Research On Radar Signal Modulation Recognition Based On Time-Frequency Image

The development of cognitive radio and radar electronic warfare raises important requirements for improving the modulation recognition ability of radar signals in complex electromagnetic environments.In the complex spectrum electromagnetic environment,only by fully perceiving and analyzing the use of enemy electromagnetic spectrum resources,can the use efficiency of non-cooperating radar equipment be weakened,and thus be at a commanding height in electronic warfare.However,with the emergence of complex system radar,the existing modulation recognition technology has not been widely applicable to signal recognition frameworks of different modulation types.In this paper,aiming at the radar signal modulation recognition system in complex electromagnetic environment,we focuse on the time-frequency(T-F)analysis,image fusion and deep learning theory to design a modulation recognition algorithm based on T-F image,and further study the radar signal modulation recognition problem with low signal-to-noise ratio(SNR).The main research contents are as follows:Firstly,twelve typical modulation types of radar signals are introduced.And the time domain characteristics,the frequency domain characteristics and the fuzzy performance are analyzed.T-F analysis of non-stationary radar signals is discussed,and the T-F effect of signal with low SNR is studied by reducing interference T-F analysis.Besides,this paper describes the basic theory of deep learning algorithm,introduces the typical deep learning algorithm,and provides a typical theoretical basis for the feature extraction and recognition module of the subsequent modulation recognition algorithm.Secondly,in view of the problems of the existing modulation recognition technology,such as the difficulty in extracting features and the poor recognition effect under the condition of low SNR,a modulation recognition algorithm based on image fusion and convolutional neural network(CNN)was proposed.Based on the T-F analysis of the signal,the algorithm starts from the image fusion theory,analyzes the basic principle of K-L transform,and adopts the image fusion algorithm based on principal component analysis to fuse T-F images obtained by reducing interference distribution(RID).The basic principle of CNN is briefly described.Combing with the transfer learning theory,the pre-training Alex Net network model is investigated.And then,the radar signal modulation recognition algorithm based on T-F image fusion and transfer Alex Net model is proposed,which addresses the issue of training deep networks with small samples,realizes the automation of feature extraction and reduces the complexity of the training.The simulation results demonstrate that the algorithm is better than the traditional algorithm which extracts signal features manually.Finally,in view of the advantages and problems of the proposed recognition algorithm,a modulation recognition algorithm based on fusion features is proposed.Based on the thought of feature fusion,the recognition algorithm employs the Alex Net model based on transfer learning and the stacked autoencoder(SAE)as feature extractors to automatically extract the features of the T-F fusion image.The dimensions of the features which extracted by different feature extractors are reduced by the probabilistic principal component analysis(PPCA)algorithm,which retains the applicable information to the maximum extent,and removes redundant information which is useless for classification decisions.And then,the serial feature fusion algorithm is adopted to fuse the features after dimension reduction.Fusion processing effectively improves the overall recognition success rate(RSR)while ensuring the effectiveness of features.Simulation results demonstrate that the algorithm has good generalization.