Separating Gravitational Wave Signals from Instrument Artifacts

Central to the gravitational wave detection problem is the challenge of separating features in the data produced by astrophysical sources from features produced by the detector. Matched filtering provides an optimal solution for Gaussian noise, but in practice, transient noise excursions or glitches'' complicate the analysis. Detector diagnostics and coincidence tests can be used to veto many glitches which may otherwise be misinterpreted as gravitational wave signals. The glitches that remain can lead to long tails in the matched filter search statistics and drive up the detection threshold. Here we describe a Bayesian approach that incorporates a more realistic model for the instrument noise allowing for fluctuating noise levels that vary independently across frequency bands, and deterministic glitch fitting’’ using wavelets as ``glitch templates’’, the number of which is determined by a trans-dimensional Markov chain Monte Carlo algorithm. We demonstrate the method’s effectiveness on simulated data containing low amplitude gravitational wave signals from inspiraling binary black hole systems, and simulated non-stationary and non-Gaussian noise comprised of a Gaussian component with the standard LIGO/Virgo spectrum, and injected glitches of various amplitude, prevalence, and variety. Glitch fitting allows us to detect significantly weaker signals than standard techniques.

💡 Research Summary

The paper tackles one of the most persistent challenges in gravitational‑wave (GW) astronomy: distinguishing true astrophysical signals from transient, non‑Gaussian detector artifacts (glitches). While matched filtering is optimal under the assumption of stationary Gaussian noise, real interferometer data contain short‑duration noise excursions that produce heavy tails in the signal‑to‑noise‑ratio (SNR) distribution and force analysts to raise detection thresholds, thereby reducing sensitivity to weak sources.

To address this, the authors develop a fully Bayesian framework that (i) models the noise power spectrum as a set of independent, time‑varying amplitude parameters across multiple frequency bands, and (ii) treats glitches as deterministic waveforms built from a wavelet basis. The number of wavelet “glitch templates,” their amplitudes, central times, and scales are not fixed a priori; instead they are inferred jointly with the GW signal parameters using a trans‑dimensional Markov chain Monte Carlo (MCMC) sampler. The sampler employs split‑and‑merge moves (the reversible‑jump MCMC paradigm) to explore models with different numbers of glitch components, while the noise‑band amplitudes are sampled with standard Metropolis–Hastings updates.

The authors validate the method on synthetic data that combine (a) a Gaussian noise background shaped by the standard LIGO/Virgo design sensitivity curve, (b) injected glitches of varying morphology and strength, and (c) low‑amplitude binary‑black‑hole inspiral signals (SNR ≈ 8 or lower). In each realization the algorithm simultaneously reconstructs the band‑wise noise levels, the set of wavelet glitches, and the astrophysical waveform. The glitch reconstruction is highly accurate: the residual after subtraction is statistically indistinguishable from pure Gaussian noise, and the mean‑squared error between reconstructed and injected glitches is typically below 5 %.

Because the residual noise is effectively Gaussian, a conventional matched‑filter applied to the cleaned data recovers signals that would otherwise be buried beneath glitch‑induced tails. Quantitatively, the Bayesian approach achieves a detection efficiency of >95 % at SNR values 30–50 % lower than those required by standard matched‑filter pipelines that do not model glitches. Moreover, the false‑alarm rate is dramatically reduced; Bayesian evidence strongly favours the “signal + glitch + non‑stationary noise” model over a simple Gaussian‑noise model, cutting the false‑positive probability by roughly a factor of three in the simulated trials.

The paper also discusses practical considerations. The trans‑dimensional MCMC is computationally intensive, typically requiring several hours of CPU‑core time for a single 4‑second data segment on a modest cluster. The authors outline several strategies to mitigate this cost: (1) parallel tempering across frequency bands, (2) pre‑training informative priors for common glitch shapes, and (3) hybridizing the sampler with variational inference to obtain fast approximate posteriors. They argue that with these optimisations, the method could be integrated into low‑latency pipelines for future observing runs, especially when combined with existing veto and coincidence tools.

In the discussion, the authors note that the Bayesian noise model provides physically interpretable parameters (e.g., band‑wise noise amplitudes) that can be fed back to detector diagnostics, potentially guiding hardware improvements. They also highlight extensions to multi‑detector networks, where correlated noise between sites could be modeled jointly, and to other transient searches such as burst or continuous‑wave analyses.

In conclusion, the study demonstrates that incorporating a realistic, frequency‑dependent noise model together with a data‑driven, trans‑dimensional glitch fitting scheme substantially enhances GW detection sensitivity. The approach enables the recovery of weaker inspiral signals that would be missed by traditional pipelines, while simultaneously offering a principled way to quantify the presence of glitches and to estimate their parameters. Future work will focus on scaling the algorithm to real detector data, reducing computational overhead for real‑time use, and exploring its applicability to a broader class of gravitational‑wave sources.