@article{M2DF7210B,
title = "Deep Learning-Based Modulation Classification of LPI Radar Signals Using Multi-Antenna Diversity Integration",
journal = "The Journal of Korean Institute of Communications and Information Sciences",
year = "2026",
issn = "1226-4717",
doi = "10.7840/kics.2026.51.1.15",
author = "Sunghwan Cho, Haejoon Jung, Yongchul Kim, Myungsik Lee, Yohan Kim",
keywords = "LPI Radar, Modulation Classification, Diversity gain, Deep Learning, Electronic warfare",
abstract = "This paper proposes a deep learning-based modulation classification system designed to effectively detect and
classify Low Probability of Intercept (LPI) radar signals in Electronic Warfare (EW) environments. While
previous studies primarily focused on signals captured from a single antenna, practical battlefield environments 
typically involve multiple antennas spatially distributed to simultaneously intercept incoming signals. Therefore,
this study emphasizes the utilization of diversity gain achieved through integrating signals from multiple
antennas. To this end, we propose a method that employs various backbone networks, such as CNN,
EfficientNet-B2, and ResNet-50, to extract feature vectors from each antenna signal, followed by feature
integration using methods including LSTM, concatenation, and average pooling. Experimental results integration using methods including LSTM, concatenation, and average pooling. Experimental results
demonstrate that, under challenging low-SNR conditions (specifically, at an SNR of -10 dB), the classification
accuracy significantly improves from approximately 65% with a single CNN-based channel to 87.6% with the
proposed CNN-LSTM-based five-channel diversity system. Thus, our research empirically verifies that the
proposed multi-antenna integration method notably outperforms the conventional single-antenna approach."
}