Tilt-to-Length Coupling Noise

• Tilt-to-Length (TTL) coupling noise is the cross-coupling of angular or lateral jitter into phase signals, resulting in apparent longitudinal displacement errors.
• The noise originates from both geometric effects, like lever-arm and piston contributions, and non-geometric factors, including beam overlap, wavefront mismatch, and detector geometry.
• Experimental and analytical studies in missions such as LISA and TianQin show that precise optical design, beam steering, and calibration maneuvers are key to mitigating TTL noise.

Tilt-to-length (TTL) coupling noise is the cross-coupling of angular or lateral jitter into an interferometric phase signal. In precision interferometry, the induced phase is interpreted as an apparent longitudinal displacement, so TTL appears as spurious length noise rather than as a direct angular readout. The contemporary literature treats the total effect as the sum of geometric contributions, originating from optical path length changes, and non-geometric contributions, originating from wavefront properties, beam overlap, and detector geometry. TTL is therefore not a single mechanism but a family of couplings whose relative importance depends on beam parameters, pivot locations, detector segmentation, signal definition, and downstream time-delay interferometry (TDI) processing. It is a major or primary noise source in LISA, LISA Pathfinder, TianQin, and Taiji, and it is likewise relevant to broader classes of precision interferometers (Hartig et al., 2022, Hartig et al., 2022, Wanner et al., 2024, Wang et al., 2024).

1. Geometric and non-geometric origins

The geometric description of TTL coupling is organized around changes in the optical path length caused by tilts and offsets. In the small-angle limit, the lever-arm effect for a mirror rotation can be written as

$$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} (\varphi^2 - \varphi \varphi_{\mathrm{PD}}),$$

with the simpler form

$$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$

for $\varphi_{\mathrm{PD}}=0$. The piston effect introduces both linear and quadratic terms,

$$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$

and for a rotating receiving system one obtains

$$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$

These expressions show that lateral offsets generate first-order TTL terms, whereas lever arms and longitudinal offsets generate second-order terms (Hartig et al., 2022).

That geometric picture is incomplete. Non-geometric TTL coupling depends on the wavefronts of the interfering beams and on detector geometry. For two interfering fundamental Gaussian beams, the relevant parameters include the absolute and relative beam angles at the detector, relative offsets, and the individual beam parameters; finite detectors and quadrant photodiodes modify the readout further. Beam walk, beam-parameter mismatch, clipping, and the choice of quadrant-combination rule all change the measured pathlength signal. The literature therefore writes the total response as

$$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$

emphasizing that the phase readout contains both path-length and wave-optical contributions (Hartig et al., 2022).

A central consequence is that purely geometric intuition can fail even in nominally simple layouts. For identical beams and ideal centering, non-geometric TTL can nearly vanish; for laterally shifted or mismatched beams, strong linear TTL reappears. Exact or near-exact cancellation between geometric and non-geometric contributions is possible when the pivot is on the beam axis, when beam walk is optically imaged so that the rotation point maps onto the detection point, or when the two contributions are equal in magnitude and opposite in sign. This is why smart geometric design can suppress noise that is not itself geometric: the cancellation problem is set by the total readout, not by a single mechanism (Hartig et al., 2022, Hartig et al., 2022).

2. Beam overlap, detector geometry, and phase-signal definitions

The non-geometric contribution is naturally expressed through the detected overlap integral. A standard pathlength signal is

$$\mathrm{LPS} = \frac{1}{k}\arg\left(\iint E_m E_r^*\, dA_{\mathrm{pd}}\right),$$

where $E_m$ and $E_r$ are the measurement and reference beam fields over the detector area and $k=2\pi/\lambda$. This formulation makes detector geometry explicit: the measured phase depends on the detector aperture, clipping, and segmentation, not only on the incident fields (Chwalla et al., 2020).

For quadrant photodiodes, the signal definition itself becomes part of the TTL problem. The literature distinguishes between a complex-sum formulation and an averaged-phase formulation:

$$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$0

and

$$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$1

Under static alignment imperfections in the LISA test mass interferometer, simulations with three cross-checked methods found that the averaged phase is the choice with equal or less TTL coupling noise in four out of the five test cases. The same study found the non-linear TTL contributions in the test mass interferometer to be negligible in all test-cases, no matter which formulation of pathlength signal is chosen (Edwards et al., 3 Feb 2025).

Related analytical work reaches a consistent conclusion at the detector-composition level. Different ways to combine quadrant signals, including arithmetic-mean phase and LISA Pathfinder-style phase, yield different TTL responses, and the arithmetic-phase definition is generally less sensitive to mismatch than the LPF definition. This does not imply that one formulation is universally optimal in every interferometer; rather, it establishes that phase-signal definition is an instrumental design variable, not a neutral post-processing choice (Hartig et al., 2022).

