LSTM을 사용한 펄스 레이다 신호 속성 분류

Abstract

In this study, the feature attributes from a received pulse radar signal are defined and certain methods are proposed to classify the attributes using machine learning. The classifier is composed of a long short-term memory (LSTM) network and a fully-connected neural network. As inputs to the LSTM network, pulse repetition interval and radio frequency are used among the pulse description words. The outputs of the classifier can be organized into three types of attributes, i.e., the change pattern of the pulse repetition interval and radio frequency, the number of steps in a period, and the number of pulses in a step. Two classifier structures, single-input and multiple-input, are proposed according to the number of input types. The number of classes varies depending on the classifier structure and the multiple-input classifier classifies 22,410 radar signal attribute sets. The missing pulses, the introduction of any non-desired pulses, and an error in measuring the pulse features are considered as the non-ideal characteristics of the pulse train. The simulation results show that the proposed method shows a high classification accuracy despite the large number of classification options.