Published October 31, 2026 | Version 1.0

Results: Choosing a generator for streaming Gaussian noise in gravitational-wave mock data

Authors/Creators

  • 1. KU Leuven, Department of Electrical Engineering (ESAT), STADIUS Center for Dynamical Systems, Signal Processing and Data Analytics; Leuven Gravity Institute, KU Leuven

Description

Measured results behind the tables, the data figures and the numbers quoted from them in the paper Choosing a generator for streaming Gaussian noise in gravitational-wave mock data: accuracy, cost and reproducibility, together with the noise spectra the results were computed for. The paper compares generators of long, streaming Gaussian noise records with a target power spectral density: the crossfade of independent frequency-domain realizations used by gwmock-noise and LALSimulation, PyCBC's reproducible filtered noise, overlap-save FIR filtering, and autoregressive and ARMA models. It reports their exact covariance error, throughput, retained state and the cost of regenerating a segment bit for bit, across Advanced LIGO, Advanced Virgo, KAGRA, Cosmic Explorer and Einstein Telescope sensitivity curves.

Files. results.tar.gz unpacks into results/ (191 JSON files) and psd/ (16 spectrum tables). README.md documents the layout, the fields common to every file and the measured quantities of each kind of result.

  • results/exact/: the covariance error of every generator configuration, per spectrum, band-edge treatment and sampling rate.
  • results/costs/: throughput, memory and state of every generator.
  • results/implementation/: generated noise compared with its exact spectrum.
  • results/reproducibility.json: bit-exact regeneration of a segment, its stored state and rebuild cost.
  • results/verification.json: the measured crossfade covariance against its closed form.
  • results/edge-leakage/: predicted and measured coherence near the band edges.
  • psd/: the one-sided power spectral density of each spectrum.

Provenance. Each result file records the run configuration, the random seeds of a sampled run or the rule that derives them (the exact errors are computed, not sampled, and have none), the gwmock-noise commit that generated the noise, the revision of the analysis code (with no uncommitted changes in any file), library versions, the hardware, and the run's wall time, CPU time and peak memory. Run labels were replaced by descriptive ones for this release, and non-finite numbers are written as null so that every file is strict JSON; no other measured value, configuration, seed or commit differs from the run's output.

Reproducing the paper. The scripts that fetch this record and reproduce the paper's tables and figures, including its two analytic schematic figures, which need no data, are at github.com/isaac-cf-wong/noise-stitching-statistics-data.

Files

README.md

Files (45.0 MB)

Name Size
md5:77692378a6f5a8fa6014cedf1c5e6127
16.7 kB Preview Download
md5:d24fab45af50addeb46ea8c03cb5a1ba
44.9 MB Download

Additional details