Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DVGAN: Stabilize Wasserstein GAN training for time-domain Gravitational Wave physics

Tom Dooney, Stefano Bromuri, Lyana Curier

TL;DR

The paper addresses the challenge of generating high-fidelity, time-domain gravitational-wave signals for data augmentation and glitch modeling. It introduces DVGAN, a three-player Wasserstein GAN with a derivative-based auxiliary discriminator that regularizes learning on 1D signals, yielding smoother and more realistic samples than a vanilla WGAN. Through ablation studies and experiments on Gaussian pulses, ringdown, BBH waveforms, and real LIGO glitches, DVGAN demonstrates improved stability, fidelity, and distribution coverage, with practical implications for online GW reconstruction and detector noise characterization. The approach offers a pathway to robust synthetic data generation for next-generation GW detectors and complex signal environments.

Abstract

Simulating time-domain observations of gravitational wave (GW) detector environments will allow for a better understanding of GW sources, augment datasets for GW signal detection and help in characterizing the noise of the detectors, leading to better physics. This paper presents a novel approach to simulating fixed-length time-domain signals using a three-player Wasserstein Generative Adversarial Network (WGAN), called DVGAN, that includes an auxiliary discriminator that discriminates on the derivatives of input signals. An ablation study is used to compare the effects of including adversarial feedback from an auxiliary derivative discriminator with a vanilla two-player WGAN. We show that discriminating on derivatives can stabilize the learning of GAN components on 1D continuous signals during their training phase. This results in smoother generated signals that are less distinguishable from real samples and better capture the distributions of the training data. DVGAN is also used to simulate real transient noise events captured in the advanced LIGO GW detector.

DVGAN: Stabilize Wasserstein GAN training for time-domain Gravitational Wave physics

TL;DR

The paper addresses the challenge of generating high-fidelity, time-domain gravitational-wave signals for data augmentation and glitch modeling. It introduces DVGAN, a three-player Wasserstein GAN with a derivative-based auxiliary discriminator that regularizes learning on 1D signals, yielding smoother and more realistic samples than a vanilla WGAN. Through ablation studies and experiments on Gaussian pulses, ringdown, BBH waveforms, and real LIGO glitches, DVGAN demonstrates improved stability, fidelity, and distribution coverage, with practical implications for online GW reconstruction and detector noise characterization. The approach offers a pathway to robust synthetic data generation for next-generation GW detectors and complex signal environments.

Abstract

Simulating time-domain observations of gravitational wave (GW) detector environments will allow for a better understanding of GW sources, augment datasets for GW signal detection and help in characterizing the noise of the detectors, leading to better physics. This paper presents a novel approach to simulating fixed-length time-domain signals using a three-player Wasserstein Generative Adversarial Network (WGAN), called DVGAN, that includes an auxiliary discriminator that discriminates on the derivatives of input signals. An ablation study is used to compare the effects of including adversarial feedback from an auxiliary derivative discriminator with a vanilla two-player WGAN. We show that discriminating on derivatives can stabilize the learning of GAN components on 1D continuous signals during their training phase. This results in smoother generated signals that are less distinguishable from real samples and better capture the distributions of the training data. DVGAN is also used to simulate real transient noise events captured in the advanced LIGO GW detector.
Paper Structure (18 sections, 10 equations, 7 figures, 3 tables)

This paper contains 18 sections, 10 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Generative Adversarial Networks
  4. Methods
  5. Dual-Discriminator Derivative GAN (DVGAN)
  6. Training Data and Procedures
  7. Dataset Construction and Preprocessing
  8. Results
  9. Data Generation and Experiments
  10. Experimental Analysis
  11. Visualization - Diversity
  12. Discriminative Score - Fidelity
  13. Match Analysis for BBH Dataset
  14. Conclusion
  15. Conclusions
  16. ...and 3 more sections

Figures (7)

  • Figure 1: Time series (top) and corresponding Q-transform (bottom) representations of blip glitches from LIGO's Hanford (H1) detectorGENGLI.
  • Figure 2: Diagrams of a typical GAN architecture (left) and DVGAN (right).
  • Figure 3: Examples of an unstable learning process from a vanilla WGAN (left) and a stable learning process from a DVGAN (right) on the Binary Black Hole (BBH) dataset.
  • Figure 4: Example waveforms (top) for all datasets (pulse, ringdown, BBH, blip) with example generations from WGAN (middle) and DVGAN (bottom). Noise artifacts are more significant in the WGAN generations compared with DVGAN, in particular for the pulse and blip examples.
  • Figure 5: T-SNE plots comparing of real and synthetic signals for a vanilla WGAN (top) and DVGAN (bottom)
  • ...and 2 more figures