Research On Intelligent Cognition Method Of Radar Electromagnetic Environment

The continuous improvement of electronic countermeasures technology and the increasingly complex electromagnetic environment of the informational battlefield has been made the acquisition of key electromagnetic information a decisive factor for the victory of electronic warfare.In the electromagnetic environment with complex and changeable active interference signals,radar needs to accurately and quickly distinguishing the interference categories in the received signals in order to select the most suitable anti-interference measures.Deep learning(DL)with powerful learning ability can extract different features between signals.Combined with DL,radar can quickly determine which category the signal belongs to,which has certain practical significance.The main work of this thesis is to study the radar active interference signals present in the electromagnetic environment.By analyzing the generation mechanism and working characteristics of the signals,the identification and classification methods of radar active interference are discussed.Using the time-domain characteristics and time-frequency diagram of the signal,combined with deep learning,the target echo signal and 13 kinds of active interference signals are identified and classified.The main applied deep learning algorithms are Convolution Neural Network(CNN)and Long Short-Term Memory(LSTM).Considering the limited characteristic dimensions of the signals extracted from a single network model,this thesis proposes a cascading model of CNN and LSTM,which improves the recognition accuracy of various types of interference signals.The specific work and research results of the thesis are as follows:1.The generation mechanism and working characteristics of echo signals and interference signals are analyzed,modeled,and simulated,which are specifically two suppressive jammers,seven deceptive jammers,and four composite jammers.Then,according to the time-frequency analysis,the time-frequency distribution diagrams of 14 kinds of signals superimposed with Gaussian white noise between pulses and pulses are obtained.Finally,the intrapulse time-frequency distribution map of the signal is preprocessed,and the preprocessed images and time series sequences are respectively constructed into data sets,which lays a data foundation for the subsequent interference identification and classification.2.This thesis studies the active interference recognition and classification algorithm based on CNN and LSTM.Firstly,the theoretical model designs process of CNN and LSTM is analyzed,and three algorithms are proposed: classification algorithm based on signal time series dataset and 1-D CNN model,algorithm based on signal time series datasets and LSTM models,and algorithm based on time-frequency image datasets and 2-D CNN models.Then the three algorithms are simulated respectively to analyze the classification accuracy of different interference signals under different interference-to-noise ratio conditions.Finally,the three proposed classification algorithms are compared and analyzed.3.Since the single neural network model has poor classification ability on the composite interference signal existing in the electromagnetic environment,this thesis adopts the composite neural network model to improve the classification ability of the composite signals.The series and parallel models of LSTM and CNN are proposed.Aiming at the different cascading methods,the classification accuracy of the model under different interference-to-noise ratio conditions and the classification ability of different interference signals is simulated and analyzed.The computational complexity,operation time,classification effect etc.be compared and analyzed.The simulation results show that the two cascaded models can improve the classification effect of various signals compared with the single model,where the parallel model has better classification effect,and the series model has less computing time and computational complexity.