RadioML 2016.04 Dataset

Dataset Contents

The dataset contains 11 modulation types generated using GNU Radio’s digital and analog modulators.


Channel Effect Variations

Version 2016.04A

This is a high SNR dataset with gaussian noise only.  A relatively clean starting point for basic experimentation.

Input Shape: 2×128

Labels: Modulation-Type

Version 2016.04B

This is a low SNR dataset with gaussian noise, clock and center frequency drift, multi-path fading components, and relatively poor SNR.

Input Shape: 2×128

Labels: Modulation-Type

Version 2016.04C

This is a variable SNR dataset with moderate LO drift, light fading, and numerous different labeled SNR increments for use in measuring performance across different signal and noise power scenarios.

Input Shape: 2×128

Labels: Modulation-Type, SNR

Download Dataset


Relevant Studies

This dataset was used for both [1] “Convolutional Radio Modulation Recognition Networks” and [2] “Unsupervised Representation Learning of Structured Radio Communications Signals


Dataset Visualization

Some high-SNR excerpts from the dataset are shown below in time and frequency.  Each dataset consists of a number of complex 128 sample labeled time-domain radio signal examples.

dataset_psd dataset_time