Source-linked AI summary
Improving the Sensitivity of Advanced LIGO Using Noise Subtraction
D. Davis, T. J. Massinger, A. P. Lundgren, J. C. Driggers, A. L. Urban, L. K. Nuttall
TL;DR
Advanced LIGO noise subtraction had previously been applied around selected events rather than across the full observing run. This paper develops a parallel approach for subtracting linearly coupled noise and applies it to O2 data. It reports a 30% increase in volume-time over O2 and identifies loud noise artifacts as a constraint on that gain.
Problem
Earlier event-focused subtraction used 4096 seconds around identified signals, while O2-wide noise subtraction required a method suited to long, nonstationary data.
Source: §1 · p. 2, §2.2 · p. 4Method
The pipeline estimates witness-sensor transfer functions, jointly handles correlated witnesses, and processes independent noise sources and data sections in parallel.
Source: §1 · p. 2, §2.1 · p. 3Results
30% increase in V-T was measured over O2, averaged across all mass bins.
Source: §5.3 · p. 12Takeaways & Limitations
The reported 30% increase in aLIGO sensitivity to compact binary coalescences expands the spacetime volume searched for gravitational waves.
Source: §6 · p. 13Loud artifacts unrelated to the targeted noise sources gain SNR against the reduced noise floor, limiting the V-T increase.
Source: §5.3 · p. 12
Abstract
from arXiv · showhide
This paper presents an adaptable, parallelizable method for subtracting linearly coupled noise from Advanced LIGO data. We explain the features developed to ensure that the process is robust enough to handle the variability present in Advanced LIGO data. In this work, we target subtraction of noise due to beam jitter, detector calibration lines, and mains power lines. We demonstrate noise subtraction over the entirety of the second observing run, resulting in increases in sensitivity comparable to those reported in previous targeted efforts. Over the course of the second observing run, we see a 30% increase in Advanced LIGO sensitivity to gravitational waves from a broad range of compact binary systems. We expect the use of this method to result in a higher rate of detected gravitational-wave signals in Advanced LIGO data.
1. Introduction
Earlier noise-subtraction efforts targeted short periods around known events, while this work develops a computationally efficient, parallel Python pipeline intended for the full O2 data set. It targets beam jitter, calibration lines, and mains power lines and reports resulting search-sensitivity improvements.
1. Introduction
- O2 produced detections from three binary black hole systems and the first binary neutron star system.The run lasted from November 30, 2016 to August 26, 2017.Source: §1 · p. 2
- Earlier noise subtraction used a MATLAB algorithm on 4096 seconds around identified events, not the entirety of O2.That targeted analysis supported parameter estimation for previously published gravitational-wave signals.Source: §1 · p. 2
- The Python pipeline targets subtraction of beam jitter, calibration lines, and mains power lines across the O2 data set.Its design prioritizes parallel processing and computational efficiency for the scale of the data.Source: §1 · p. 2
2. Subtraction Pipeline Overview
The pipeline estimates linearly coupled noise from witness sensors, subtracts projected contributions from strain, and adapts its estimates over time. Independent noise sources and detector-data sections can be processed in parallel.
