Adaptive cancellation of mains power interference in continuous gravitational wave searches with a hidden Markov model
T. Kimpson, S. Suvorova, H.Middleton, C.Liu, A. Melatos, R. Evans, W. Moran · 2024
Read Adaptive cancellation of mains power interference in continuous gravitational wave searches with a hidden Markov model by T. Kimpson, S. Suvorova, H.Middleton, C.Liu, A. Melatos, R. Evans, W. Moran (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-01058.txt.
Introduction
Tom Kimpson and co-authors show, in a 2024 paper, how to strip a mains power hum from data in the hunt for steady gravitational waves. Detectors that listen for these waves, such as LIGO, show long, thin noise lines in their data. A true signal that lands on a line is often thrown out, lost to the search. The mains line sits at 60 hertz in the United States. Their fix borrows a noise-cancel trick from electrical work, built on a fast least-squares update.
They pair the fix with a hidden Markov model, a standard search tool that tracks a signal whose tone can drift. A fake signal, set to cross the mains line, was missed before the fix in both fake and real LIGO noise. After the fix, the line fell by 20 to 40 decibels and the signal stood out. The pair can save signals once lost.
Catalog record
| CREATOR | T. Kimpson, S. Suvorova, H.Middleton, C.Liu, A. Melatos, R. Evans, W. Moran |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-01058
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Adaptive cancellation of mains power interference in continuous gravitational wave searches with a hidden Markov model T. Kimpson, S. Suvorova, H.Middleton, C.Liu, A. Melatos, R. Evans, W. Moran Abstract Continuous gravitational wave searches with terrestrial, long-baseline interferometers are hampered by long-lived, narrowband features in the power spectral density of the detector noise, known as lines. Candidate GW signals which overlap spectrally with known lines are typically vetoed. Here we demonstrate a line subtraction method based on adaptive noise cancellation, using a recursive least squares algorithm, a common approach in electrical engineering applications such as audio and biomedical signal processing. We validate the line subtraction method by combining it with a hidden Markov model (HMM), a standard continuous wave search tool, to detect an injected continuous wave signal with an unknown and randomly wandering frequency, which overlaps with the mains power line at 60 Hz 60\,{\rm Hz} in the Laser Interferometer Gravitational Wave Observatory (LIGO). The performance of the line subtraction method is tested on a injected continuous wave signal obscured by (a) synthetic noise data with both Gaussian and non-Gaussian components, and (b) real noise data obtained from the LIGO Livingston detector. In both cases, before applying the line subtraction method the HMM does not detect the injected continuous wave signal. After applying the line subtraction method the mains power line is suppressed by 20–40 dB, and the HMM detects the underlying signal, with a time-averaged root-mean-square error in the frequency estimate of ∼ 0.05 \sim 0.05 Hz. The performance of the line subtraction method with respect to the characteristics of the 60 Hz line and the control parameters of the recursive least squares algorithm is quantified in terms of receiver operating characteristic curves. DOI I Introduction Instrumental noise artifacts in gravitational wave (GW) searches with terrestrial, long-baseline interferometers [ 1 , 2 , 3 ] are classified according to their duration and spectral properties. Short-lived, non-stationary, recurrent noise events such as optomechanical glitches typically last for seconds and exhibit distinctive spectral signatures, e.g they can be chirp-like [ 4 , 5 , 6 , 7 ] . Long-lived, quasi-stationary, broadband noise sources include seismic disturbances at low frequencies, test mass thermal fluctuations at intermediate frequencies, and photon shot noise at high frequencies [ 8 , 9 , 10 , 11 ] . Long-lived narrowband spectral artifacts — termed instrumental lines — are caused by electrical subsystems (e.g. mains power, clocks, oscillators), mechanical subsystems (e.g. test mass and beam-splitter violin modes) and calibration processes [ 12 ] , although sometimes the origin of a specific feature is unknown. Instrumental lines are disruptive, especially for continuous wave (CW) searches where the target astrophysical signal is quasi-monochromatic and resembles the noise artifact spectrally. Many