3. Analytical models from single interferometers to TDI observables

At the level of a single interferometer, a generic first-order TTL model is

$$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$2

For LISA, first-order models were derived both for the individual interferometers and for the TDI Michelson observables, including noise contributions from angular and lateral jitter of the six test masses, six movable optical subassemblies (MOSAs), and three spacecraft. Those models show that the dominant terms in the science interferometers are angular jitter couplings in the long-arm interferometers, while lateral jitter and test-mass-interferometer contributions are strongly suppressed in TDI relative to the raw interferometers (Wanner et al., 2024).

In link-based TDI modeling, TTL noise is written as a delayed sum of receive and transmit angular couplings. For TianQin, a representative expression is

$$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$3

and the link noise enters as

$$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$4

The same framework admits coefficient drift through

$$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$5

and can be extended with quadratic terms when point-ahead angle compensation errors and wavefront errors are relevant (Wang et al., 2024).

A crucial modeling distinction follows. TTL coupling in the individual interferometers and TTL coupling in the TDI observables are not the same estimation problem. In TDI, some physically distinct coefficients become indistinguishable: test-mass and inter-satellite interferometer contributions enter only through combined coefficients, and certain receiver and transmitter jitter coefficients are fully correlated in any single Michelson observable. Simultaneous use of all three Michelson observables resolves the receiver/transmitter degeneracies, whereas a model valid only in TDI is not valid for the raw individual interferometers (Wanner et al., 2024).

4. Experimental suppression by optical design and beam steering

A large experimental literature treats TTL reduction as an optical-design problem. In a LISA-representative long-arm interferometer testbed with a flat-top beam generator, two-lens and four-lens imaging systems were introduced to image the plane of the Rx-clip onto the quadrant photodiode. After careful optimization, TTL coupling factors below $$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$6m/rad for beam tilts within $$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$7rad were obtained. The same study showed that additional TTL coupling due to lateral alignment errors of elements of the imaging system can be compensated by introducing lateral shifts of the detector, and vice versa (Chwalla et al., 2020).

A later experiment demonstrated an active beam-alignment mechanism (BAM) based on two parallel glass windows whose independent rotations produce arbitrary two-dimensional beam displacements while leaving the propagation direction unchanged. In a ground-based optical system, the BAM achieved lateral displacement compensation of the optical axis with a resolution of $$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$8m across a dynamic range of about $$\mathrm{OPD}_{\mathrm{lever}}^{\mathrm{2D}} \approx 2 d_{\mathrm{lever}} \varphi^2$$9mm, and reduced the TTL coefficient from about $\varphi_{\mathrm{PD}}=0$0mm/rad to about $\varphi_{\mathrm{PD}}=0$1m/rad (Qiu et al., 2024).

Ground calibration of TTL has also motivated the development of dedicated actuation hardware. An advanced pure tilt actuator (APTA), monitored by a four-beam interferometer, was designed to suppress the residual translational motion that ordinarily contaminates tilt calibrations. Comparative testing showed that the imaging system is capable of effectively suppressing TTL coupling errors, reducing the TTL coupling coefficients from over plus-minus $\varphi_{\mathrm{PD}}=0$2m/rad to within plus-minus $\varphi_{\mathrm{PD}}=0$3m/rad across a range of plus-minus $\varphi_{\mathrm{PD}}=0$4rad (Lin et al., 2024).

Approach	Reported performance	Source
Two- and four-lens imaging systems	Below $\varphi_{\mathrm{PD}}=0$5m/rad for $\varphi_{\mathrm{PD}}=0$6rad	(Chwalla et al., 2020)
Beam alignment mechanism	About $\varphi_{\mathrm{PD}}=0$7mm/rad to about $\varphi_{\mathrm{PD}}=0$8m/rad; $\varphi_{\mathrm{PD}}=0$9m resolution; about $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$0mm range	(Qiu et al., 2024)
Advanced pure tilt actuator with imaging	Over $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$1m/rad to within $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$2m/rad across $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$3rad	(Lin et al., 2024)

These results establish that optical suppression is feasible, but they also show that suppression remains alignment- and beam-dependent. Earlier LISA test-mass-interferometer experiments already found a dependency of the TTL coupling on beam properties such as the waist size and location, characterized both theoretically and experimentally (Tröbs et al., 2017).

5. Mission-specific manifestations in LISA Pathfinder, LISA, TianQin, and Taiji

In LISA Pathfinder, TTL coupling was the limiting noise source between $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$4 and $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$5mHz before subtraction in post-processing. Analytical modeling identified lever-arm and piston effects, coupling due to transmissive components, and non-geometric beam-walk contributions, and showed that a pure geometric model would not have been sufficient to describe the coupling. The same work demonstrated that only angular, not lateral, realignment of the test masses affects TTL (Hartig et al., 2023).

Subsequent data analysis verified that the alignment dependence predicted analytically was present in the flight data. Realignments performed during the mission were only partially successful, and a similar minimization approach based directly on experimental data was proposed as a useful strategy for LISA (Armano et al., 2023). Long-term analysis then showed that the coupling coefficients drifted slowly, by $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$6m/rad and $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$7 in $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$8 days, with strong temperature correlations of $$\mathrm{OPD}_{\mathrm{piston}}^{\mathrm{2D}} \approx -2 d_{\mathrm{lat}} \varphi + d_{\mathrm{long}} \varphi^2,$$9m/rad/K and $$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$0/K. On the basis of analytical models, these changes were attributed to rotations of the test masses and small distortions in the optical setup, including bending of the optical baseplate during the cooldown experiment (Armano et al., 2024).

