WASD: Wireless Anomaly Signal Dataset

WASD is a wireless anomaly dataset for spectrum-level anomaly detection in fixed urban environments. It combines measured LTE/5G signals with synthetic anomaly overlays (tone, chirp, pulse) for detection and localization tasks.

Paper

• IEEE Access (2024): https://ieeexplore.ieee.org/document/10813361
• DOI: 10.1109/ACCESS.2024.3521946

Related Works

• Kim et al., "Spectrum Anomaly Detection Using Deep Neural Networks: A Wireless Signal Perspective", IEEE Access (2025): https://doi.org/10.1109/ACCESS.2025.3603216
• Kim et al., "Anomaly Detection for Wireless Cellular Communication based on Synthetic Anomaly", IEEE Access (2025): https://doi.org/10.1109/ACCESS.2025.3584113

Highlights

• Real-world measurements over 19 LTE/5G bands.
• Synthetic anomalies: tone, chirp, pulse.
• Channel/fading-aware synthesis following practical assumptions.
• 85,500 spectrogram samples total (19 bands x 4,500 samples/band).
• Bounding-box labels for anomaly localization.
• Public interactive viewer: https://remilab.ai/wasd/

Quick Visualization (Local)

Generate baseline figures from local sample assets (5G IQ zip, LTE IQ bins, anomaly npy + labels):

python3 scripts/wasd_basic_visualization.py

The basic 5G/LTE STFT figures use the same spectrum crop as the REMI Lab viewer:

• Full FFT: 2048 bins (±30.72 MHz)
• Guard band removed: 224 bins (6.72 MHz) on each side
• Visible spectrum after crop: 1600 bins (48.00 MHz, about ±24.00 MHz)

Generated files:

• images/basic_visualization/wasd_5g_iq_stft_basic.png
• images/basic_visualization/wasd_lte_iq_stft_basic.png
• images/basic_visualization/wasd_anomaly_overlay_basic.png
• images/basic_visualization/wasd_label_summary_basic.png
• images/basic_visualization/basic_visualization_manifest.json

Interactive Viewer (REMI Lab)

• Main viewer: https://remilab.ai/wasd/
• Large viewer: https://remilab.ai/wasd/viewer-large/

Repository Contents

.
|-- WASD_data_loader.ipynb
|-- WASD_abnormal_signal_generation.ipynb
|-- images/
|   `-- basic_visualization/
|       |-- wasd_5g_iq_stft_basic.png
|       |-- wasd_lte_iq_stft_basic.png
|       |-- wasd_anomaly_overlay_basic.png
|       |-- wasd_label_summary_basic.png
|       `-- basic_visualization_manifest.json
|-- scripts/
|   `-- wasd_basic_visualization.py
|-- LICENSE
`-- README.md

Dataset Packaging (Full Release)

Dataset/
|-- IQ data.zip
|   `-- <band_name>/
|       |-- bin/
|       |   `-- IQ_<timestamp>.bin
|       `-- IQ_<timestamp>.json
`-- npy data.zip
    |-- Abnormal/
    |   |-- Abnormal_spectrogram.npy
    |   `-- label/Spectrum_label.csv
    `-- Normal/
        `-- Normal_spectrogram.npy

Data Notes

• Each .npy file stores spectrogram matrices.
• Spectrum_label.csv can contain multiple bounding boxes per spectrogram.
• The number of bounding boxes can exceed the number of spectrogram samples.

IQ Calibration Note

• iq_level_offset: correction offset applied to raw ADC-domain IQ power.
• reference_level: reference level used for absolute calibration.
• Typical conversion path: apply iq_level_offset to convert scale, then convert to dBm as defined by measurement metadata.

Notebooks

• WASD_data_loader.ipynb
  • Example loading of IQ binary, spectrogram .npy, and label .csv.
• WASD_abnormal_signal_generation.ipynb
  • Example synthesis of tone/chirp/pulse anomaly signals and time-frequency visualization.

Access Links

• IEEE DataPort (split dataset for quick evaluation):

  • https://ieee-dataport.org/open-access/wasd-wireless-anomaly-signal-dataset

• Google Drive (full dataset chunks):

  • Chunk 1
  • Chunk 2
  • Chunk 3
  • Chunk 4
  • Chunk 5
  • Chunk 6
  • Chunk 7
  • Chunk 8
  • Chunk 9
  • Chunk 10
  • Chunk 11
  • Chunk 12
  • Chunk 13
  • Chunk 14
  • Chunk 15
  • Chunk 16

Authors

• Jinha Kim: jinha.kim@o.cnu.ac.kr
• Hyeongwoo Kim: hyeongwoo.kim@o.cnu.ac.kr
• Byungkwan Kim: byungkwan.kim@cnu.ac.kr

Acknowledgement

This work was supported by MSIT/IITP through the ICT R&D program (RS-2023-00229541).