above-threshold candidates discovered in published CW searches are vetoed because they coincide with known instrumental lines [ 13 , 14 , 15 ] , for instance, in data from Observing Run 3 (O3) with the Laser Interferometer Gravitational Wave Observatory (LIGO), Virgo and the Kamioka Gravitational Wave Detector (KAGRA) [ 16 , 17 , 18 , e.g.] Several techniques have been implemented by the LIGO-Virgo-KAGRA (LVK) Collaboration to identify, characterize and suppress instrumental noise artifacts [ 5 , 6 ] . Some techniques identify and veto an artifact (or gate data segments) based on its time-frequency signature [ 19 , 20 ] . Other techniques perform offline noise subtraction with reference to auxiliary data from physical environmental monitors (PEMs) [ 21 , 22 , 23 , 24 ] . PEMs can be used to witness correlated noise and generate a reference signal directly or elucidate and quantify multichannel couplings [ 25 , 26 ] . Additionally some techniques are based on machine learning [ 27 , 28 , 29 , 30 ] or multi-detector consistency [ 31 , 32 ] . In CW searches specifically, the distinctive amplitude and frequency (Doppler) modulations associated with the Earth’s rotation and revolution respectively can be exploited to discriminate between terrestrial noise artifacts and astrophysical signals [ 33 , 34 ] . In most of the situations above, the practical effect of an instrumental line is to excise the relevant part of the observing band from a CW search. That is, if an above-threshold CW search candidate coincides with a known instrumental line, the candidate is often vetoed under current practice without further analysis, e.g. without comparing the expected strength of the noise line with the measured strength of the candidate 1 1 1 A regularly updated log of narrowband instrumental lines in the LVK detector is maintained at dcc.ligo.org/LIGO-T2100200/public for public reference. , although some line robust algorithms exist which combine statistics from different detectors to suppress single detector line artifacts [ 31 , 35 , 36 , 37 ] ; we refer the reader to [ 38 ] for an overview of different continuous wave search pipelines and the associated vetoing procedures. In this paper we take a step towards lifting the above limitation. We introduce an adaptive noise cancellation (ANC) scheme based on an adaptive recursive least squares (ARLS) algorithm which suppresses narrowband noise proportional to a known PEM reference signal. We then apply the ANC scheme to a CW search algorithm based on a hidden Markov model (HMM), which detects and tracks quasi-monochromatic GW signals with wandering frequency and has been tested and validated thoroughly in multiple LVK searches [ 39 , 13 , 14 , 15 ] . We demonstrate with synthetic data that the ANC scheme and HMM algorithm together can successfully detect a GW signal overlapping the 60 Hz mains power line, if the signal exceeds a well-defined minimum amplitude. The approach extends naturally to other instrumental lines, provided they have an associated witness PEM channel, a topic for future work. The paper is organized as follows. In Section II we outline a mathematical model for the 60 Hz mains power line and the PEM reference signal, and justify the assumptions of the model by reference to data from the LIGO Livingston interferometer. In Section III we introduce ANC formulated as an ARLS method. In Section IV we validate the ANC scheme on synthetic GW strain data, in conjunction with a pre-existing HMM search pipeline, and demonstrate the successful recovery of a frequency-wandering CW signal. In Section V we quantify the performance of the HMM pipeline and ANC scheme as a function of the key parameters of the mains power interference and ANC filter. The validation tests are extended to real LIGO noise in Section VI . Concluding remarks are made in Section VII . II Mains power interference The goal of this paper is to detect a quasi-monochromatic GW signal in a data stream contaminated by two kinds of noise: additive Gaussian noise which is fundamental and irreducible, and additive non-Gaussian interference from a long-lived narrow spectral feature which can be filtered out in principle given an accurately measured reference signal. For this work we consider the spectral line at 60 Hz that results from the mains electricity in North America as the additive non-Gaussian interference, as it is a long-standing impediment to many CW searches with the LIGO interferometers. Similar lines due to mains electricity arise at 50 Hz for Virgo [