Research On Characteristics Of Radio Frequency Fingerprint Based On Amplifier Nonlinearity

Recognition based on radio frequency(RF)fingerprint characteristics of wireless devices is a physical layer solution for wireless device identity authentication.With the rapid development of wireless networks and increasing security threats,RF fingerprint identification technology is a good choice.The advantages of uniqueness,stability and unclonability of RF fingerprint can provide a higher security performance than traditional wireless networks Authentication method,thereby reducing threats such as malicious attacks.In the existing research on the characteristics of RF fingerprints,many signal characteristics have been studied.But few of them use the non-linear characteristics of the amplifier in the transmission system to extract the fingerprint of the device to make recognition.The power amplifier is a key component that affects the performance of the communication system.Because it normally works in the saturation region,which means nonlinearity always exists,and there are differences between each device due to the deviation of the production process,so the power amplifier has unique and stable physical layer characteristicsBased on above information,this thesis proposes a method to extract the fingerprint feature of the device based on the nonlinear feature of the power amplifier in the signal transmitter,and proposes an estimation method for the nonlinear feature coefficient of the power amplifier model.The nonlinear feature coefficient is used as the fingerprint feature of the device,and then the devices are classified and identified through different classification algorithms.We have done simulations and experiments to prove that this feature can be used for the classification and recognition of wireless devices,and the recognition accuracy is high.The main work of this thesis lists as follows:· For the polynomial model without memory effect,the least square method is used to estimate its nonlinear characteristic coefficients,and the simulation experiment proves that this method can accurately estimate the nonlinear characteristic coefficients of the polynomial model without memory effect.For the polynomial model with memory effect,the expression of the memory polynomial model is expressed in the form of linear matrix equation,and the coefficient extraction problem is transformed into a matrix equation problem.In addition,the QR decomposition of the matrix is introduced to further optimize the process of solving the model coefficients.· A set of quantitative evaluation criteria is introduced to evaluate the results of model coefficient extraction.The results are evaluated by three evaluation factors : MER(memory effect ratio),MEMR(memory effect modeling ratio)and RMSE(root mean square error).· A novel method for estimating the memory length of the memory polynomial power amplifier model based on Generalized Akaike Information Criterion is proposed,so that the memory length of the model can be obtained in advance when estimating the coefficients of the memory polynomial model,which improves the accuracy and calculation efficiency of coefficient extraction.· The influence of channel conditions on the model coefficients of the power amplifier is explored,and the degree of influence of channel equalization on the memory effect is analyzed.We found that the channel conditions will cause the nonlinear characteristic coefficients of the power amplifier model changing greatly.Channel equalization can reduce the deviation caused by the channel but will eliminate part of the memory effect,resulting in the estimated coefficient after channel equalization is not exactly the same as the coefficient of the unaccepted channel,which will affect the recognition accuracy.· Simulation experiments and actual measurements prove that the nonlinear characteristic coefficients of the power amplifier can be used for the classification and recognition of wireless devices,and the recognition accuracy is high.Three different types of signals,ZigBee signal,Wi-Fi signal and LTE RACH preamble signal are collected,and three classification algorithms,LDA,SVM,and KNN are used for classification and identification.The classification accuracy of the LDA classification algorithm and the SVM classification algorithm is higher than 90%.For ZigBee devices,the best-performing classification algorithm is the LDA classification algorithm with an accuracy rate of 93.9%.For LTE RACH signal devices and Wi-Fi signal equipments,SVM classification algorithm performs best,and the classification accuracy rate is as high as 99%.