RGCN-S2G: A Graph Structure Learning Framework for Robust Radar Intrapulse Modulation Recognition
IEEE Transactions on Aerospace and Electronic Systems · 2026 · 62 · 11433-11451
DOI: 10.1109/taes.2026.3696254 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Aerospace and Electronic Systems, 2026. Verfasst von Liting Sun und 2 weiteren Autor:innen. Gefördert von: National Natural Science Foundation of China, Innovation Research Foundation of National University of Defense Technology.
Abstract (Originalsprache)
To address the challenges of robustness and generalization in radar intrapulse modulation recognition (IPMR) under complex electromagnetic environments, this article proposes the radar graph convolutional network (RGCN) with signal-to-graph (S2G) framework. This framework introduces graph structure learning to IPMR to enhance resistance against noise and multipath fading and enables direct application to unseen time-varying multipath (TVMP) fading channels. Specifically, an S2G representation paradigm is established, which models the time–frequency coupling characteristics by constructing graph nodes and edges from the differential information of signal segments in both time and frequency domains. This adaptively generates a topological representation with physical interpretability and inherent robustness. For classification, an RGCN is designed. It employs dual-channel complex graph convolution to collaboratively extract features from the I/Q components, augmented with residual connections and a differentiable pooling mechanism to improve gradient stability and structural compression. Experimental results on 18 modulation types, including composite modulation, show that the proposed method achieves an accuracy exceeding 96% under 0 dB additive white Gaussian noise (AWGN). Moreover, without retraining, it successfully generalizes from AWGN to various TVMP channels, attaining an average accuracy of 88.40% across six distinct TVMP scenarios with different statistical characteristics and different paths, at 0 dB. This work demonstrates a promising approach for practical IPMR in dynamic electromagnetic environments.
Förderung
- National Natural Science Foundation of China
- Innovation Research Foundation of National University of Defense Technology
Zitieren
APA
Sun, L., Liu, Z., & Huang, Z. (2026). RGCN-S2G: A Graph Structure Learning Framework for Robust Radar Intrapulse Modulation Recognition. IEEE Transactions on Aerospace and Electronic Systems, 62, 11433-11451. https://doi.org/10.1109/taes.2026.3696254
Vancouver
Sun L, Liu Z, Huang Z. RGCN-S2G: A Graph Structure Learning Framework for Robust Radar Intrapulse Modulation Recognition. IEEE Transactions on Aerospace and Electronic Systems. 2026;62:11433-11451. doi:10.1109/taes.2026.3696254
BibTeX
@article{Sun2026,
author = {Sun, Liting and Liu, Zheng and Huang, Zhitao},
title = {RGCN-S2G: A Graph Structure Learning Framework for Robust Radar Intrapulse Modulation Recognition},
journal = {IEEE Transactions on Aerospace and Electronic Systems},
year = {2026},
volume = {62},
pages = {11433--11451},
doi = {10.1109/taes.2026.3696254}
}
EUVIMED