Application of Non-Linear Noise Regression in the Virgo Detector (2410.06220v3)

Abstract: Since the first detection of gravitational waves (GWs) in 2015, the International Gravitational-wave Network has made substantial strides in improving the sensitivity of ground-based detectors. Despite these advancements, many GW signals remain below the detection threshold due to environmental noise that limits sensitivity. In recent years, algorithms such as DeepClean have been developed to estimate and remove contamination from various noise sources, addressing linear, non-linear, and non-stationary coupling mechanisms. In this paper, we present noise reduction in the Virgo detector using DeepClean, serving as a preliminary step toward integrating Virgo into the online noise reduction pipeline for the O5 observing run. Our results demonstrate the applicability of DeepClean in Virgo O3b data, where noise was reconstructed from a total of 225 auxiliary witness channels. These channels were divided into 13 subsets, each corresponding to a specific frequency band, with training and subtraction performed layer-wise in a sequential manner. We observe that the subtraction improves the binary neutron star inspiral range by up to 1.3 Mpc, representing an approximately 2.5% increase. To ensure robust validation, we conduct an injection study with binary black hole waveforms. Matched-filter analyses of the injections showed an average improvement of 1.7% in the recovered signal-to-noise ratio, while parameter estimation confirmed that DeepClean introduces no bias in the recovered parameters. The successful demonstration provides a pathway for online non-linear noise subtraction in Virgo in the future observing runs.