiantenna receiving scenarios, a classification model combined prototype learning with Gaussian mixture modeling is proposed to solve antenna interference issue. Above all, this paper proposes an open-set RFFI framework integrating feature fusion and Gaussian mixture model (GMM). We carried out extensive experimental evaluation on several WiFi datasets, experimental results demonstrate the framework’s robustness and superior performance in complex open-set environments.