Skip to main content

RADCOM (Modulation & Signal Type)

Dataset for modulation and signal type classification of radar and radio wavesforms supporting single-task and multi-task learning.

Tasks supported​

  • Modulation classification / signal type classification

Class labels​

  • Modulation labels: AM-DSB, AM-SSB, ASK, BPSK, FMCW, PULSED.
  • Signal-type labels: AM radio, short-range, SATCOM, Radar Altimeter, Air-Ground-MTI, Airborne-detection, Airborne-range, Ground mapping.
  • The cache stores combined (modulation, signal_type) pairs for multi-task training.

Split​

  • 80/20 random split

Preprocessing​

  1. Read tuple-keyed samples (modulation, signal_type, snr, sample_idx) from HDF5.
  2. Split each sample into I and Q channels (first 128, second 128).
  3. Compute global I/Q mean and std across all samples (first pass).
  4. Normalize each sample and store with labels (modulation + signal type).
  5. Save flattened cache with modulation, signal type, and SNR metadata.

Script: preprocess_radcom.py

python preprocessing/preprocess_radcom.py \
  --input <RADCOM_RAW_H5> \
  --output data/radcom.h5

Metric​

  • Mean Per-Class Accuracy per task (report modulation and signal type).

Citations​

@article{AjagannathCOMNET21,
  title = {Dataset for modulation classification and signal type classification for multi-task and single task learning},
  journal = {Computer Networks},
  volume = {199},
  pages = {108441},
  year = {2021},
  doi = {10.1016/j.comnet.2021.108441},
  author = {Anu Jagannath and Jithin Jagannath}
}

@inproceedings{JagannathMTL,
  title={{Multi-task Learning Approach for Automatic Modulation and Wireless Signal Classification}},
  author={Jagannath, Anu and Jagannath, Jithin},
  booktitle = {Proc. of IEEE International Conference on Communications (ICC)},
  address = {Montreal, Canada},
  month = {June},
  year = {2021}
}