Research On Radio Frequency Fingerprint Recognition Based On Deep Learning

With the continuous development of wireless communication technology,4G,5g and Wi Fi devices have gradually entered thousands of households.While facilitating the life of the general public,it also makes the number of wireless communication devices grow explosively,mixing legal and illegal communication devices,causing a large number of security problems.In order to authenticate the user identity of wireless communication equipment,radio frequency fingerprint identification technology came into being.This technology can identify the radio frequency characteristics caused by the hardware of wireless communication equipment,deduce the hardware characteristics of communication equipment from the radio frequency characteristics,and then identify the identity of the wireless communication equipment.Radio frequency fingerprint identification technology based on deep learning identifies the user’s identity from the physical layer,not analyzing communication information,but identifying the equipment transmitting information,which can effectively resist replay attack and man in the middle attack.Through the in-depth research on the principle of RF fingerprint identification technology and deep learning algorithm,the main research contents are as follows:1.For the lack of data set in the direction of RF fingerprint identification,this paper establishes a simulation and physical experiment platform based on gnuradio software,hackrf one transmitter,walkie talkie and 3900 a receiver.On this platform,firstly,this paper uses gnuradio software to generate signals using QPSK,am and 2FSK modulation methods;Then use 10 hackrf one and two walkie talkies to transmit radio signals with different modulation modes and powers;Finally,the 3900 a receiver receives radio signals in different channels and demodulates the signals to verify the correctness of the signal transmission link.The experimental results show that the radio frequency fingerprint identification technology based on deep learning,which is implemented on the simulation and physical experiment platform based on gnuradio software,hackrf one transmitter,walkie talkie and 3900 a receiver,can achieve better radio frequency fingerprint identification results for radio signals affected by different modulation modes and different channels.2.The traditional RF fingerprint identification algorithm is vulnerable to channel interference and the recognition accuracy is not high.To solve the above problems,starting with the preprocessing method and the classification algorithm based on deep learning,this paper proposes a radio frequency fingerprint recognition method based on comparative learning.Firstly,the IQ signal transmitted by the steady-state wireless communication equipment is preprocessed into the spectrum waterfall as the input,and then the spectrum waterfall of different transmitters is compared by using the comparative learning model.Finally,the characteristics of RF fingerprint are extracted and the identity of the transmitter is recognized.The experimental results on the measured transmitter data set show that,The recognition accuracy rate of using spectrum waterfall as preprocessing method is92.11%,the accuracy rate is 100%,the recall rate is 86.49%,and the F1 score is 92.76%.Compared with the preprocessing scheme using spectrum,IQ and phase,it has higher accuracy.Compared with the traditional supervised comparative learning network,the recognition accuracy of the improved comparative learning network is improved by about5.93%,and has higher accuracy than seven networks such as resnet50 and vgg16.The recognition accuracy of its walkie talkie is as high as 92.11%.3.Aiming at the problem that the traditional RF fingerprint identification algorithm can only identify the radio signal of one modulation mode,this paper proposes a preprocessing method based on the improved spectrum waterfall.This method first intercepts the baseband range of the spectrum waterfall,no longer pays attention to the irrelevant frequency range,and then retains the relationship between the signal strength with time and frequency in the spectrum waterfall to form the improved spectrum waterfall,Finally,the improved spectrum waterfall is input into the supervision and comparison network.The simulation results show that the improved spectrum waterfall has 82% recognition accuracy for the data set composed of 2FSK,am and QPSK;For the training and verification of 2FSK and am data sets,the test result of QPSK data set is that the accuracy of RF fingerprint recognition is 83.5%.