Research On Methods Of Multi-component Radar Signal Intra-pulse Modulation Recognition

Intra-pulse modulation recognition of radar signals is an important part of modern electronic intelligence reconnaissance and electronic support systems.Due to the increasingly complex environment of modern electronic warfare,radar reconnaissance system often intercepts overlapping pulses in the time domain to form multi-component radar signals.Most of the existing radar signal modulation recognition technologies are not adaptive to the multicomponent radar signal environment,resulting in the failure of radar signal recognition.Therefore,the analysis and processing of multi-component radar signal is an urgent problem to be solved in the current radar reconnaissance system.In this paper,by studying timefrequency analysis method,time-frequency image feature extraction method and multi-label classification method for multi-component radar signal,a multi-component radar intra modulation recognition algorithm with high recognition accuracy and wide adaptability to radar signal types in low signal-to-noise-ratio(SNR)environment is proposed.The main contents of this paper are as follows:Firstly,for the time-frequency analysis of multi-component radar signals,this paper proposes a Cohen class time-frequency analysis method based on multiple kernel functions to obtain the time-frequency images of multi-component radar signals.Firstly,according to the energy distribution characteristics of radar signals with different intra-pulse modulation modes in the ambiguity domain,several targeted kernel functions are designed,and then multiple time-frequency images are obtained based on these kernel functions.Finally,multiple time-frequency images are fused by combining the multi-exposure image fusion method.The simulation results show that the kernel functions designed in this paper can improve the time-frequency aggregation of some radar signals in the time-frequency images.The fusion time-frequency image obtained by multi-exposure image fusion method retains the main characteristics of signal energy in multiple time-frequency images,and obtains good time-frequency analysis results.Secondly,for the feature extraction of the time-frequency images,this paper designs a feature extraction network based on a convolutional neural network structure.According to the time-frequency image input,this paper designs the network structure of the convolution part of the feature extraction network.Besides,aiming at the problem of multi-component radar signal recognition,the output part of the convolutional neural network is expanded to get the initial recognition results of the recognition system,and the training of the timefrequency image feature extraction network is realized.The simulation results show that the recognition system has a good initial recognition performance and that the trained feature extraction network has good adaptability to the time-frequency image data.Finally,for the classification of the multi-component radar signals,this paper designs the internal structure of the multi-label classification network based on recurrent neural network and trains the multi-label classification network based on reinforcement learning.Firstly,the feature vector corresponding to the radar signal is obtained through the feature extraction part of the recognition system.Then the feature vector is used as the input of the multi label classification network,the reward value of each radar signal type is calculated and output through the network model,and the radar signal type with the highest reward value is selected as the recognition result of the current radar signal component.Finally,the final recognition result set of the recognition system is obtained by the iterative process of the recurrent neural network.The simulation results show that the performance of the recognition system can be effectively improved by combining the multi-label classification network.The multi-component radar signal intra-pulse modulation mode recognition method proposed in this paper realizes the recognition of multi-component radar signals formed by the random overlapping of 8 typical radar signals and achieves good recognition performance in low signal-to-noise-ratio environment.Meanwhile,the algorithm is also adaptive to the recognition of single-component radar signals.