2.1. Measurement of Transfer Functions
- The subtraction models noise in strain as a witness-sensor output convolved with an unknown transfer function, then filters and subtracts that contribution.Transfer functions are calculated in the frequency domain using data sampled over time and frequency-band inner products.Source: §2.1 · pp. 2–3
- A 2.5 × 10^-9 fractional threshold zeros transfer-function bands below the threshold to reduce spurious correlations from uncorrelated noise.The threshold was sufficient for the noise sources studied and can be tuned for other use cases.Source: §2.1 · p. 3
- Correlated witness sensors require solving jointly for transfer functions whose noise contributions are independent, preventing oversubtraction when sensors witness the same source.The method also permits additional sensors witnessing different features of a source to be included without risking oversubtraction.Source: §2.1 · p. 3
2.2. Calculation of Coupled Noise
- Because Advanced LIGO data are nonstationary over hours, overlapping finite windows and Hann-weighted projected noise sections accommodate changing transfer functions.The method removes high-amplitude non-Gaussian artifacts before estimating transfer functions; Figure 2 illustrates the time-window tiling.Source: §2.2 · p. 4, Figure 2 · p. 5
- When transfer functions are constant, the windowed method is equivalent to applying one transfer function throughout; independent projected noise series can then be subtracted separately.After all targeted contributions are subtracted, the data are called “cleaned”.Source: §2.2 · p. 4
2.3. Workflow Implementation
- Parallel transfer-function calculations for independent noise sources and detector-operation sections enable high-throughput processing, limited by available computing nodes.With 14 witness sensors, available resources processed about 65 gigabytes—two weeks from one detector—in a few hours.Source: §2.3 · p. 5
3. Noise sources
During O2, the targeted noise sources included broadband beam-jitter contributions and narrow line artifacts from mains power and calibration signals. The sources had different detector and sensor coverage, and jitter subtraction was bounded by witness-sensor sampling.
3. Noise sources
- O2 instrumental noise fell into two targeted classes: broadband beam-jitter contributions and narrow lines from power mains and calibration signals.Both classes had also been identified and subtracted in analyses of previously published events.Source: §3 · p. 5
3.1. Jitter Noise
- Hanford jitter arose amid the O2 PSL high-powered oscillator configuration and a test-mass point absorber that may have strengthened beam-fluctuation coupling.Mitigation required replacing the HPO stage and test mass, work that could not occur until after the observing run.Source: §3.1 · p. 6
- Beam jitter was present at Hanford throughout O2, with stronger coupling late in the run and broadband contributions as well as coupling at optic-mount resonances from 100 to 700 Hz.Seven beam-angle and beam-size readouts were used as witnesses; Livingston had no noticeable jitter coupling and no jitter subtraction was performed there.Source: §3.1 · p. 6
- Jitter subtraction at Hanford was limited to 1024 Hz by the 2048 Hz witness-sensor sampling rate; broadband coupling above that frequency was not addressed.The paper attributes possible higher-frequency broadband noise to beam-angle coupling but does not subtract it.Source: §3.1 · p. 6
3.2. Line Artifacts
- Power-mains noise produced lines at 60 Hz and harmonics at both sites, measured using three witness sensors for the mains' three-phase voltage.Calibration lines were also targeted, using recorded digital excitation signals or a photodetector monitoring photon-calibrator beam power, depending on how they were applied.Source: §3.2 · pp. 6–7
4. Diagnostics
The diagnostics tested whether noise subtraction preserves astrophysical signals, handles artificial noise, maintains calibration, and behaves safely when witness sensors contain transient or spurious correlations. The tests found preserved or improved signal recovery, small spectral and calibration changes, and specific transient-related oversubtraction risks.
