Research On Classification And Recognition Methods Of Communication Jamming Signals Based On Deep Neural Networks

The classification and identification of communication jamming signals is one of the key technologies of complex electromagnetic environment sensing,and it is the premise and foundation of communication anti-jamming,which is widely applied in wireless communication and communication electronic warfare.Classification and recognition of communication jamming signals is essentially a pattern recognition problem.The traditional communication jamming recognition algorithms mostly stay in shallow learning,and cannot describe the feature information in the data in detail.The recognition effect is often limited by the noise and signal power.Therefore,in the environment of lower jamming-to-noise ratio(JNR),the recognition rate is often very low.In recent years,deep neural network(DNN)has been successfully applied in many fields,such as image,speech and modulation recognition.Deep neural network can mine the potential features of data,which has the advantages of high recognition rate and less influence by noise.Even when the JNR is low,it can still obtain high recognition rate.Therefore,based on the deep learning theory,this thesis studies the classification and recognition of communication jamming signals.The main work of this paper is as follows:Firstly,the generation mechanism and feature extraction approaches of communication jamming signals are studied.Five communication jamming signals are generated by simulation,including single-tone jamming signals,multi-tone jamming signals,linear sweep jamming signals,partial band jamming signals and noise frequency modulation jamming signals.According to the characteristics of jamming signals in time domain and frequency domain,the features that can highlight the difference of different jamming signals are extracted,including: frequency domain moment kurtosis coefficient,frequency domain moment skewness coefficient,single frequency energy aggregation degree,average spectrum flatness coefficient,frequency domain parameters,time domain moment kurtosis coefficient,and these characteristics are combined and used as the training set.Secondly,using the extracted feature information,a communication jamming signals classification and recognition method based on multi-classification support vector machine(SVM)is realized.The simulation results show that the recognition rate of communication jamming signals classification based on SVM is better when the JNR is high,however,it becomes very low when JNR is low.Then,aiming at the problem of low recognition rate of the SVM-based classification methods under low JNR,a classification and recognition method of communication jamming signals based on deep convolution neural network(CNN)is proposed.The features of five jamming signals are spliced into pictures as the CNN training sets.The training sets are input into the network for training,and the CNN network model is obtained.Using the test set to test,the classification results of five jamming signals are achieved.The simulation results show that the performance of CNN method is better than that of SVM method when JNR is relatively low.However,when JNR is very low and small samples are processed,the recognition rate of the CNN method begins to decline.Although the CNN method has high classification recognition rate,its recognition rate is still to be improved under small samples and much low JNR.In order to solve this problem,a new classification method of communication jamming signals based on deep neural network(DNN)is further proposed.The training set is input into the network containing four hidden layers for training,and the DNN network model is obtained.Then the communication jamming signals to be identified are normalized and feature extracted,the corresponding test set is generated and input into the DNN model,and the classification results of communication jamming signals are obtained.The simulation results show that,under the background of small sample and much low JNR,the recognition rate based on DNN is higher than that based on SVM and CNN.