Automatic Modulation Classification with Deep Neural Networks

Download Dataset

The dataset can be downloaded from https://www.deepsig.ai/datasets.

We use the RADIOML 2018.01A dataset.

Setup

To run our code as a package, we provide a setup.py file. To run the setup, inside this directory, run:

python setup.py develop

Prepare Dataset

Our code assumes the data to be in TFRecord format; however, the dataset provided by DeepSig is in hdf5 format. To convert the dataset using our train/test split, run the script process_data.py. For example:

process_data.py --src_dataset_path="./dataset/2018.01/GOLD_XYZ_OSC.0001_1024.hdf5" --dest_dataset_path="./dataset/GOLD_XYZ_OSC_tfrecord"

To perform the same operation with less code and put the data directly into tensorflow_datasets (tfds) format, please see another repository of ours: https://github.com/caharper/smart-tfrecord-writer/tree/main

With the following example: https://github.com/caharper/smart-tfrecord-writer/blob/main/examples/radioml/radioml_to_tfrecord.py

We will use this code in future work as it is more flexible and easier to use. Particularly, loading the data becomes much easier with tfds using tfds.load().

Run Code

To run all experiments using Bocas, run the following inside the ./experiments/radioml2018.01a/ directory:

python3 -m bocas.launch run.py --task run.py --config configs/sweep-models.py

Citation

If you make use of this work or use our dataset split, please cite our work:

@Article{electronics12183962,
  AUTHOR = {Harper, Clayton A. and Thornton, Mitchell A. and Larson, Eric C.},
  TITLE = {Automatic Modulation Classification with Deep Neural Networks},
  JOURNAL = {Electronics},
  VOLUME = {12},
  YEAR = {2023},
  NUMBER = {18},
  ARTICLE-NUMBER = {3962},
  URL = {https://www.mdpi.com/2079-9292/12/18/3962},
  ISSN = {2079-9292},
  DOI = {10.3390/electronics12183962}
}

The link to our paper can be found here: https://www.mdpi.com/2079-9292/12/18/3962

The preprint can be found here: https://arxiv.org/abs/2301.11773