A Drone Classification Method Based On RF Fingerprints

UAV detection and identification is attractive topic for researchers.In recent years,with the developments of communication technologies,RF-based classification of drones shows a good performance by overcoming the shortages of traditional methods depending on radar signals or visual signals.However,current literatures are mostly utilize the ordinary feature of communication signals emitted by drones to construct the feature dataset.This kind of method cannot deal with the situation when the drones signal use the same mode.Therefore,inspired by the idea of emitter fingerprint recognition,this thesis is aiming at classifying different drones based on their communication signal fingerprints.The main content of this paper can be divided into following parts.The formation mechanism of UAV Communication Signal fingerprint is analyzed,and the simulation signal model is constructed to make the simulation signal source effectively reflect the UAV communication signal characteristics including signal fingerprint characteristics.On this basis,the fingerprint feature extraction and classification of UAVs are carried out,and the recognition performance of fingerprint feature algorithm is verified based on the measured signal data set of UAV.As the traditional research of individual identification of radiation source didn't consider the influence of motional carrier,the dynamic characteristics of flying UAV are analyzed,and the vibration test is designed to verify that mechanical vibration can enhance the phase noise characteristics of the transmitted signal.On this basis,the fingerprint feature model of UAV is established.In the process of feature extraction,the original signal is decomposed into several sub modes by using the variational mode decomposition algorithm.Considering the poor noise robustness of the multi-scale entropy,this paper proposed an improved multi-scale entropy algorithm to extract the complexity features of each sub mode.Additionally,in order to reduce the complexity of the algorithm,the PCA algorithm is employed to reduce the dimension of the feature.The result shows that the improved algorithm shows better recognition performance at low SNR.In order to solve the problem of feature redundancy of rectangular integral bispectrum in high-order spectral analysis,a new algorithm based on the maximum proportion interval is proposed.The simulation results show that the recognition rate of the improved rectangular integral bispectrum feature is improved in different SNR.In this paper,the performance of the proposed fingerprint extraction algorithm for UAV is analyzed and verified by the simulation signal and measured signal.The results show that the method proposed in this paper can effectively extract the phase noise and nonlinear characteristics of UAV communication signals,and achieve a recognition rate of more than 90 % for real UAV signals.