The Design And Implementation Of Radio Frequency Fingerprint Blind Recognition Methods

With the popularity of the Internet of Things and the advance of 5G,the security of wireless communication devices has received more and more attention.As the physical layer feature of the device,RF fingerprint can only mark the device physically,and is expected to bring new and reliable means for wireless network security based on the traditional authentication mechanism.At present,almost all the existing researches on radio frequency fingerprint technology are based on supervised methods,which require expert experience to manually label signals in advance.However,in practical applications,such as electromagnetic offensive and defensive warfare,intrusion detection and other scenarios,it is often difficult for researchers to obtain the real information of the opponents signal samples.Moreover,the above-mentioned environment often has relatively high-intensity noise,accompanied by obvious channel fading effects,and the above-mentioned various factors make the existing supervised radio frequency fingerprint technology unable to be effectively applied in practice.In view of the above problems,this thesis focuses on the field of blind radio frequency fingerprint recognition,and proposes an unsupervised blind radio frequency fingerprint recognition method.In addition,this thesis also focuses on complex environments under high electromagnetic noise and multipath channels,and studies how to maintain the robustness of blind recognition models in low SNR environments.Finally,relevant experiments are designed to verify the effect of the proposed blind recognition model,and a RF blind recognition system is designed and implemented based on the blind recognition model.The main contributions of this thesis include:(1)Apply the unsupervised blind identification method to radio frequency fingerprint technology.Based on the success of deep iterative clustering and deep contrastive clustering methods in the field of image clustering,this thesis improves the related algorithm according to the characteristics of radio frequency I/Q signals,making it suitable for the field of radio frequency fingerprinting.The deep clustering model researched and implemented in this thesis can directly perform feature extraction and clustering on the original I/Q signal,and automatically complete the blind identification of RF fingerprints without the intervention of source signal prior information and expert experience,greatly reducing The time and labor cost of manual marking provides a new idea for the practical application of radio frequency fingerprint technology.(2)The RF fingerprint identification method in the actual channel environment is studied and implemented,which enhances the robustness of the model.This thesis combines the deep learning noise reduction method with the traditional noise reduction method in the communication field,and designs a noise processing module based on a noise reduction autoencoder in the deep iterative clustering model;Enhanced,combined with the adaptive threshold shrinkage algorithm in the field of wavelet noise reduction,which reduces the impact of actual electromagnetic environment on model performance,enhances the robustness of the model,and has practical research value.(3)On the basis of the above algorithm research,according to the actual project requirements,a radio frequency fingerprint blind identification system is designed and implemented.The main functions of the system include signal management,statistical analysis,signal snapshot and visualization of blind recognition results,etc.,providing an efficient platform for relevant researchers to systematically study RF fingerprints.