Research On Recognition Algorithm Of Radar Emitter Sequence Signal Based On Deep Learning

Radar emitter recognition is a key link and an important part of electronic intelligence reconnaissance,and its level has become an important symbol to measure the technical level of electronic reconnaissance.Quickly and accurately identify the modulation type of the enemy radar signal,which can effectively obtain battlefield information and situation,and provide good support for subsequent decision-making.At present,the electromagnetic environment of the battlefield is becoming more and more complex,and the electromagnetic environment has dense signals and many sources of interference.The new system radar gradually replaces the traditional radar,and the modulation mode of the radar emitter signal is gradually complicated and diverse.Traditional recognition methods has been difficult to meet the needs of radar emitter recognition in the current the electromagnetic environment of the battlefield.Therefore,this paper proposes a radar emitter sequence signal recognition algorithm based on deep learning: using the deep learning method to extract the features of the sequence signal,classify and identify,and propose solutions to the problem of imbalanced data sets in reality.The specific work is as follows:1.Signal modulation types and basic knowledge of deep learning.In this paper,eight kinds of radar emitter intrapulse modulation signals,including BPSK,Costas,FMCW,Frank,P1,P2,P3 and P4,are studied.The signal models are established respectively to analyze the modulation signals in the time domain,frequency domain and timefrequency domain.In addition,this paper summarizes two routes based on deep learning methods and the relevant basic theories of deep learning,including convolutional neural network and cyclic neural network,and analyzes the advantages and disadvantages of the two networks.2.In view of the low accuracy of radar emitter recognition under the condition of low SNR,a radar emitter recognition method,SACNN,is proposed in this paper.Based on SRNN,the recognition effect is improved effectively.Firstly,one-dimensional radar emitter signals are preprocessed,and the processed signals are input into SACNN to complete classification and identification.Under the condition of SNR=-10 d B,the recognition accuracy of eight kinds of radar emitter signals is 100%.Under the condition of SNR =-20 d B,the average recognition accuracy is more than 80%.The experimental results show that,under the condition of low SNR,the recognition effect is significantly improved compared with some existing methods,and the model can take into account the requirements of accuracy and speed.3.Aiming at the unbalanced real data sample category,this paper proposes a method of generating radar emitter sequence signal based on1D-CWGAN-GP.The 1D-CWGAN-GP generation model was designed and built,combined with WGAN-GP and CGAN,and the generator network was improved.The U-shaped structure composed of upper sampling module and lower sampling module replaced the fully connected network,so that it could generate one-dimensional radar emitter signal and improve its generation effect.Eight kinds of radar emitter signal generation experiments are carried out.The validity of the signal generated by the signal generation model is analyzed and verified by time-frequency image,sample generation similarity experiment and sample generation substitution experiment.Experimental results show that the established1D-CWGAN-GP signal generation model can effectively generate radar emitter signals,which can replace real data for deep learning training.Moreover,it still has good generation effect under the condition of a small number of samples training.