A Noise-Robust Radio Frequency Fingerprint Identification Scheme for Internet of Things Devices

Radio frequency fingerprint (RFF) identification is a potentially effective technique to address the authentication security of Internet of Things (IoT) devices. Since the complex working environment and limited resources of IoT devices, noise is non-negligible in RFF identification of IoT devices. It is a challenge to suppress the noise without damaging the RFF information. In this paper, we propose a robust RFF identification scheme, which consists of a frequency point selection (FPS) based denoising algorithm, and a convolutional neural network (CNN) classifier. The FPS algorithm performs denoising by filtering out all the frequency components that are independent of the RFF. The CNN is designed with a dynamically decreasing learning rate to accelerate learning and obtain optimal identification performance. Experiments were conducted with 54 ZigBee devices to evaluate the performance of the proposed scheme under three different RFF identification scenarios. The results show that the FPS algorithm brings the highest accuracy improvement of about 25 % when the training signal-to-noise ratio (SNR) is hybrid and the test SNR is 0 dB.