Design And Implementation Of Radio Frequency Fingerprint Identification System Based On Deep Learning

With the development of wireless communication technology,various wireless signals and Internet of Things applications have become closely related to people's daily lives.Therefore,communication security in wireless networks is getting more important.Due to the inherent openness of wireless networks,communications between normal devices are vulnerable to malicious attacks such as eavesdropping and deception,which can lead to information leakage and destruction,and threat to wireless network security.The traditional identity authentication mechanism based on high-level encryption cannot effectively deal with malicious attacks such as eavesdropping and identity forgery.Moreover,complex identification algorithms for identity verification in this mechanism are limited by Io T scenarios,for simple device structures and limited computing capabilities of Io T devices.Because of the stability and uniqueness,Radio Frequency Fingerprint(RFF)can effectively identify device ID and improve the security of the wireless network at the physical layer.There are still some shortcomings in the current research on the RFF identification technology based on deep learning.On the one hand,a RFF extraction usually requires preprocessing of the signal,which will cause the loss of information in the signal to a certain extent.On the other hand,the current RFF identification method based on deep learning improves the accuracy at the cost of more network layers and more parameters,which is less feasible in real Io T application scenarios.To solve the problems above,this article has launched two researches on the RFF identification based on deep learning.Firstly,this article proposes a RFF identification method based on CNN-GRU network,which can directly learn and extract RFF features from the collected original signals.This CNN-GRU network model has combined advantages of feature extraction advantages in Convolutional Neural Network(CNN)and time features extraction advantages in Gated Recurrent(GRU).The CNN-GRU network can extract both I/Q and time domain features of the original signal sampling data as the RFF feature of the transmitting device.Experiments have proved that the RFF identification method based on CNN-GRU network has excellent identification ability.Compared with other CNN methods,it has fewer parameters,faster calculation speed and better performance.Then,starting from the real application scenario,this paper studies the embedded implementation of RFF identification based on the CNN-GRU network,and designs and completes a set of NFC signal RFF identification system.Considering the real scenario of NFC communication,this article compares and analyzes the performance,cost,power consumption and actual feasibility of various Software-Defined Radio equipment and embedded development boards in the market,selecting the equipment and building a physical system.After that,transplant the trained CNN-GRU network model to the system.Aiming at the automatic recognition and feedback function in real application scenarios,this paper designs and programs a real-time detection module to identify valid signals and return the results while the system is running,filtering invalid signals.The experimental results of the system in the real NFC communication scenario show that the system designed and completed in this paper has higher recognition accuracy,speed,and lower power consumption and cost basically meet the requirements of application scenarios.At the same time,it also verified that the RFF identification method based on CNN-GRU network can be applied to real embedded systems.The design and implementation of this system provides a possibility for the practical application of deep learning-based RFF identification technology in the Internet of Things scene,and provides a solution for ensuring the security of wireless networks...