Research On Modulation Recognition Algorithm Of Radar Signal Based On Deep Neural Networks

With the emergence of various new types of radars,the existing radar signal modulation and identification technology can no longer meet the identification of these new system radars,especially for complex modulated radar signals such as polyphase codes and other low probability of intercept radar signals.In recent years,with the rise of deep learning,deep neural networks have been introduced into radar signal recognition methods.The methods based on deep learning have greatly improved the shortcomings of traditional modulation recognition methods,but there are also some problems.First of all,many methods based on deep learning only perform feature analysis from a single transform domain,which often results in insufficient feature extraction,resulting in poor recognition of some complex modulation radar signals under low signal-to-noise ratios.Feature fusion methods also do not take full advantage of the complementarity between multi-domain features.In addition,many existing methods are not effective in identifying radar signals with low probability of interception such as polyphase codes under low signal-to-noise ratio.In view of the above problems,the main work and innovations of this paper are as follows:Firstly,a single time-domain or frequency-domain analysis cannot effectively identify radar signals with complex modulation methods,and a radar signal identification method based on multi-domain feature fusion is designed.This method combines the ideas of feature selection and feature fusion.A number of statistical features are extracted in the time domain and frequency domain respectively,and the feature selection method is used to remove redundancy and select the most effective feature for subsequent feature fusion.In the time-frequency domain,this paper selects a pre-trained deep neural network to extract the features of timefrequency images.Finally,the features of the three domains are fused for classification.The simulation results of the algorithm show that the algorithm can effectively identify 7 types of radar signals including polyphase codes.Secondly,aiming at the problem of poor recognition of signals such as polyphase codes under the condition of low signal-to-noise ratio,a residual attention network structure is designed,which combines the residual unit and channel attention sum and spatial attention,using The feature redirection function of the attention mechanism increases the weight of useful time-frequency features to suppress the noise weight under low SNR,thereby further improving the recognition accuracy of complex types of radar signals under low SNR.The experimental results show that the algorithm still has good anti-noise performance and generalization performance at low SNR.Finally,in order to verify the realizability of the algorithm proposed in this paper in the actual scene,a demonstration platform system based on radar signal modulation recognition was built through the joint development of software and hardware by software development executable software and ADALM-Pluto portable software radio equipment.This demonstration platform verifies the feasibility and effectiveness of the proposed algorithm in real scenarios.