Aperture Masking Interferometer

Updated 2 July 2026

Aperture Masking Interferometer (AMI) is a modular optical technique that uses non-redundant masks to convert a single-aperture telescope into a sparse interferometric array for high-resolution imaging.
It utilizes closure phase and Fourier modeling to achieve self-calibration, mitigating wavefront errors and enhancing dynamic range in challenging observational regimes.
AMI’s design supports breakthrough applications such as exoplanet detection, circumstellar imaging, and mapping fine solar system features with sub-diffraction resolution.

Aperture Masking Interferometer (AMI) is a modular optical instrument that transforms a single-aperture telescope into a sparse interferometric array by selectively transmitting light through a small set of non-redundant subapertures. This configuration enables direct measurement of complex visibilities and robust self-calibrating observables such as closure phase, providing angular resolution and contrast capabilities beyond classical imaging. AMI is now implemented on leading facilities ranging from ground-based 8–10 m telescopes with adaptive optics to the James Webb Space Telescope, realizing sub-diffraction-limited imaging at optical and infrared wavelengths (Tuthill, 2013, Soulain et al., 2022, Lau et al., 2023).

1. Theoretical Principles and Imaging Formalism

AMI operates by placing a non-redundant mask at a re-imaged pupil plane, restricting the full telescope aperture to $N$ small holes (subapertures) so that all baseline vectors $\mathbf{B}_{ij}$ between holes $i$ and $j$ are unique. The resulting pupil transmission function $M(x,y)$ acts to modulate the incoming wavefront $A(x,y)$ such that the observed point-spread function (PSF) is a linear superposition of interference patterns ("fringes") from all unique baselines: $PSF(\alpha, \delta) = \left| \mathcal{F}[A(x,y)M(x,y)e^{i\phi(x,y)}](\alpha, \delta) \right|^2,$ where $\mathcal{F}$ denotes the Fourier transform, and $\phi(x,y)$ encodes residual wavefront error (Ford et al., 2014, Sivaramakrishnan et al., 2022, Desdoigts et al., 10 Oct 2025). The van Cittert–Zernike theorem guarantees that the mutual coherence (complex visibility) measured on each baseline is the Fourier transform of the sky intensity at the corresponding spatial frequency: $V(u,v) = \iint I(\alpha, \delta) \exp[-2\pi i(u \alpha + v \delta)] \, d\alpha d\delta,$ with $\mathbf{B}_{ij}$ 0. Unique baseline sampling ensures that each observed fringe encodes an independent visibility.

The key observable for phase self-calibration is the closure phase, defined for a triangle of holes as: $\mathbf{B}_{ij}$ 1 which is mathematically immune to any hole-based (piston) phase error, since such errors cancel around a closed loop (Tuthill, 2013, Sivaramakrishnan et al., 2022, Ford et al., 2014).

2. Mask Design, Instrumentation, and Observing Modes

Mask designs are typically fabricated in metal or silicon to micron-level precision and mounted in a pupil wheel at a re-imaged telescope stop. The design goal is strict non-redundancy—no two hole pairs share the same vector separation—maximizing Fourier-plane coverage and self-calibration robustness.

Ground-Based Masks: Examples include the 7-, 9-, and 18-hole masks on VLT/NAOS–CONICA and 7-, 9-, and 21-hole masks on Keck/NIRC2, often paired with adaptive optics for high Strehl ratios (Tuthill et al., 2010, Tuthill, 2013). The number of baselines is $\mathbf{B}_{ij}$ 2, routinely ranging from 21 to over 150, yielding dense coverage for model-independent imaging and high-contrast detection (<5–7 magnitudes at $\mathbf{B}_{ij}$ 31–2 $\mathbf{B}_{ij}$ 4 in $\mathbf{B}_{ij}$ 51–2 h exposures) (Tuthill, 2013).

