tomdooney/deepextractor-glitch-reconstructions

DeepExtractor Glitch Reconstructions

Time-domain reconstructions of seven gravitational-wave detector glitch classes from LIGO's third observing run (O3), produced using DeepExtractor. This dataset was used to train GlitchGAN, a class-conditional generative model for realistic glitch synthesis described in:

T. Dooney et al., Realistic Time-Domain Synthesis of Gravitational-Wave Detector Glitches using Class-Conditional Derivative Generative Adversarial Networks, 2026.

Dataset Description

Each sample is a 2-second whitened time-series centred on a glitch trigger time, reconstructed at 4096 Hz (8192 samples). Glitches were selected from LIGO Hanford (H1) and Livingston (L1) during O3a and O3b under the following criteria:

• Gravity Spy classification confidence ≥ 0.9
• Signal-to-noise ratio (SNR) ≥ 15

All waveforms are normalised to the range [−1, 1] and mean-subtracted. The dataset is class-balanced via bootstrap resampling to 5,000 samples per class (35,000 total).

Files

File	Shape	dtype	Description
glitch_GAN_samples_scaled_balanced.npy	(35000, 8192)	float64	Whitened, normalised glitch waveforms
glitch_GAN_deriv_samples_balanced.npy	(35000, 8191)	float64	First-order time derivatives of the waveforms (for cDVGAN training)
glitch_GAN_labels_balanced.npy	(35000, 7)	float64	One-hot encoded class labels
glitch_GAN_label_order.npy	(7,)	str	Class names corresponding to each label column

Glitch Classes

The seven classes, in label-column order:

Index	Class
0	Blip
1	Fast Scattering
2	Koi Fish
3	Low Frequency Burst
4	Scattered Light
5	Tomte
6	Whistle

Usage

With glitchgan

from glitchgan import download_data

download_data(data_dir="data/")

With deepextractor

from deepextractor.data import download_glitch_data

download_glitch_data(data_dir="data/")

Direct download

from huggingface_hub import hf_hub_download

samples = hf_hub_download(
    repo_id="tomdooney/deepextractor-glitch-reconstructions",
    filename="glitch_GAN_samples_scaled_balanced.npy",
    repo_type="dataset",
)

Citation

If you use this dataset, please cite:

@article{dooney2026glitchgan,
  title   = {Realistic Time-Domain Synthesis of Gravitational-Wave Detector Glitches
             using Class-Conditional Derivative Generative Adversarial Networks},
  author  = {Dooney, Tom and others},
  journal = {Physical Review D},
  year    = {2026},
}

License

Creative Commons Attribution 4.0 International (CC BY 4.0)