Characterization of gravitational-wave detector noise with fractals

Abstract

We present a new method, based on fractal analysis, to characterize the output of a physical detector that is in the form of a set of real-valued, discrete physical measurements. We apply the method to gravitational-wave data from the latest observing run of the Laser Interferometer Gravitational-wave Observatory. We show that a measure of the fractal dimension of the main detector output (strain channel) can be used to determine the instrument status, test data stationarity, and identify non-astrophysical excess noise in low latency. When applied to instrument control and environmental data (auxiliary channels) the fractal dimension can be used to identify the origins of noise transients, non-linear couplings in the various detector subsystems, and provide a means to flag stretches of low-quality data.

UniMAP: model-free detection of unclassified noise transients in LIGO-Virgo data using the temporal outlier factor

“…Finally, more radical pre-processing schemes for the detector data may make lower-SNR signals more visible to the TOF algorithm. Candidates for such schemes include removal of stationary noise background via existing methods, and conversion of raw strain data into a measure of fractal dimension [30] before analysis by UniMAP.…”

Detection of anomalies amongst LIGO’s glitch populations with autoencoders

“…The first step towards characterizing glitches through safe auxiliary channels requires identifying anomalous data stretches within them [43,45,46]. In [47], the author proposes the measurement of FD as an additional effective tool for characterizing the instrument output in low latency. FD is an index that characterizes the self-similarity of a set and provides a measure of the complexity of the signal in the context of signal processing [48].…”

“…Following [47] we numerically estimate the measured FD with the variation (VAR) method (see [47] for details). For a discretely-sampled set of data with N measurements C ∈ R N , we can define a sliding window to compute the variation of the data with centre l and scale k,…”

“…As we can see in algorithm 1, the implementation in [47] computes the maximum and minimum over a range of values at each iteration k, l (line 5 and 6). The runtime of this implementation is O(N 3 ).…”

“…In this work we focus on LIGO Livingston, as the author in [47], but this investigation could be extended to LIGO Hanford and Virgo. The details on how this data was pre-processed for its posterior usage in our model, can be found in the next section.…”