Microseismic Noise Mitigation with Machine Learning for Advanced LIGO

TL;DR

The paper addresses the challenge of residual microseismic noise in LIGO's active seismic isolation, particularly in the $0.1-0.3$ Hz band, which degrades lock stability and duty cycle. It compares traditional linear filtering with a data-driven, causal ML approach and demonstrates that an LSTM model can predict residual platform motion from SEI sensor data, achieving up to two orders of magnitude reduction in the target band. The results indicate significant nonlinear cross-couplings behind residual motion and show promising generalization to data beyond the training period, suggesting practical viability for real-time control. The study proposes a path toward real-time deployment and adaptive retraining, potentially extending the detectors’ low-frequency sensitivity and astrophysical reach through improved seismic isolation.

Abstract

The unprecedented sensitivity of the Laser Interferometer Gravitational-Wave Observatory, which enables the detection of distant astrophysical sources, also renders the detectors highly susceptible to low-frequency ground motion. Persistent microseisms in the 0.1-0.3 Hz band couple into the instruments, degrade lock stability, and contribute substantially to detector downtime during observing runs. The multi-stage seismic isolation system has achieved remarkable success in mitigating such disturbances through active feedback control, yet residual platform motion remains a key factor limiting low-frequency sensitivity and duty cycle. Further reduction of this residual motion is therefore critical for improving the long-term stability and overall astrophysical reach of the observatories. In this work, we develop a data-driven approach that uses machine learning to model and suppress residual seismic motion within the isolation system. Ground and platform sensor data from the detectors are used to train a neural network that predicts platform motion driven by microseismic activity. When incorporated into the control scheme, the network's predictions yield up to an order-of-magnitude reduction in residual motion compared to conventional linear filtering methods, revealing that nonlinear couplings play a significant role in limiting current isolation performance. These results demonstrate that machine-learning-based control can provide a powerful new pathway for enhancing active seismic isolation, improving lock robustness, and extending the low-frequency observational capabilities of gravitational-wave detectors.

Figures (6)

• Figure 1: Illustration of the HAM seismic controls. Sensors on the ground and on the platform are combined to act as a witness of the platform’s inertial motion. Control signals are derived and sent to actuators on the platform.
• Figure 2: Amplitude spectral density (ASD) of the original residual table motion GS13 in X-direction (blue) compared to the prediction based on the sensor outputs from the linear model (green) and residual between the motion and the prediction (orange). All data is taken on October 16, 2023 (GPS time 1381528818) and calibrated with a top-hat function (a), f2 filter (b) and causal filter (c).
• Figure 3: Average coherence between the channels, between 0.1 and 100Hz (a), vs. 0.1--0.3Hz at one time (b), and a different time (c). The system appears to show time dependent cross couplings in the microseismic band.
• Figure 4: Time dependence of a some significant coefficients of channels as predictors of GS13X.
• Figure 5: GS13 sensor output in horizontal (X) direction (a) and corresponding Amplitude spectral density (ASD). The original measured residual table motion (blue) is compared to the predicted table motion (green). The orange line shows the difference between the actual motion and the prediction, indicating the possible improvement if the neural network output is used for active feedback control over the current residual (pink). The grey area indicates the frequency band used during training.