A Novel RFF Extraction and Recognition Method Based on DMRS and Deep Learning

Radio Frequency Fingerprint (RFF) identification is a physical-layer security technology based on hardware imperfections, which provides unique device authentication for wireless devices. However, current RFF identification faces challenges in multi-device environments and low signal-to-noise ratio (SNR) scenarios. This paper proposes an RFF extraction and recognition method based on Demodulation Reference Signal (DMRS) and deep learning to address these challenges. The DMRS is preprocessed to enhance RFF features, enabling effective extraction of RFF. To reduce the influence of outliers on recognition performance, the Local Outlier Factor (LOF) algorithm is employed to filter outliers in the extracted feature data. An RFF identification network under the ResNet framework is constructed for recognition. Experimental results demonstrate that the proposed method achieves a recognition accuracy of over 94% in a classification scenario of 40 devices and reaches 99.39% in a scenario of 10 devices with an SNR of 13.34 dB.