JWST/NIRISS Implementation: NIRISS features a fixed seven-hole non-redundant mask (hexagonal, $\mathbf{B}_{ij}$ 60.82 m holes on selected segments of the 6.5 m primary), covering $\mathbf{B}_{ij}$ 715% of the pupil area and yielding 21 uniquely sampled baselines from 1.3 to 5.3 m (Soulain et al., 2022, Sivaramakrishnan et al., 2022). The AMI mode operates in four filters (F277W, F380M, F430M, F480M; 2.77–4.8 $\mathbf{B}_{ij}$ 8m), achieving an inner working angle (IWA) of $\mathbf{B}_{ij}$ 9– $i$ 0 mas and an outer working angle (OWA) of $i$ 1 mas, with overall system throughput $i$ 210–15% (Sivaramakrishnan et al., 2022, Lau et al., 2023).

3. Data Reduction, Calibration, and Self-Calibrating Observables

AMI data analysis follows a pipeline optimized for extracting robust interferometric quantities:

Image Calibration: Detector corrections (bias, dark, flat), cosmic ray flagging, and careful bad-pixel masking are applied.
Fourier Extraction: Each interferogram is centered and transformed. At each baseline's unique spatial frequency in the (u,v) plane, the complex visibility is extracted—either via aperture photometry on the Fourier peaks or by fitting a parametric PSF model in the image plane (Greenbaum et al., 2014, Soulain et al., 2022).
Self-Calibration: Closure phases are constructed for all independent triangles (35 for a 7-hole mask), and fringe amplitudes (squared visibilities) for all baselines. Calibration uses point-source reference stars to correct for instrumental and detector biases, with kernel-phase techniques employed to extend sensitivity to more general pupil geometries (Sivaramakrishnan et al., 2022, Desdoigts et al., 10 Oct 2025).
Forward Modelling: For high-contrast applications and accurate error budgets, end-to-end differentiable forward-models (e.g., Amigo) now jointly model the optics, detector nonlinearities (notably the Brighter-Fatter Effect), and calibrator/target data to achieve contrast floors near the photon-noise limit (Desdoigts et al., 10 Oct 2025, Charles et al., 13 Oct 2025).

These techniques enable contrasts of $i$ 3– $i$ 4 mag at $i$ 5 for JWST, significantly outperforming previous ground-based NRM systems (Sivaramakrishnan et al., 2022, Sallum et al., 2023).

4. Image Reconstruction, Deconvolution, and Algorithms

AMI is intrinsically underdetermined, as the number of sampled Fourier modes is far less than for a filled aperture. Several algorithmic strategies have been developed:

Nonlinear Model Fitting: Binary or multi-component models are fit to calibrated visibilities and closure phases, using either grid search or Markov Chain Monte Carlo for detection and parameter inference (Soulain et al., 2022, Ray et al., 2023).
Iterative Image Reconstruction: Maximum entropy (BSMEM), regularized maximum likelihood (DORITO), total variation (TV), quadratic variation (QV), and CLEAN deconvolution are all now routinely employed (Lau et al., 2023, Charles et al., 13 Oct 2025, Carilli et al., 15 Oct 2025). These algorithms use positivity, entropy, or sparsity priors, and may constrain images using either visibilities, closure phases, or, in advanced pipelines, kernel-phase/amplicude observables in a decorrelated DISCO ("Delay-Insensitive Subspace of Calibrated Observables") basis (Charles et al., 13 Oct 2025, Desdoigts et al., 10 Oct 2025).
Neural Network Deconvolution: For complex targets such as Jupiter's moon Io, modern pipelines incorporate convolutional neural networks (supervised/unsupervised) to deconvolve AMI interferograms and recover spatial structure (Sanchez-Bermudez et al., 20 Aug 2025).

Performance typically reaches the formal interferometric resolution $i$ 6, routinely $i$ 7 mas at NIRISS wavelengths ( $i$ 8– $i$ 9 mas at 3.8–4.8 $j$ 0m) and achieves dynamic range up to $j$ 1– $j$ 2 on calibrators and $j$ 3 on science targets (Lau et al., 2023, Carilli et al., 15 Oct 2025).

