| With the progress of communication technology,electromagnetic space has become the "fifth dimension of battle space" behind land,sea,air and sky.Electromagnetic confrontation has become an important part of modern military confrontation.Emitter electronic countermeasure is the key point of modern electronic warfare,in which emitter signal feature extraction is an essential prerequisite.The characteristics of radiation source signal include feature engineering and feature learning.By feature engineering and feature reduction,the characteristics of different radiation sources are extracted from signal domain to feature domain,so that it can be applied to different applications.The feature extraction of radiation source signal is mainly oriented to the practical application of signal sorting and individual signal recognition.Radiation source signal sorting is a key technology in modern electronic countermeasures.The signal sorting system can extract characteristic information and de-interleave the received signals of complex radiation sources,realize radiation source identification and threat level assessment,and provide important military information for combat.The traditional method of radiation source signal sorting is to preprocess Pulse parameters and process them according to the interleaving of Pulse Repetition Interval(PRI),so as to achieve signal sorting.However,with the increasing complexity of battlefield electromagnetic environment,the traditional signal sorting method is no longer suitable for the current and future signal sorting requirements.On the other hand,individual identification of radiation source signals after sorting is also an important work.The classical method of extracting signal modulation parameters(such as carrier frequency,bandwidth and other simple characteristics)by communication modulation identification method is not competent enough to meet the needs of modern battlefield.So this paper focuses on the method of radiation source signal sorting and individual recognition.In order to solve the problems of mixing signal processing difficulty and low sorting accuracy in the existing radiation source signal sorting technology,a multi-feature fusion sorting method for Pulse Description Word(PDW)is proposed in this paper,which adopts divide-and-conquer idea and has a two-stage sorting structure.In the first step,a spatio-temporal density clustering model is constructed to separate the radiation source signals that are overlapped in the time-frequency domain.The second step is to further improve the accuracy of the algorithm through the multi-parameter intersection ratio method.This method can effectively separate the time-frequency domain aliasing pulse train,is less affected by PRI,and has higher accuracy than traditional methods.For the problem of large number of radiation sources and changeable modes,a signal sorting method based on Hof transform is proposed by analyzing the time-varying characteristics of pulse phase parameters of radiation sources in Rotating Long Baseline Interferometer(RLBI)system.Firstly,frequency parameter clustering was used for the initial sorting,and hough transform was used to extract the phase time-varying characteristic lines of the pulse with different wave directions.Finally,phase periodic continuation was used to extract the pulse sequence.Simulation results show the effectiveness of the algorithm.Compared with the traditional method,this method can use the phase parameters to directly separate the pulses with different direction directions,which can solve the problem of the decrease of the sorting effect caused by the lack of azimuth information in the signal sorting of the rotating interferometer to a certain extent.To solve the problem that it is difficult to judge the hardware attributes of emitter after signal sorting,this paper sorts out the individual identification process of emitter,and introduces the individual identification method of emitter in four sections: signal preprocessing,feature extraction,feature dimension reduction and individual identification.In addition,non-neural network algorithm and typical neural network algorithm are respectively used to conduct individual recognition technology experiments on experimental data.Finally,random forest and convolutional neural network are selected as the representatives of non-neural network algorithm and neural network algorithm respectively.To sum up,this paper will introduce the basic ideas of radiation source signal sorting and individual recognition in detail.In the sorting part,two unique sorting methods are proposed on the basis of the traditional method,which are proved to be better than the traditional method by experiment.In the part of individual identification,a variety of traditional methods are repeated and compared to screen out better algorithm representatives. |
