Radar Emitter Signal Sorting And Recognition In Complex Environment

As the electromagnetic environment becomes more and more complex,the pulse flow density becomes larger and larger,and the waveform of modern radar signal becomes more and more complex,the signal sorting method using traditional fiveparameter sorting can not meet the requirements.It is necessary to propose some new methods to sort and identify radar signals.In this thesis,the sorting and recognition of radar signals are studied.Using the knowledge of clustering and deep learning,the radar signals are sorted and recognized,and good results are obtained.The work is as follows:1.The radar signals with different intra-pulse modulation modes and different pulse repetition interval working modes are simulated and analyzed.The algorithms CDIF,SDIF and PRI transformation algorithms using PRI parameters are studied and simulated,and how to improve the PRI algorithm to extract PRI eigenvalues in the case of large jitter is studied.2.Different intra-pulse features of radar emitter signals are extracted,including complexity features,ambiguity function features and bispectral similarity coefficient features.The complexity features are mainly box dimension and sparsity.These intrapulse features of different radar signals have great differences and high discrimination between classes.3.A new sorting model is used to sort radar signals.The core cluster support clustering(CCSVC)method is used to comprehensively complete radar signal sorting by using inter-pulse parameters and intra-pulse features.First,the RSVC clustering algorithm is used to cluster the parameters between pulses to obtain as many missed pulses as possible,with a low false selection rate.Then the missed pulses are clustered by SE-MSVC clustering algorithm by using intra-pulse features.The results of inter-pulse parameter clustering and intra-pulse feature clustering are combined to finally complete the sorting of radar signals,and the accuracy is more than 98%.4.A framework is explored to extract the time-frequency characteristics of signals.Convolutional neural network is used to classify and recognize different modulation signals,and a high recognition accuracy is obtained;The smooth pseudo Wigner-Ville distribution characteristics are extracted,the migration learning is used to identify different types of radar signals,the high-precision deep learning network is used to identify the signals,the recognition accuracy is high,and the lightweight network is used to identify,which reduces the amount of parameters and improves the recognition efficiency.5.The improved synchronous transform(IMSST)is studied,the IMSST features of different signals are extracted,and the extracted IMSST features are used to recognize the radar signal through the Efficient Net network model,and finally a relatively high recognition accuracy is obtained.