4.1. Sensor Safety
- All witness sensors used for subtraction were found incapable of witnessing and subtracting away gravitational-wave signals.Safety was assessed by injecting sine-Gaussian waveforms and checking whether they coupled significantly into witness sensors.Source: §4.1 · p. 7
4.2. Recovery of Simulated Compact Binary Coalescence Signals
- All simulated signals had coincident ranking statistics consistent with or better than before subtraction, and previously unrecovered signals and hardware injections were recovered.The simulations covered five days of data from both detectors.Source: §4.2 · p. 7
4.3. Simulated Noise Tests
- For artificial correlated noise, the algorithm reconstructed the injected transfer function and the resulting amplitude spectral density matched the original within ±3% at all frequencies.Source: §4.3 · pp. 7–8
- For uncorrelated noise absent from the strain data, the algorithm subtracted only a minimal amount and the resulting amplitude spectral density matched the original within ±2% at all frequencies.Source: §4.3 · p. 8
4.4. Effect on Calibration
- Across 6.65 days, calibration-line amplitude ratios and phase offsets remained consistent with 1 and 0°; the authors concluded subtraction did not significantly impact overall strain-data calibration.At 331 Hz, where beam-jitter subtraction removes non-negligible power, the amplitude ratio was consistent with 1 within ±0.15% and phase offset with 0° within ±0.087°.Source: §4.4 · pp. 8–9
4.5. Impact of Nonstationary Data
- Transient witness-sensor noise can cause oversubtraction when uncorrelated with strain, while spurious correlations over multi-hour averages may affect less than 0.1 Hz of bandwidth.The process was halted for 3 seconds around photon-calibrator transients; narrowband spurious correlations can be removed from long-timescale analyses with notch filters.Source: §4.5 · p. 9
5. Results
Noise subtraction processed the O2 data set with only a small loss of strain data, removing narrowband lines at both detectors and broadband beam-jitter noise at Hanford. Across O2, the PyCBC search’s volume-time sensitivity increased by 30%, although unchanged instrumental artifacts limited the gain.
5.1. The O2 Data Set
- The pipeline processed the nine-month O2 data set, discarded 0.05% of strain data during subtraction, and produced 118 days of coincident cleaned data.Periods unfit for astrophysical analysis were not processed; additional losses came from filter-corrupted periods and unsuitable witness sensors.Source: §5.1 · pp. 9–10
5.2. Effects on the Noise Curve
- Subtraction removed calibration and power-mains lines at both detectors and broadband beam-jitter noise at Hanford, between 80 and 1000 Hz.Broadband subtraction was limited to 13–1024 Hz; Livingston had no suitable broadband witness sensor.Source: §5.2 · pp. 10–11, Figure 4 · p. 10
- Hanford’s inspiral range rose by about 20% on average over O2 and peaked at about 50% near the run’s end; Livingston’s change was negligible.The range measures average distance for observing a 1.4–1.4 M⊙ binary neutron star at SNR 8; the decrease after week 32 was attributed to an earthquake.Source: §5.2 · p. 11, Figure 5 · p. 11
5.3. Effect on Astrophysical Analyses
- PyCBC volume-time sensitivity was compared before and after subtraction using identical injection sets, chirp-mass bins, and IFAR thresholds of 100 and 1000 years.Volume-time accounts for search sensitivity across the detector network and non-Gaussian artifacts, which inspiral range alone does not fully capture.Source: §5.3 · pp. 11–12
- 30% increase in PyCBC volume-time sensitivity was measured across O2, with the largest gains for chirp masses from 1.74 to 8.07 M⊙.The reported gains were averaged across mass bins; this mass range had not previously yielded detected signals and includes the observed NS-BH mass gap.Source: §5.3 · p. 12, Figure 6 · p. 12
- Instrumental artifacts unrelated to the targeted noise sources retained their amplitude but increased in SNR against the lower noise floor, limiting the volume-time gain.Loud noise artifacts are a primary limitation on an astrophysical search’s ability to recover signals.Source: §5.3 · p. 12
6. Future prospects
Noise subtraction increased compact-binary search sensitivity and is presented as useful for additional noise sources and gravitational-wave searches. The authors identify future candidates and note expected hardware changes that could reduce beam-jitter noise.
6. Future prospects
- Noise from feedback loops sensing and controlling cavity length and alignment is a potential future subtraction target, with contributions reported at lower frequencies.Source: §6 · p. 13
- Before O3, replacing the Hanford test mass with a point absorber and updating the PSL configuration are expected to reduce beam-jitter noise contributions.Source: §6 · p. 13
- 30% increase in aLIGO sensitivity to compact binary coalescences after noise subtraction will allow a greater volume of spacetime to be searched for gravitational waves.Source: §6 · p. 13
- Noise from continuous-wave, stochastic, and unmodeled-burst searches is also expected to benefit, although those sensitivity gains were not quantified.Source: §6 · p. 13