On The Identification Technique Of Communication Individual Transmitter

As an important research topic in the field of communications confrontation,transmitter identification technique is able to identify individual equipment based on the relationship between the subtle features and the individual resulting from differences in hardware.This subject has important impact on some areas such as wireless network security,military communication confrontation and radio monitoring.Nevertheless,it still confronts with many great difficulties like very small difference among individuals,feature extraction,high dimension feature selection and so on.This thesis attempts to focus on the research of individual recognition based on transient feature,steady feature and deep learning by analyzing some key technologies of individual identification of communication transmitters,including feature extraction,dimension reduction,feature fusion and classification,and all of the proposed scheme are verified to be efficient by the measured data.The main areas in this thesis are as follows.(1)The individual recognition scheme based on transient characteristics is realized.First,the Bayesian detector is used to detect the transient starting point,the coefficient of the polynomial fitting is taken as the transient feature.Then,the feature dimensionality is reduced by PCA algorithm.Finally,The SVM classifier outputs the individual class.The average recognition rate for the given 3 different types of walkie-talkie is 99.61%,and 94.87% is acquired for six walkie-talkie individuals.(2)The feature selection for high dimensional spurious features is studied.The improved multi-class EC-FS algorithm is compared to the Fisher and Laplacian feature selection algorithm,experimental result indicates that the improved algorithm obtains better performance with the acceptable increasment of the time complexity,and it is non-sensitive to the number of features.(3)The DCA algorithm is used to solve the fusion problem for spurious features and integral bispectrum features,and the algorithm is improved for the original algorithm low recognition rate and the weekness of poor stability when the signal is affected by noise.The experimental result indicates that the average recognition rate of the six mobile phones is 91.7%,and the improved DCA algorithm has a 2% ~ 3% improvements than the pre-algorithm when SNR is below 18 dB and has better robustness.(4)This thesis explores the application of the deep learning in the individual recognition of the communication transmitter.The auther designs and constructs a 5-layer one dimension convolutional neural network to extract the deep features of the integral bispectrum.The performance of this deep network is compared with other five commonly used shallow classifiers.The individual recognition rate is about 5% higher than that of random forest classifier(RF)with the highest performance in shallow classifiers,and the average recognition rate of the six given mobile phone is 99.07%.The recognition scheme based on deep learning proposed in this paper provides a new method for the research of individual identification of communication transmitters.