BRS-T360 Seismic Sensing for Einstein Telescope

• BRS-T360 is a hybrid seismic sensing system that integrates a beam rotation sensor for tilt and a T360 seismometer for horizontal translation detection.
• It employs spectral noise models and acausal-optimum weighting to design coupled MIMO control loops, enhancing active seismic isolation.
• The system co-optimizes blending filters and feedback controllers to adaptively manage sensor noise under varying ground motion conditions.

Searching arXiv for the specified paper and topic context. BRS-T360 is a candidate seismic sensing system for active seismic isolation in the Einstein Telescope, a third-generation gravitational-wave observatory. In the formulation presented in “Multi-scale optimal control for Einstein Telescope active seismic isolation” (Saffarieh et al., 21 Jul 2025), BRS-T360 combines a Beam Rotation Sensor (BRS) as an inertial tilt sensor with T360 as a horizontal seismometer and is embedded in a cross-coupled opto-mechanical control architecture. Its role is not limited to sensing in isolation: it is defined through its integration with blending filters, feedback controllers, actuator dynamics, and a unified cost function based on the “acausal optimum,” which quantifies sensor signal-to-noise ratios across frequencies. Within that framework, BRS-T360 is one of two candidate sensing systems considered, alongside OmniSens (Saffarieh et al., 21 Jul 2025).

1. Definition and constituent sensors

The BRS-T360 configuration consists of two sensing elements with distinct physical observables. The Beam Rotation Sensor is an inertial tilt sensor: it consists of a proof mass suspended from a precision flexure, and ambient ground tilt exerts a torque on this mass. Up-to-date implementations use an interferometric readout (“HoQI” style) to sense the differential vertical displacement of two optical readout points, yielding an angle signal

$$\theta_{\rm BRS}(t) \;=\;\frac{y_2(t)-y_1(t)}{2\,p_r}\,,$$

where $y_2-y_1$ is the vertical differential readout and $p_r$ is the lever arm (Saffarieh et al., 21 Jul 2025).

The second component is the Nanometrics T360 seismometer, described as a standard broadband horizontal seismometer whose manufacturer-quoted acceleration noise is converted into an equivalent amplitude spectral density $S_{x,\rm T360}(f)$ in nm / s$^2$/$\sqrt{\rm Hz}$. In the control design, a transfer-function fit to this curve defines the noise-coloring filter $\eta_{x,\rm sens}(s)$.

Taken together, these two sensors provide tilt and horizontal translation information in separate channels. This suggests that the BRS-T360 designation should be understood not merely as a hardware pairing, but as a sensing basis for a multivariable isolation problem in which rotational and translational dynamics are explicitly coupled.

2. Noise models and sensing characterization

The BRS noise model is defined spectrally. Its total noise spectral density $S_{\theta,\rm BRS}(f)$ is the quadrature sum of flexure thermal noise and interferometric readout noise. The flexure thermal contribution is given by

$$S_{\rm th}(f)=4k_BT\,\phi_{\rm eff}\,k_{el}/(2\pi f)\,.$$

A fit of $S_{\theta,\rm BRS}(f)$ is then used as the coloring filter $y_2-y_1$0 in the control design (Saffarieh et al., 21 Jul 2025).

The T360 is characterized analogously through its quoted acceleration noise, converted to an equivalent amplitude spectral density $y_2-y_1$1. A transfer-function fit to that curve defines $y_2-y_1$2. In this framework, both sensors are therefore represented not only by their physical measurements but by fitted spectral models that enter the synthesis problem directly.

The broader control formulation distinguishes three input noise classes: $y_2-y_1$3 These correspond to ground tilt/translation, displacement-sensor self-noise, and inertial-sensor noise. For BRS-T360, the relevant inertial-sensor terms are supplied by the BRS and T360 fits. A plausible implication is that the fidelity of those transfer-function fits is structurally important, because the optimization weights are derived from them rather than from an abstract white-noise idealization.

3. Cross-coupled opto-mechanical architecture

BRS-T360 is integrated into two primary control loops, one for rotation $y_2-y_1$4 and one for horizontal translation $y_2-y_1$5. The two are coupled by a tilt-to-translation plant term $y_2-y_1$6, so that rotation feeds into translation via

$y_2-y_1$7

The plant transfer matrix is a $y_2-y_1$8 MIMO model,

$y_2-y_1$9

where $p_r$0 is the ground-to-platform rotational transfer function and $p_r$1 is the ground-to-platform translational transfer function (Saffarieh et al., 21 Jul 2025).

Measurements are combined through a sensor-blending matrix

$p_r$2

with complementary high-pass/low-pass pairs satisfying $p_r$3. Actuator dynamics are included as diagonal transfer functions $p_r$4 and $p_r$5 in the loop.

This architecture matters because BRS-T360 is explicitly not a pair of decoupled scalar sensors. The translation channel inherits rotational contamination through $p_r$6, and the control problem is therefore posed as a cross-coupled MIMO synthesis rather than as independent tilt and displacement filtering. A common misconception would be to treat the BRS only as an auxiliary tilt monitor; in the stated formulation, it is part of the main inertial sensing structure for the $p_r$7 loop and indirectly influences horizontal isolation through the coupling term.

