Adaptive Denoising of Acoustic Noise Injections Performed at the Virgo Interferometer

Abstract

A methodology using adaptive time series analysis is tested on data from a seismometer monitoring the north end building (NEB) of the Virgo interferometer during four acoustic noise injections. Empirical mode decomposition (EMD) is used for adaptive detrending, while the recently developed time-varying filter EMD algorithm is used for narrowband mode extraction. Mode persistency is evaluated with detrended fluctuation analysis, and denoising is achieved by setting a threshold $H_{\text {thr}}$ on the Hurst exponent of the obtained modes. The adopted methodology is proven useful in adaptively separating the seismic noise induced by the acoustic noise injections from the underlying nonlinear non-stationary recordings of the seismometer monitoring NEB. The Hilbert–Huang transform provides a high-resolution time–frequency representation of the data. Furthermore, the local Hurst exponent exhibits a drop due to the injections that is of the same order of $H_{\text {thr}}$. This suggests that the local Hurst exponent could be calculated as an initial step in order to select the threshold $H_{\text {thr}}$. The algorithms could be used for detector characterisation purposes such as the investigation of non-Gaussian noise.