Research On Radio Frequency Fingerprint Recognition Technology Of Electromagnetic Signal

This thesis mainly focuses on the identification of mobile phone radiation sources based on 4G LTE signals,the fusion of multi-dimensional features,and the closed set identification of electromagnetic radiation sources in case of missing enough data.The content of this thesis mainly includes the modules of signal feature extraction,signal feature selection and individual classification.Focusing on the study of the combination of deep learning and traditional feature extraction algorithms,the application of deep learning algorithms in the classification and recognition of electromagnetic radiation sources,and the identification of closed sets of electromagnetic radiation sources in a missing data-driven environment.The feasibility of the algorithm is verified by the measured data.The work and innovations of this paper are mainly concentrated in the following aspects.(1)Collect the 4G signal of the smartphone signal under two different channel environments: the communication control channel and the communication service channel of the mobile phone,and analyze and preprocess the signal,analyze and extract the signal characteristics from each signal characteristic dimension,and select the short-term Fourier transform,Hilbert-Huang transform,bispectral transform and other signal feature extraction algorithms,so as to obtain the influence of different feature extraction algorithms on the final individual recognition effect.The effectiveness of the individual identification scheme is verified through experimental demonstrations in a variety of different environments such as the ideal environment in the darkroom,indoor environment,and outdoor environment and different collection schemes.(2)In view of the different characteristics of signals in different dimensions,three feature fusion algorithms based on neural network architecture are proposed.The first two methods fuse features from the perspective of feature stacking and from the perspective of learning different features from dual networks.,The last method starts from the characteristics of different channels for feature fusion.Experiments show that the feature set selected through the multi-dimensional feature fusion algorithm in the classification process has better recognition performance than the single-dimensional feature set.(3)Aiming at the problem that some modes of electromagnetic radiation sources are difficult to identify due to lack of data support due to the lack of data support in the multi-mode working state of electromagnetic radiation sources,a proximity judgment algorithm based on the siamese network is proposed,which induces the network to learn the differences between classes and ignore the differences within classes.The guidance network strengthens the proportion of electromagnetic fingerprint characteristics in the classification process,weakens the proportion of characteristics caused by environmental factors such as radio frequency channels or frequency selective fading,and finally achieves the purpose of individual identification.In the same environment,the proposed siamese network is based on Compared with the conventional convolutional neural network,the proximity decision algorithm has better recognition performance in the low-sample situation.Aiming at the problem of low recognition accuracy caused by insufficient data set,a network decision feedback mechanism was explored and proposed,which finally achieved the effect of improving recognition accuracy.