For LISA itself, first-order modeling of the interferometers and TDI Michelson observables indicates that expected TTL noise will initially violate the entire mission displacement noise budget, establishing the known necessity to fit and subtract TTL noise in data post-processing. For post-alignment coefficients of about $$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$1mm/rad, the median single-link TTL noise was reported as about $$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$2pm/$$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$3; for pre-alignment coefficients of about $$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$4mm/rad, the median rose to about $$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$5pm/$$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$6 (Wanner et al., 2024). This suggests that alignment prior to subtraction is not merely helpful but structurally coupled to the achievable subtraction residual.

TianQin and Taiji extend the same logic into mission-level post-processing. In TianQin, a null TDI channel $$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$7 was used in simulation to estimate TTL coefficients with drifts to an accuracy of $$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$8m/rad, and it was found necessary to estimate linear drift coefficients and quadratic TTL coefficients to keep TTL noise residuals below the $$\mathrm{OPD}_{\mathrm{RS}}^{\mathrm{2D}} \approx -d_{\mathrm{lat}} \varphi_{\mathrm{RS}} + \frac{1}{2} d_{\mathrm{long}} \varphi_{\mathrm{RS}}^2.$$9pm noise reference curve (Wang et al., 2024). For Taiji, an improved subtraction algorithm was proposed to handle unknown secondary noise floors and the presence of gravitational-wave signals, and simulation based on spacecraft dynamics and attitude-control modeling was reported to validate the algorithm’s robustness across different TTL-coefficient levels (Deng et al., 24 Sep 2025).

6. Calibration, subtraction, estimator bias, and maneuver design

Because TTL is not removed by TDI, the standard strategy is to estimate coupling coefficients from auxiliary angular measurements and subtract the modeled TTL contribution from the science channels. Fisher-information analysis for LISA showed that, after one day of integration, residual TTL noise post-subtraction in the TDI variables can be limited to the $$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$0pm/$$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$1 level by angular sensing noise under a requirement-level angular-jitter model, whereas a more realistic angular-jitter model makes the coefficient uncertainties $$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$2 times larger and limits the subtraction to the $$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$3pm/$$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$4 level (George et al., 2022).

Estimator choice matters. A dedicated uncertainty study found that angular readout noise leads to a systematic bias of the least-squares estimator, depending on the TTL coupling coefficients, jitter, and readout-noise level, while the instrumental-variables estimator converges to an unbiased result as the data-set length increases. The same work presented an equation predicting the estimation bias of least squares due to angular readout noise (Hartig et al., 2024). This implies that TTL subtraction is not only an optical or dynamical calibration problem but also a system-identification problem with nontrivial estimator pathologies.

The presence of gravitational-wave signals does not appear to invalidate the subtraction strategy. In simulated LISA data containing four different gravitational-wave signal types, the gravitational-wave responses had little effect on the quality of the TTL coupling fit and subtraction, and the signal characteristics were not altered by the subtraction procedure (Hartig et al., 2024).

The recent literature therefore emphasizes dedicated calibration maneuvers. For LISA, an explicit maneuver-based approach estimated the $$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$5 linear coupling coefficients by modulating spacecraft attitude or MOSA yaw and found that sinusoidal signals with amplitudes of around $$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$6 nrad and frequencies near $$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$7 mHz are practical and nearly optimal. Simulations enabled estimation of the TTL coefficients with precision below $$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$8m/rad after a total maneuver time of $$\mathrm{LPS}_{\mathrm{TTL}} = \mathrm{OPD} + \mathrm{LPS}_{\mathrm{ng,TTL}},$$9 minutes (Wegener et al., 2024). Closed-loop nonlinear dynamics simulations later showed that regular datasets with amplified TTL coefficients are limited to a relative error of $$\mathrm{LPS} = \frac{1}{k}\arg\left(\iint E_m E_r^*\, dA_{\mathrm{pd}}\right),$$0 by bias and correlations, whereas sinusoidal maneuvers improve the inference to a high accuracy of $$\mathrm{LPS} = \frac{1}{k}\arg\left(\iint E_m E_r^*\, dA_{\mathrm{pd}}\right),$$1 and remove all correlations in the inferred parameters (Heisenberg et al., 2 Feb 2026).

A final operational layer appears in point-ahead angle control. An open-loop control strategy for LISA’s point-ahead angle correction was shown to maximize periods between adjustments at the constellation level and to be optimal from the perspective of estimating and correcting TTL coupling, because long intervals without abrupt coupling-factor changes create longer stable estimation windows (Houba et al., 2022). This suggests that TTL mitigation is inherently cross-disciplinary: optical design, attitude control, estimator design, and TDI channel selection jointly determine the residual noise floor.