RF fingerprint recognition based on convolutional operator networks

With the widespread adoption of wireless devices, device security faces severe challenges. Traditional RF fingerprint recognition based on intelligent methods tends to focus on local features under low signal-to-noise ratio conditions and relies on a single method for extracting time-domain information, resulting in significantly degraded recognition performance. To address this, we propose an RF fingerprint recognition method based on extracting functional domain features. Based on a universal mathematical model of RF fingerprint signals, a functional domain mapping relationship is established between wireless transmitter signals, time series, and device identities. Continuous operators are employed to suppress additive noise disturbances while extracting subtle hardware characteristics at deeper levels. Building upon this foundation, a deep operator network (RFF-CONet) is designed for RF fingerprint recognition. Simulation results demonstrate that the proposed method achieves higher accuracy than traditional approaches under low signal-to-noise ratio conditions while exhibiting faster convergence. In experiments on the WiSig dataset, RFF- CONet achieves a 51% higher recognition accuracy than the optimal reference model (ORACLE) for 135 transmitters at a low SNR of 0dB, converging in just 12 epochs—outperforming the fastest model, LSTM. Compared to traditional time/frequency domain feature methods, the proposed approach offers a novel perspective for wireless device security authentication.