Active-jamming Recognition Method Of Radar Based On Bayesian Deep Learning

Accurate recognition of radar jamming signals is the premise to ensure radar antijamming capability,and is crucial to the survival and performance of radar systems.Traditional radar jamming signal identification methods usually require cumbersome analysis and calculation to extract features.These methods have poor versatility and weak generalization ability,and are difficult to adapt to complex and changeable radar working environments.In order to solve these problems,more and more researchers have begun to explore radar jamming signal recognition methods based on deep learning in recent years.Considering that the traditional deep learning network cannot measure the uncertainty in the prediction results well due to the use of point estimation,this thesis proposes a network model based on Bayesian deep learning for Jamming identification.Bayesian estimation is introduced,and the point estimation of network parameter model is replaced by probabilistic modeling,which solves the problem of network overfitting caused by uncertain random data.The Bayesian deep learning network model structure proposed in this thesis is suitable for both one-dimensional and two-dimensional scenarios.The thesis will also conduct experiments from these two dimensions to verify the effectiveness of the proposed network model in radar active jamming identification.The main work of the thesis revolves around the modeling of interference signals,the preparation of data sets,the recognition of one-dimensional and two-dimensional patterns,and the specific research work is as follows:1.Carry out mathematical modeling and data collection for three types of suppression jamming and four types of spoofing jamming based on chirp radar,and give their waveforms in time domain and frequency domain to provide theoretical support for subsequent jamming identification experiments.2.Prepare seven kinds of interference signals into a data set.The data set contains two parts,one-dimensional part,including eight data forms of real part,imaginary part,modulus and phase of the original sequence and the sequence after pulse compression processing;In the two-dimensional part,the interference signal is preprocessed to obtain a time-frequency map and a range-Doppler map.This dataset provides data support for subsequent training and test recognition.3.One-dimensional experiment,the Bayesian deep learning network we proposed,adding the Long Short Term Memory(LSTM)layer can effectively use the timing characteristics of the radar echo signal.Based on the one-dimensional data set,the network training and testing were completed,and a comparative experiment was carried out with the network with the same structure but without the Bayesian method.The results showed that the accuracy of echo modulus recognition increased by 7.36% after the Bayesian learning was introduced.At the same time,after pulse compression processing,the accuracy of real part and imaginary part are increased by 9.40% and 8.82%respectively.4.In the two-dimensional experiment,the same network structure is used to identify the time-frequency diagram of the interference signal and the range-Doppler diagram,and only the convolutional layer is replaced by a two-dimensional convolution.Based on the two-dimensional data set,the network training and testing were completed,and compared with the network with the same structure but without the Bayesian method,the results showed that the recognition accuracy was improved after the introduction of Bayesian learning,and the time-frequency map The recognition accuracy rate reached 99.35%.