Skip to main content

CommRad RF Signals (RFS)

Spectrogram-based communication signal dataset for detection, identification, and classification.

  • Raw data: https://zenodo.org/records/14192970
  • Citation: M. Zahid, “CommRad RF: A dataset of communication radio signals for detection, identification and classification”. Zenodo, Nov. 20, 2024. doi: 10.5281/zenodo.14192970.

Tasks supported​

  • RF signal classification (Mean Per-Class Accuracy)

Class labels​

  • ads-b
  • airband
  • ais
  • automatic-picture-transmission
  • bluetooth
  • cellular
  • digital-audio-broadcasting
  • digital-speech-decoder
  • fm
  • lora
  • morse
  • on-off-keying
  • packet
  • pocsag
  • Radioteletype
  • remote-keyless-entry
  • RS41-Radiosonde
  • sstv
  • vor
  • wifi

Split​

  • 80/20 random split

Preprocessing​

  1. Load each spectrogram image file and rotate 90 degrees.
  2. Convert to RGB if needed, then to tensor and grayscale.
  3. Resize to 224x224 (bicubic).
  4. Normalize with mean 0.5 and std 0.5 (maps to roughly [-1, 1]).
  5. Use the filename prefix before _ as the label and write to HDF5.

Script: preprocess_rfs.py

python preprocessing/preprocess_rfs.py \
  --data-path <RFS_DIR> \
  --output data/rfs.h5

Metric​

  • Mean Per-Class Accuracy on the test split.

Citation​

@dataset{zahid2024commrad,
  title = {CommRad RF: A dataset of communication radio signals for detection, identification and classification},
  author = {Zahid, M.},
  publisher = {Zenodo},
  year = {2024},
  doi = {10.5281/zenodo.14192970}
}