Optimal Sensor Fusion Method for Active Vibration Isolation Systems in Ground-Based Gravitational-Wave Detectors

TL;DR

This paper introduces an optimization-based $ ext{H}_ extinfty$ synthesis method for designing optimal complementary filters in active seismic isolation systems, improving noise suppression in gravitational-wave detectors.

Contribution

It presents a novel $ ext{H}_ extinfty$ optimization approach for automatically designing superior complementary filters based on sensor noise characteristics, replacing manual tuning.

Findings

Synthesized filters closely match the lower noise bound across frequencies.

The method outperforms pre-designed filters in noise suppression.

Applicable to various sensor noise scenarios.

Abstract

Sensor fusion is a technique used to combine sensors with different noise characteristics into a super sensor that has superior noise performance. To achieve sensor fusion, complementary filters are used in current gravitational-wave detectors to combine relative displacement sensors and inertial sensors for active seismic isolation. Complementary filters are a set of digital filters, which have transfer functions that are summed to unity. Currently, complementary filters are shaped and tuned manually rather than optimized, which can be suboptimal and hard to reproduce for future detectors. In this paper, an optimization-based method called $\mathcal{H}_\infty$ synthesis is proposed for synthesizing optimal complementary filters according to the sensor noises themselves. The complementary filter design problem is converted into an optimization problem that seeks minimization of an objective function equivalent to the maximum difference between the super sensor noise and the lower bound in logarithmic scale. The method is exemplified by synthesizing complementary filters for sensor fusion of 1) a relative displacement sensor and an inertial sensor, 2) a relative displacement sensor coupled with seismic noise and an inertial sensor, and 3) hypothetical displacement sensor and inertial sensor, which have slightly different noise characteristics compared to the typical ones. In all cases, the method produces complementary filters that suppress the super sensor noise equally close to the lower bound at all frequencies in logarithmic scale. The synthesized filters contain features that better suppress the sensor noises compared to the pre-designed complementary filters. Overall, the proposed method allows the synthesis of optimal complementary filters according to the sensor noises themselves and is a better and versatile method for solving sensor fusion problems.