Radar Modulation Recognition of Intra-Pulse Overlapping Signals Based on Object Detection

Abstract

Existing radar modulation recognition research predominantly assumes the presence of a single intra-pulse signal, despite the frequent occurrence of overlapping or interleaving of two pulse signals. This article addresses the challenging problem of modulation recognition for overlapping intra-pulse signals, where the complexity arises from the stepwise growth in the number of samples due to the permutation of subsignals. For the first time, a series of object detection methods is employed to tackle this issue. Specifically, focusing on air-to-ground reconnaissance systems, we construct a time-frequency dataset comprising overlapping signals. Using this dataset, we successfully detect dual signals while training solely on single pulse signals. Mainstream frameworks, such as “You Only Look Once” (YOLO) and region-based convolutional neural network (RCNN) are explored, and an improved YOLOv7 algorithm is proposed, incorporating DCNv2 and explicit visual center block. Experimental results demonstrate a 91% recognition rate for overlapping signals at −4 dB, surpassing state-of-the-art methods. Notably, this article introduces a new dataset of overlapping radar signals, which can be publicly accessible at IEEEDataPort https://ieee-dataport.org/documents/radar-modulation-recognition-intra-pulse-overlapping-signals.