license: other
tags:
- vision-transformer
- radar
- signal-processing
- electronic-warfare
- modulation-recognition
GC-ViT: AI-Based Global Context Vision Transformer for Radar Signal Modulation Recognition
An AI-based Global Context Vision Transformer (GC-ViT) that leverages the short-time Fourier transform (STFT) phase spectrum for feature extraction to identify phase-coded radar waveforms — a key capability for electronic warfare (EW) systems facing the growing use of low-probability-of-intercept (LPI) radars. Combines local and global self-attention to improve recognition robustness at low SNR.
Dataset: (https://www.kaggle.com/datasets/sidrabhatti/latestdataset-cnn)
Role & Attribution
Sidra Ghayour Bhatti — first author; conceived, implemented, and evaluated the GC-ViT model. Co-authored with Mohsin Ullah (Dept. of Electrical Engineering, Capital University of Science and Technology, Islamabad).
Method
Input representation. Each intercepted radar pulse is converted to its short-time Fourier transform (STFT) phase spectrum rather than the more commonly used magnitude spectrum — the deliberate choice is because phase-coded waveforms (Barker codes, P1–P4 polyphase codes, etc.) are defined by their intrapulse phase-modulation pattern, not by how their energy is distributed across frequency. Two different phase codes can have near-identical magnitude spectra while differing sharply in phase, so a magnitude-only spectrogram throws away the very information that distinguishes the waveform classes; the phase spectrum keeps it. The resulting 2D time-frequency phase map is cropped to its informative region and resized to 224×224×3, turning a 1D IQ waveform recognition problem into a 2D image classification problem that a vision architecture can exploit.
Model. A GC-ViT (Global Context Vision Transformer, Tiny variant) backbone, pretrained then fine-tuned with a 6-class softmax classification head for phase-coded waveform families. GC-ViT combines local window-based self-attention (fine-grained spectrogram texture) with a global query token (long-range structure across the full time-frequency map) in each block — the architectural reason it outperforms CNN baselines at low SNR, where discriminative structure is spread across the spectrogram rather than localized.
Training. Fine-tuned with SGD (lr=0.001) and sparse categorical cross-entropy loss on top of GC-ViT's pretrained backbone.
Evaluation protocol. Tested across a wide SNR sweep (roughly −14 dB to +8 dB in 2 dB steps) with held-out test sets per SNR level, so performance is reported as a function of noise level rather than a single aggregate number — directly relevant to EW receivers that must operate across unknown, often very low, intercept SNRs.
Results
- ~80% recognition accuracy at −12 dB SNR, substantially outperforming prior methods at that noise level
- Performance degrades gracefully across the full tested SNR sweep rather than collapsing sharply, consistent with the global self-attention mechanism recovering structure that local-only (CNN) models miss at low SNR
- Demonstrates robustness for EW situational-awareness applications in complex electromagnetic environments
Code
code/Global_Context_ViT_for_RadarSiganls.ipynb — the training/evaluation
notebook used for this paper. It's Colab-specific (mounts Google Drive and
loads a locally-trained .h5 weights file not included here), so it won't
run standalone, but it documents the exact preprocessing, model setup, and
SNR-sweep evaluation protocol behind the reported results. Training data:
see the Kaggle dataset link above.
Published: Bhatti, S.G. & Ullah, M. (2024). "Radar signal modulation identification using global context vision transformer." Engineering Research Express, 6(4), 045331. doi.org/10.1088/2631-8695/ad8b96 (subscription required — not open access, so no PDF is included here; link only)
Citation
@article{bhatti2024radar,
author = {Bhatti, Sidra Ghayour and Ullah, Mohsin},
title = {Radar signal modulation identification using global context vision transformer},
journal = {Engineering Research Express},
volume = {6},
number = {4},
pages = {045331},
year = {2024},
doi = {10.1088/2631-8695/ad8b96}
}