Research On Application Of Radio Frequency Fingerprint Recognition Based On Multiwavelet

With the widespread use of mobile wireless communication devices,wireless networks are vulnerable to eavesdropping and interference and other security threats.Therefore,how to identify and track radio frequency(RF)devices by using the subtle differences of radio frequency signals of transmitters is an important topic in the field of physical layer security.The launch of the steady state radio frequency devices for high similarity(transient)RF signal feature extraction is difficult.Wavelet analysis can extract more complete fingerprint features by fine division of frequency bands.Compared with simple wavelet,multiwavelet has the excellent properties of orthogonal,symmetric and high-order vanishing moment at the same time.Therefore,this paper explores a feature extraction scheme based on multiwavelet packet band decomposition.In this paper,based on the theory of multiwavelet,combined with the excellent timefrequency characteristics of high-multiplicity finite element multiwavelet,according to the actual application scenarios,a large(small)sample recognition system based on multiwavelet packet decomposition is proposed to further broaden the application of multiwavelet in signal processing.The following research work is completed in this paper:1.Research the related theories of finite element multiwavelet construction and multiwavelet packet decomposition,focusing on the process of constructing the scale coefficient matrix and wavelet coefficient matrix of high-multiplicity finite element multiwavelets.Based on the classical Mallat theory,the difference between multiwavelet packet decomposition and multiwavelet decomposition is compared,which lays the foundation for the next step of feature extraction.2.On the basis of the traditional cyclic spectrum-graph domain feature extraction method,this paper proposes a small sample RF fingerprint identification system scheme based on finite element multiwavelets.In order to solve a problem,that is,the row index characteristic sequences of the radio frequency signals of the same type of equipment are exactly the same at the overlapping point of the cyclic frequency.Since the increase of multiwavelet multiplicity will lead to higher dimension of the extracted fingerprint features,PCA is used to optimize the decomposed subbands.It is proved theoretically that the multiwavelet decomposition subband algorithm has lower computational complexity than DFT analysis filter.3.A multiwavelet-LPC feature extraction method under the condition of large samples is proposed,and the Cusum algorithm,variance trajectory starting point,fractal Bayesian and LPC sliding window signal starting point detection algorithms are compared and analyzed.Then,high-multiplicity finite element multiwavelets are used to decompose subbands and extract the features of linearly encoded LPC coefficients.In order to eliminate redundancy,LDA is used to reduce the dimensionality of eigenvectors,and the separability index of eigenvectors before and after dimensionality reduction is analyzed.Simulation and comparison of the correct recognition rate of Bluetooth signals in four cases of finite element multiwavelet and GHM multiwavelet illustrate the advantages brought by high multiplicity.In addition,in order to prove that the LPC feature extraction method is suitable for a variety of classifiers,three machine learning classifiers,KNN,random forest and SVM,are designed to more strongly illustrate the significant advantages of high-multiplicity finite element multi-wavelets in feature extraction.In conclusion,the RF fingerprint feature extraction system based on finite element multiwavelet proposed in this paper can optimize the overall recognition performance in steady state signal set and transient signal set compared with the classical GHM multiwavelet and single wavelet systems.