Research On Application Of Ensemble Clustering For Sorting Radar Signals

With the increasing complexity of modern radar electronic countermeasures environment,the pulse density of radar increases rapidly,the interweaving is serious,and the modulation modes are complex and diverse,the radar emitter signal sorting technology is required to be higher.The existing radar signal cluster-sorting methods often adopt a single clustering method.However,these clustering methods have their own specific application scenarios,and it is difficult to process radar signal data in complex radar electromagnetic environment,which leads to the low sorting accuracy.Aiming at the above problems,this paper introduces the idea of ensemble clustering,combined with Graph Autoencoder,and the sorting model and method of radar emitter signals are studied in two scenes with no prior knowledge and with prior knowledge.The prior knowledge of radar signals generally includes the types,characteristics and parameters of radar signals.The research work and results of this paper are as follows.(1)without prior knowledge,a novel Radar Signal Sorting method based on Ensemble Clustering Without Radar Rules(ECWRRRSS)is proposed.The model is mainly composed of ensemble clustering module and Graph Autoencoder module.Specifically,it contains the construction of cluster components generator and filter,and the calculation of similarity matrix and the construction of the graph,and the construction of Graph Autoencoder.The idea of sorting is as follows: the radar signal dataset is converted from European data to graph structure data by the method of ensemble clustering,and the graph is input to Graph Autoencoder to learn,and the extracted new node representations are clustered using KMeans algorithm,and the radar signal sorting is completed.Experimental results show that,compared with 12 comparison methods,this method effectively improves the sorting accuracy and clustering performance of radar signals.The innovation of this method is that the ensemble clustering and Graph Autoencoder are introduced into radar signal sorting for the first time,and the similarity matrix is weighted by Gaussian distance.The disadvantage is that the stability of the model in complex environment is not high..(2)with prior knowledge,a novel Radar Signal Sorting method Combining Radar Rules and Ensemble Clustering(CRRECRSS)is proposed.This model is improved on the ECWRRRSS model,and the improvements are as follows: the Gaussian distance weighting module is eliminated,and the Graph Autoencoder model and the semi-supervised clustering model that introduce constrainted sample pairs are constructed respectively.Specifically,the radar rule table is constructed,constraint sample pairs are extracted,a new loss function is designed to design a new Graph Autoencoder model,and the extracted new node representations are semi-supervised clustered to complete the radar signal sorting.Experimental results show that,compared with the ECWRRRSS method,this method can further improve radar signal sorting accuracy and clustering performance,and it has advantages in model stability.The innovation of this method lies in introducing radar prior knowledge into the radar signal sorting model.The difficulty are that the establishment of radar rule table and how to integrate radar rules into radar signal sorting model.