An Intelligent Recognition Method for Radar Comb Spectrum Jamming Based on Dual-Channel Deep Convolutional Network

Abstract

This paper presents a deep learning method to recognize comb spectrum jamming in radar systems. Unlike traditional methods requiring manual feature extraction, our approach learns features directly from signal data. We built a dataset of radar echoes with four comb jamming types and five non-comb interference types. A dual-channel method creates 2D images preserving both magnitude and phase information from the signal spectrum. A CNN classifier with convolutional blocks, batch normalization, and dropout achieves 99.75% accuracy with 1.5% false alarm rate after only 7 training epochs.