Researches On The Key Techniques For Specific Emitter Intelligent Identification

Specific emitter identification(SEI)is a technique that uses the unintentional modulation information contained in the received signal to identify which emitter it belongs to.Since the unintentional modulation information of signal(also called as radio fingerprint feature(RFF)of the emitter)can uniquely identify emitter,the SEI technique has been widely applied in military and civilian fields.However,traditional SEI system based on expert-defined RFFs has many insurmountable problems,such as low system development efficiency.The root cause of these problems is that it is impossible to establish a unified and accurate model for RFF.This makes the RFFs extracted by the traditional expert-defined methods have many problems such as poor versatility and unclear scope of application.In order to effectively solve the shortcomings of traditional SEI system,deep learning method is introduecd into SEI fields in this paper,and the emitter individual intelligent representation method based on transient signal and the emitter individual intelligent representation method based on steady state signal are also studied.In order to further improve the versatility of SEI system,this paper studies the intelligent signal detection and signal-to-noise ratio(SNR)estimation methods in the SEI preprocessing operation.In addition,this paper also studies designing a SEI classification system and how to intelligently represent emitter individuals under weakly labeled dataset.To sum up,the main work and contributions of this paper can be summarized as follows:(1)The processing framework and general processing flow of intelligent SEI system are proposed.This paper first analyzes the shortcomings and root causes of traditional SEI system based on expert-defined fingerprint features.On this basis,the processing framework of intelligent SEI system based on deep learning is proposed.Then,the composition and processing flow of the intelligent SEI system are expounded,and the differences and advantages of the intelligent SEI system are analyzed.Finally,a new method for intelligent SEI system based is introduced.(2)The research proposes an intelligent preprocessing method based on deep learning.Aiming at the requirements of signal detection and signal-to-noise ratio estimation in the intelligent identification preprocessing of radiation source in non-cooperative scenarios,this paper introduces the deep learning theory after in-depth analysis of typical signal detection methods and SNR estimation methods.A deep learning based signal detection method and a deep learning based signal to noise ratio estimation method.The experimental results show that the proposed signal detection method and signal-to-noise ratio estimation method are better and more versatile.(3)An intelligent representation method for emitter individual based on deep learning is proposed.In this paper,the generation mechanism of RFF is analyzed firstly.Then,for the emitter individual identification requirements based on transient signal and steady state signal in non-cooperative scenarios,the intelligent representation methods for emitter individual based on transient signal and steady state signal are proposed respectively.Finally,actual experiment data are used to demonstrate that the proposed intelligent representation method for emitter individual is better,more robust and more versatile.In addition,the visual analysis shows that the features extracted based on deep neural network have the characteristics of large inter-class differences and good intra-class aggregation.(4)A classification system with open-set recognition ability is designed in this paper.Aiming at the requirements of the open-set identification task for intelligent SEI system,this paper first analyzes the characteristics of RFFs extracted based on deep neural network,and proposes the corresponding SEI classification system.Then,the research proposes an abnormal target sample detection model based on MGMM and a cluster analysis method based on SD-DPC.The experimental results show that this SEI classification system can effectively find abnormal target samples from the sample to be classified,improve the recognition accuracy of known targets,reduce the generation of fake new targets and extract the core samples of new targets.(5)An intelligent representation method for emitter individual under weakly labeled dataset.Aiming at the task requirements of intelligent SEI system under weakly labeled dataset,this paper first analyzes the reasons why the adversarial training method can improve the generalization ability of deep neural network,then introduces a virtual adversarial training method,and finally expounds the intelligent representation method for emitter individual based on virtual adversarial training under weakly labeled dataset.The experimental results show that the proposed method can effectively improve the performance of enitter individuals based on deep learning methods under weakly labeled dataset.