Research On Airborne Radar Signal Sorting Based On Convolutional Neural Network

The main purpose of airborne radar signal sorting is to effectively separate highvalue air-radiation source signals from the intercepted mixed pulse stream.It affects the successful implementation of the subsequent recognition and analysis of high-threat airborne targets,so as to provide credible intelligence information,combat strategy formulation and other important combat methods.It's the core of the radar counterreconnaissance system,laying a solid foundation for the smooth seizure of "electromagnetic power" and "air dominance".It has already become a research hotspot.This thesis plays the research emphasis on the airborne radar signal sorting,especially the extraction of intra-pulse features of airborne radar signals by convolutional neural network.Then,with the analysis foundation of the features of airborne radar signals,the deep learning methods are introduced to modify the traditional airborne radar signal sorting structure.It improves the generalization performance of the developed features,alleviates the cumulative error of traditional signal sorting approach,and abates the reliance of signal sorting on manual experience.So the airborne radar signal sorting can be more intelligent.The main work of present thesis can be concluded as follows:1.Aiming at the problem that traditional parameters can no longer effectively when the airborne radar signal is complex,on the basis of one-dimensional convolutional neural network(1D-CNN),a new approach for the extraction of intra-pulse features of radar signals was developed.First,we establish a feature extraction model based on 1D-CNN,and design the structure of the proposed 1D-CNN,extracting the intra-pulse feature as an auxiliary feature for airborne radar signal sorting.2.Aiming at the problem of sorting dense pulse streams in a complex electromagnetic environment,considering the parameter features of airborne radar signals and the extracted pulse description words,this thesis utilized the K-means clustering algorithm for conducting multi-parameter joint clustering,which improves the selection of clustering center at start.This method realizes the dilution of the dense pulse stream in the airspace and relieves the pressure of later processing.3.Aiming at the problem that a single feature can no longer effectively deal with the sorting of complex airborne radar signals,a joint sorting method is proposed with both intra-pulse features and inter-pulse features considered.And a new structure of airborne radar signal sorting is established.Then,the intra-pulse features are extracted by the proposed 1D-CNN.Combined with the sequence difference histogram algorithm,the inter-pulse features are estimated and analyzed,so that the proposed joint sorting method is realized.In the end,on the basis of simulations,the validity of the developed approach are verified,which can provide accurate airborne radar signal sorting.