Modulation Recognition And Parameter Extraction Of Radar In Pulse Signal Based On Depth Learning

With the rapid development of information technology and electronic warfare(EW),it is very worthwhile to accurately identify the modulation mode of radar pulse signal(RPMS)and invert the corresponding modulation parameters.The level of its level affects the performance of reconnaissance and electronic Adversarial effectiveness and success rate.Aiming at the problems of low accuracy of radar modulation pattern recognition and insufficient parameter inversion ability in low signal-to-noise ratio environment,this paper proposes a neural network model based on U-shaped neural network(Unet)for noise reduction and recognition of radar intrapulse signals using semantic segmentation(RDLnet).After the successful identification of the modulation mode,the frequency modulation and phase modulation signals in the low signal-to-noise ratio environment are extracted by the least squares method with genetic factors and the neural network designed based on long short-term memory network(LSTM),respectively.The main research content as follows:(1)Propose RDLnet,in which the combined structure of the auto-encoder(Encoder)and the decoder(Decoder)can expand the receptive field and integrate the multi-scale features of the central part while retaining the detailed information,so as to preserve the original signal structure more completely,energy distribution and semantic information.Integrating the attention mechanism into the network and obtaining a larger receptive field of the modulation signal makes the Encoder contain more useful modulation feature information.The noise-free map is recovered by deconvolution in the Decoder.The hierarchical distribution of the Decoder module makes the semantic mapping reference more stable and ensures the stability of RDLnet training.After testing,the network output noise reduction pictures have good numerical performance in terms of peak signal-to-noise ratio(PSNR),structural similarity(SSIM)and recognition rate of low signal-to-noise ratio RPMS.In the-10 db environment,the PSNR value of the time-frequency image obtained after RDLnet noise reduction can reach more than 50,and the structural similarity with the original image can exceed98.9%.The pattern recognition rate under-10 db can still exceed 98.6%,which significantly improves the recognition accuracy of traditional neural networks in a low signal-to-noise ratio environment.(2)Through the semantic segmentation network,high robustness can also be achieved in the training of small sample data,which solves the problems of sparse collection of radar data samples and insufficient labeling.Using the Encoder and Decoder combined structure can solve the problem of heavy training tasks caused by the heavy use of deep convolutional neural networks(DCNN)in traditional semantic segmentation.(3)In the parameter extraction,in order to solve the problem that the timefrequency data of the radar pulse signal contains noise data.A least squares method with forgetting factor is proposed to fit the time-frequency data of FM signals.Experiments show that this method has good anti-interference performance for FM signal data in low signal-to-noise ratio environment,and the average error between the extraction results of FM signal modulation parameters and the real value is 3.71%.For the phase-modulated signal,a neural network is designed based on LSTM to extract the carrier frequency parameters of the phase-modulated signal.Experiments show that the error rate of parameter identification is 6.37% under the low signal-to-noise ratio of-10 db,which provides a new solution for the parameter extraction of phase-modulated signals.