Key Technologies Research For Specific Emitter Identification

Specific Emitter Identification(SEI)is the technology or system that identifies emitters of interest by the features of the signal.As a key technology in Electronic Support Measure(ESM),the SEI has always been attracting much attention.In the civilian field,the SEI also has great application value in such as wireless network security,Io T device identification,and physical layer authentication.In recent years,many excellent SEI algorithms have proposed^[30-37],however,there are still many technical challenges.First,as far as the evaluation criteria of SEI is concerned,most of the existing SEI methods evaluate their performance based on comparison of the accuracy among algorithms.However,due to insufficient public data sets and different SEI scenarios,the horizontal comparison is not completely"fair".Furthermore,researchers lack sufficient theoretical basis to determine whether the accuracy of their SEI schemes has reached the optimal upper bound.Second,the machine learning-based SEI technology also faces many technical difficulties and challenges.For example,the features extracted by most networks are not interpretable,and supervised learning methods are difficult to obtain labels,and the training of the machine learning model lacks sufficient training data.The above problems are all urgent problems to be solved by SEI.For those problems,this thesis discussed the distortion models in detail,and derived the Cramer-Rao lower bound for the I/Q imbalance distortion parameters in wireless channels.Then,under the assumption of the optimal I/Q imbalance distortion parameter estimation,the upper bound of accuracy is derived according to the Neyman-Pearson rule.And the feature-based SEI method is compared with the signal-based SEI method.This identification performance upper-bound of the SEI provides evaluation criteria for related problems in SEI.Then this thesis proposed an unsupervised feature extraction method that combines the distortion model with neural networks.The I/Q imbalance parameters extraction is realized based on a Convolutional Neural Network(CNN),and then the phase noise model parameter is extracted by a Generative Adversarial Network(GAN).