Optimized Coherent Integration-Based Radio Frequency Fingerprinting in Internet of Things

Abstract

This paper focuses on the security risks in the access authentication of Internet of Things, to provide an optimized classification algorithm of radio frequency fingerprinting (RFF). The novel method is based on coherent integration, multiresolution analysis, and Gaussian support vector machine (SVM). First, we proposed a de-noising algorithm for RFF as the present performance of RFF is seriously affected by the noise. The optimized coherent integration first developed in this paper effectively improves the signal-to-noise ratio (SNR) of the waveform without increasing the number of required signals, by which a de-noising processing is performed and the identification accuracy is improved. Then a wavelet-based multiresolution analysis is applied to extract feature points in the waveform that has passed the de-noising optimizer, because the less sample points are needed for the SVM classification processing, which reduces the computational complexity of SVM compared to the classical SVM classification methods where massive sample points are necessary. Extensive experiments are performed. Simulation result shows that the optimized classification algorithm achieves a high accuracy (exceed 99%) at a relatively low SNR (0 dB), which is the best result compared with other existing methods.