4. Acausal-optimum weighting and joint optimal control

The central design principle is a unified “acausal optimum” cost function. In the non-causal ideal limit, one could simultaneously subtract all three noise-coloring contributions: ground, displacement-sensor, and inertial-sensor. Defining $p_r$8, $p_r$9, and $S_{x,\rm T360}(f)$0 as the three amplitude spectra from their transfer-function fits, the acausal optimum spectrum is

$S_{x,\rm T360}(f)$1

The associated frequency-dependent weighting is

$S_{x,\rm T360}(f)$2

optionally with an extra $S_{x,\rm T360}(f)$3 tilt to equalize decades (Saffarieh et al., 21 Jul 2025).

The control design is posed as an H$S_{x,\rm T360}(f)$4 or mixed H$S_{x,\rm T360}(f)$5/H$S_{x,\rm T360}(f)$6 optimization on the closed-loop transfer from all white noises $S_{x,\rm T360}(f)$7 to performance outputs $S_{x,\rm T360}(f)$8. With $S_{x,\rm T360}(f)$9 denoting that block, the cost is

$^2$0

Minimizing $^2$1 drives the closed-loop residual motion down toward the non-causal bound $^2$2, while automatically penalizing gains in frequency bands where the signal-to-noise ratio of the available sensors is poor.

Within this construction, BRS-T360 enters through its sensor noise fits and through the resulting shape of $^2$3. This suggests that the sensor pair is evaluated less by a single scalar noise number than by how its spectral characteristics interact with ambient ground motion and displacement-sensor noise over the full band relevant to seismic isolation.

5. Blending filters, feedback synthesis, and implementation constraints

The blending filters $^2$4 and $^2$5 satisfy

$^2$6

and are parameterized as low-order state-space or rational filters, with orders 1–7 in Table 4 of the source paper. They are initialized with a polynomial complement filter, with crossover near the inertial/displacement noise-intersection, and are then refined by the optimizer. The feedback controllers $^2$7 and $^2$8 are also cast as fixed-order LTI filters, with orders 6–8 (Saffarieh et al., 21 Jul 2025).

Both blending and controller filters are adjusted simultaneously via a nonsmooth subgradient descent, specifically MATLAB’s systune, subject to three stated constraints: stability of all loops, an H$^2$9 constraint on the $\sqrt{\rm Hz}$0-loop sensitivity to impose a finite unity-gain frequency, and an H$\sqrt{\rm Hz}$1 objective on the combined performance outputs.

This one-step co-optimization is significant because BRS-T360 is not tuned by choosing a fixed blending crossover and then designing feedback afterward. Instead, blending filters and feedback controllers are synthesized together under a single objective. The paper further states that, in principle, even mechanical suspension parameters could be co-optimized. A plausible implication is that the BRS-T360 configuration is best understood as one point in a larger design space linking sensor inventory, mechanical compliance, and control architecture.

6. Closed-loop behavior, environmental dependence, and comparative context

The closed-loop translational power spectral density is written as

$\sqrt{\rm Hz}$2

In the BRS-T360 configuration, the reported closed-loop behavior has two specific regimes. At very low frequencies $\sqrt{\rm Hz}$3, the BRS inertial-tilt sensor dominates the loop because of high SNR, and $\sqrt{\rm Hz}$4 falls well below ground. Around the secondary microseism $\sqrt{\rm Hz}$5–$\sqrt{\rm Hz}$6, the T360 contribution takes over, producing an overall reduction of platform motion by $\sqrt{\rm Hz}$7–20 dB relative to open-loop (Saffarieh et al., 21 Jul 2025).

The framework also makes the BRS-T360 response explicitly dependent on environmental conditions. Because the acausal optimum $\sqrt{\rm Hz}$8 depends on the ambient ground PSD $\sqrt{\rm Hz}$9, changes in seismic conditions such as storm versus quiet shift the inversion weighting $\eta_{x,\rm sens}(s)$0 and hence the ideal blending crossover. Under high ground noise, one leans more heavily on the inertial sensor band, whereas under quiet conditions the displacement sensor may be employed over a wider band. Swapping sensor types, such as changing BRS readout noise or using a different horizontal seismometer, simply means refitting the corresponding $\eta_{x,\rm sens}(s)$1 curve and recomputing $\eta_{x,\rm sens}(s)$2; the same optimization procedure then yields new optimal $\eta_{x,\rm sens}(s)$3 and $\eta_{x,\rm sens}(s)$4.

For BRS-T360 specifically, the finite low-frequency tilt compliance of the flexure sets a floor to $\eta_{x,\rm sens}(s)$5. Under very high microseism, the optimal filter temporarily shifts the crossover out of a band of poor BRS SNR, trading a small loss in tilt suppression for overall improved loop stability. This directly addresses a possible misunderstanding that the BRS should always dominate at the lowest frequencies: the synthesis is conditional on SNR and stability, not on a fixed heuristic preference for tilt sensing.

In comparative context, the same study states that OmniSens demonstrates superior low-frequency isolation, reducing platform motion by up to two orders of magnitude near the microseism (Saffarieh et al., 21 Jul 2025). That comparison does not invalidate BRS-T360; rather, it places it as a candidate sensing system within a common optimization framework that supports ready re-optimization and projection of sensor noise to metrics relevant to instrument performance.