Research And Application Of Radar Radiation Source Signal Sorting Technology Based On Machine Learning

The sorting of radar emitter signals refers to a process where the pulse sequence of each radar is extracted for the estimation and identification of parameters for each of them in the presence of staggered pulse fr-om multiple radars.At present,the application of novel radars in contemporary battlefield breaks the old-fashioned patterns of such characteristic parameters as carrier frequency,pulse repetition interval and so on.In the meantime,staggered PRI radar pulse,RF jittering radar and the likes emerge in succession,which is accompanied by the increasingly complex electromagnetic environment on the battlefield.In this case,the density of radar emitter signals detected by the receiver simultaneously increases,which makes it difficult for the conventional algorithms reliant on TOA and PRI parameters for sorting to meet the requirements of contemporary electronic warfare on the radar signal sorting system.In order to solve the aforementioned problems,there have been some researchers seeking breakthrough to the conventional algorithms for adapting to the characteristics of signals released by novel radars.Based on the approaches to machine learning,a study is conducted in this paper on the techniques applied to the primary sorting of signals from both known and unknown radar emitters according to the data collected on the radiation sources of radar.The pulse description word of aliased signals is extracted according to the classification of radiation source,which lays a foundation for the sorting of radar emitter signals both theoretically and technically in a complicated setting.The major details are presented below.As for the primary sorting of signals from known radar emitters,decision tree,random forest,XGBoost,Plain Bayes,K-nearest neighbor and other supervised learning algorithms are applied to conduct simulation experiments,for exploring its sorting performance given the proportion of various testing sets in the full set.According to the experimental results,random forest and XGBoost algorithms perform best,with the accuracy of sorting maintained around 95%.Besides,AUC exceeds 0.995 when the training set accounts for 70%.At this time,the accuracy reaches up to 95.4%.As for the primary sorting of signals from unknown radar emitters,Density-Based Spatial Clustering of Applications with Noise is adopted to perform sorting.In the meantime,kernel density estimation is introduced into DBSCAN for automatic determination of the original parameters required by algorithm.As indicated by the experimental results,the algorithm is capable to differentiate the pulse signals released by various unknown radar emitters,with the sorting accuracy finally reaching 92.887%.In order to address such problems encountered by DBSCAM as heavy workload of calculation and lengthy computing,this paper proposes a combined sorting algorithm on the basis of clustering algorithm and supervised learning.According to the experimental results,the combined sorting algorithm achieves a 92.387%accuracy,and improves by 34.7%in operating speed compared with DBSCAN,which makes it more capable to meet the requirement of contemporary battlefield on timeliness.