Research On Radio Frequency Fingerprint Identification Technology Based On Deep Learning

Radio frequency fingerprint technology realizes wireless individual identification by extracting information on individual differences of radiation sources carried by electromagnetic waves.It has a wide application prospect in the fields of wireless network security authentication,spectrum resource management,electronic countermeasures and so on.This paper studies the methods of RF fingerprint feature acquisition and emitter individual recognition based on deep learning.The main research results are as follows:(1)The generation mechanism of RF fingerprint is analyzed,and the factors that may constitute the characteristics of RF fingerprint are analyzed from the two aspects of frequency domain distortion and time domain distortion of RF amplifier.Then it introduces the bispectrum theory which can characterize the nonlinear characteristics of the signal and the bispectrum estimation algorithm based on limited observation data,and gives the bispectrum of WiFi signals of different individuals of the same model for visual comparison.(2)A radio frequency fingerprint identification method based on signal bispectrum and residual neural network(RESNET)is proposed.Firstly,the collected signals from different devices are estimated bispectrum,and the bispectrum contour map is obtained and labeled with equipment label.Then the bispectrum contour map is trained by using the built improved residual neural network model,and the network weight is updated by back propagation(BP)and gradient descent to obtain the optimization model.Finally,another group of bispectrum contour maps is used to verify the recognition performance.The experimental results on the measured WiFi signals of six routers show that the recognition rate of the improved residual neural network algorithm based on signal bispectrum is 93% in the actual electromagnetic environment,which is an effective RF fingerprint recognition method.(3)A multi feature fusion radio frequency fingerprint recognition method based on deep learning is proposed.This method is extended on the basis of bispectrum and residual neural network.Firstly,the multi view sample data set is made by using the three representations of time domain,frequency domain and bispectrum of the signal to be identified.The complex value residual neural network is built respectively to extract the fingerprint features of time-domain complex baseband signal,and the real value residual neural network extracts the frequency domain and bispectrum features.Then,after splicing the ends of the three neural networks,multiple view samples are used to train and save the optimal weight data.Finally,the trained combined neural network is used to extract the RF fingerprint features of multiple views in the test set samples and input them into the classifier for classification and recognition.The experimental results of the combination of three views show that the recognition rate of six types of WiFi signals can be improved to 98.2% under 15 dB signal-to-noise ratio,and the recognition rate under each signal-to-noise ratio can be greatly improved under the same number of initial samples.