5. Achievable Resolution, Contrast, and Scientific Applications

AMI’s unique combination of self-calibrating phase invariants, robust Fourier modeling, and tailored imaging methods produces several distinctive advantages:

Angular Resolution: Effective resolution is set by the longest baseline, $j$ 4 ( $j$ 5– $j$ 6 mas for NIRISS), consistently twice as fine as the Rayleigh limit of the filled aperture (Sivaramakrishnan et al., 2022, Lau et al., 2023).
Contrast and Dynamic Range: Photon-noise-limited closure phase errors of $j$ 7 rad permit raw contrasts of $j$ 8 mag at $j$ 9 in simulation and up to $M(x,y)$ 0– $M(x,y)$ 1 mag on-sky, with closure amplitude and kernel observables further improving robustness (Soulain et al., 2022, Desdoigts et al., 10 Oct 2025, Ray et al., 2023).
Science Cases:
- Exoplanets: Detection and photometry of young giant planets and brown dwarfs interior to the coronagraphic IWA, with sensitivity to objects at $M(x,y)$ 210– $M(x,y)$ 3 au (at 100 pc) and mass limits down to 1–3 $M(x,y)$ 4 (Artigau et al., 2014, Ray et al., 2023).
- Circumstellar Environments: Imaging of dusty Wolf–Rayet binaries (e.g., WR 137), pinwheel nebulae, and protoplanetary disks, revealing structure at sub-diffraction scales (Lau et al., 2023, Carilli et al., 15 Oct 2025).
- Active Galactic Nuclei (AGN): Mapping of nuclear torii, dual SMBH, and narrow-line structures in nearby galaxies, exploiting absolute phase stability and high contrast at $M(x,y)$ 5– $M(x,y)$ 6 mas (Ford et al., 2014).
- Solar System Science: Unprecedented imaging of features on Io at 4.3~ $M(x,y)$ 7m, resolving volcanic hotspots and tracking surface motion with $M(x,y)$ 8 km resolution (Sanchez-Bermudez et al., 20 Aug 2025).

6. Limitations, Detector Systematics, and Future Developments

Despite these advances, AMI faces significant instrumental and methodological constraints:

Detector Systematics: On JWST, non-linear charge migration (Brighter-Fatter Effect) in the H2RG sensor severely distorted raw visibilities, necessitating sophisticated, data-driven forward models (e.g., Amigo) and matching of calibrator and science well depths, groups, and dithers to minimize bias (Sallum et al., 2023, Desdoigts et al., 10 Oct 2025, Charles et al., 13 Oct 2025).
Wavefront Error and Phasing: Segment piston and tilt errors currently limit the practical image dynamic range to $M(x,y)$ 9– $A(x,y)$ 0, with per-segment piston measurable to $A(x,y)$ 1 nm via self-calibration (Carilli et al., 15 Oct 2025).
Sparse uv Coverage: The number of independent Fourier modes is limited by mask design; denser sampling via mask rotation or larger $A(x,y)$ 2 is an active area of development (Carilli et al., 13 Mar 2025, Carilli et al., 2024).
Calibration Overheads: Achieving ultimate contrast requires repeated interleaved calibrator observations, library-based PSF interpolation, and careful matching of observing parameters (Sallum et al., 2023).
Algorithmic Bias: Imaging reconstructions are sensitive to regularizer priors (e.g., entropy, TV) and may suppress sharp features or bias diffuse flux. Future work seeks to leverage generative models, Bayesian sampling, and deep priors for systematic error control (Charles et al., 13 Oct 2025).

Planned enhancements include photonic pupil remapping, advanced regularization (wavelets, sparsity), and extension to kernel-phase imaging to reach contrasts of $A(x,y)$ 3 at sub- $A(x,y)$ 4 scales on upcoming ELTs and future space missions (Tuthill, 2013, Sivaramakrishnan et al., 2022).