GlitchGAN Documentation¶
Conditional generative model for synthesising realistic LIGO gravitational-wave detector glitches in the time domain.
GlitchGAN uses a class-conditional Dual-discriminator Variational GAN (cDVGAN) architecture — a Wasserstein GAN with gradient penalty augmented by a first-derivative discriminator that enforces realistic temporal structure. It supports seven O3 glitch classes and generates 2-second waveforms at 4096 Hz.
Note
GlitchGAN is described in Dooney et al., Phys. Rev. D 114, 082005 (2026). If you use it in your work, please cite the paper — see Citation for the BibTeX entry.
User Guide
API Reference