Research On The Fingerprint Extraction And Identification Methods Of Wireless Devices

In recent years,wireless communication technologies such as 5G networks and the Internet of Things have developed rapidly.However,compared with traditional wired transmission,the broadcast nature of wireless communication brings more serious security problems.As the first line of security communication defense,the importance of device authentication is self-evident.Device fingerprint identification is an effective and lightweight device authentication solution.Due to the inevitable imperfection of hardware design and production,every transmitter has unique hardware errors.This error is hard to counterfeit.Thus,the error is also called the device fingerprint,which will slightly distort transmitted signals but does not affect the normal communication functions.In addition,the extracted device fingerprint can be used to identify different transmitters,which is meaningful for enhancing the security of wireless networks.Therefore,it is important to carry out research on the device fingerprint extraction and identification.This dissertation has conducted extensive research in the field of device fingerprint extraction and identification,including the device fingerprint extraction and identification methods of LFM radar in multiple working modes,a device fingerprint extraction and identification scheme of DSSS system under low SNR,the superposition of classification sub-waveforms based denoising method,a denoising method based on the standard unit waveform and cross-correlation,and a channel-robust device fingerprint identification scheme.The innovative achievements of this dissertation are summarized as:1.Designed a device fingerprint extraction and identification scheme for multi-mode LFM radar.Since the LFM pulse in different modes has different fingerprint features,this dissertation proposes a dynamic and self-refining classification(DRC)algorithm for radar operation mode classification.This algorithm can dynamically establish and update operation mode library,and clear the interference pulses by periodically refining the library.The experiment show that the algorithm has excellent pulse classification performance and interference pulse removal ability.Then,a piecewise curve fitting based denoising(PCFD)algorithm and a hybrid RFF identification method are proposed,which can extract both transient fingerprint and steady state fingerprint.The scheme performance has been verified on three real radars spanning four operation modes.Experimental results show that the identification accuracy for four operation modes all exceeds 90%,when SNR is about 5d B.For Mode 3,the identification rate can reach about 100% even at-10 d B.2.Proposed a device fingerprint extraction and identification scheme for DSSS systems under low SNR.For scenarios such as DSSS communications,the received signal SNR will be very low.Thus,an efficient denoising algorithm suitable for fingerprint identification of DSSS system is proposed here for the first time.This algorithm eliminate the noise by stacking the repetitive sequence in the DSSS signal.The mathematical derivation and and simulation experiments have proved that the stacking of M sequences can achieve an M-fold increase in SNR.In addition,the simulation results show that the fingerprint library established under appropriate SNR can be used for various SNR scene fingerprint identification.This is of great significance to the practical application.Furthermore,this denoising algorithm can be extended to any communication system with repetitive sequences,such as the 8 repetitive symbols in the Zig Bee signal preamble.3.Proposed a classified sub-waveform superposition based denoising algorithm for device fingerprint identification.In order to improve the noise robustness of device fingerprint identification technology,this dissertation proposes a denoising algorithm based on classified sub-waveform superposition.Since most signals are formed by a combination of sub-waveforms with limited types,this denoising algorithm will show great universality.In addition,due to the memorability of fingerprint,the sub-waveforms should be classified according to the number of consecutive peaks and valleys.In this way,the damage of fingerprint can be minimized when the sub-waveforms is superimposed.Compared with the performance of the original I/Q signal,the identification accuracy of the denoised signal has great improvement in the identification experiment of 54 Zig Bee devices,when the SNR is in the range of [0d B,20 d B].4.Proposed a unit-waveform and cross-correlation based denoising algorithm for device fingerprint identification.Based on the uncorrelation between noise and communication signals,cross-correlation is a denoising scheme.However,the cross-correlation integration operation will average the device fingerprint,which means that the fingerprint is damaged.The key to this algorithm is to select a suitable reference signal to achieve a proper balance between denoising and fingerprint damage.This dissertation proposes to use the standard unit-waveform as the reference signal.First of all,the standard unit-waveform has the highest similarity with the received signal,which is very important for recovering the transmitted signal from the noise signal.At the same time,the signal length of the unit waveform is small,thus the fingerprint damage after cross-correlation is small.Finally,54 Zig Bee identification experiments proved the effectiveness of the algorithm.Within the SNR range of [-10 d B,30 d B],the average accuracy rate brought by this algorithm is improved by 1.15% to 53.89%.5.Proposed a channel robust device fingerprint extraction and identification scheme.In order to extract channel independent fingerprint features,a channel robust device identification scheme based on the difference of logarithmic spectra is proposed.The algorithm first obtains two different symbols with different amplitudes and phases from the received signals;these symbols exhibit different RFF characteristics.When the two symbols are within channel coherence time,during which the channel can be assumed stationary,the difference between the logarithmic spectrum of the two symbols can effectively eliminate the channel response but retain the fingerprint features.In addition,denoising operations and statistics-based data cleaning operations are applied to further improve the identification accuracy.The IEEE 802.11 OFDM signal is used as an study case.Combined with the CNN classifier,the experiment is carried out on a data-set of 20 Wi-Fi devices with the same model.Compared with the method without removing channel effects,the highest increase in the identification rate is 83.11%.