NIRISS AMI: High-Contrast JWST Interferometry

• NIRISS AMI is a high-contrast, high-angular resolution mode on JWST that uses a seven-hole non-redundant mask to transform the telescope into a sparse interferometric array for probing structures below the diffraction limit.
• It employs precise Fourier analysis and self-calibration techniques to extract complex visibilities and closure phases, enabling robust detection of faint companions and detailed imaging of circumstellar features.
• The instrument’s advanced design and data reduction pipelines push the boundaries of direct imaging, offering unprecedented insights into exoplanetary systems, compact objects, and circumstellar environments.

The NIRISS Aperture Masking Interferometer (AMI) is a high-contrast, high-angular-resolution mode operated aboard the James Webb Space Telescope (JWST) as part of its Near Infrared Imager and Slitless Spectrograph (NIRISS). By implementing a non-redundant mask in the JWST pupil, NIRISS AMI transforms the telescope into a sparse interferometric array capable of probing structure at and even within the traditional diffraction limit. It has enabled robust detections of faint companions and resolved fine structure in regions inaccessible to conventional coronagraphic imaging, establishing a new scientific regime for exoplanet discovery, circumstellar environment studies, and high-fidelity imaging of compact objects.

1. Historical Development and Core Principles

Aperture masking interferometry originates in 19th-century optical experiments pioneered by researchers like Young, Fresnel, and Fizeau, whose aperture-masked setups enabled measurements at the scale of the diffraction limit by transforming a single telescope aperture into an array of mutually coherent subapertures (Tuthill, 2013). Modern revival and adaptation of these principles in the 1980s led to implementations on large ground-based telescopes, culminating, with the advent of stable space platforms like JWST, in the deployment of the AMI mode on NIRISS.

The central concept is the placement of a physical non-redundant mask (NRM) in the pupil plane, blocking most of the aperture and leaving only a set of small, carefully located holes (“sub-apertures”). Each unique pair of holes forms a baseline that samples a unique spatial frequency in the object’s Fourier plane. Because the baselines are non-redundant—no two holes have the same pairwise separation—each Fourier component in the image encodes unique information, yielding unambiguous fringe phases and amplitudes.

AMI enables the direct measurement of complex visibilities (amplitudes and phases) and higher-order self-calibrating observables such as closure phases (ϕ₁₂ + ϕ₂₃ + ϕ₃₁). Closure phases are particularly immune to common-mode phase errors induced by atmospheric or instrumental effects, which notoriously affect filled-aperture imaging. In the space-based NIRISS context, where atmospheric turbulence is absent, AMI leverages full complex visibilities, allowing for absolute astrometry and high-fidelity model-free image reconstruction (Ford et al., 2014, Sivaramakrishnan et al., 2022).

2. Instrument Design and Observational Strategies

The JWST NIRISS AMI employs a seven-hole non-redundant mask, yielding 21 unique baselines between 1.32 m and 5.28 m. Each subaperture is a hexacon, undersized with respect to the JWST mirror segment. The patented mask configuration is optimized for robust Fourier coverage while minimizing redundancy, necessary for self-calibration and the unique recovery of source structure over a wide range of spatial frequencies (Artigau et al., 2014, Sivaramakrishnan et al., 2022). The PSF is characterized by an Airy envelope from the subaperture and a superimposed high-frequency fringe pattern corresponding to the baseline vectors.

NIRISS AMI is operated with a choice of filters: one wide band (F277W: 2.77 μm) and three medium bands (F380M: 3.8 μm, F430M: 4.3 μm, F480M: 4.8 μm), carefully matched to both instrument sensitivity and the uv-plane coverage necessary for interferometric imaging. The detectors have 65 mas pixels, yielding a field of view of ∼5", and the sampling is Nyquist at the central wavelengths for F430M and F480M.

Target acquisition is achieved with precise sub-pixel placement (POS1 in SUB80 mode), minimizing flat-field and intra-pixel sensitivity errors; dithering is generally avoided unless sub-pixel repeatability is ensured (Sivaramakrishnan et al., 2022). Exposures employ NISRAPID readout to mitigate detector nonlinearities and charge migration. For science data, calibrator star observations are critical—chosen for brightness and spectral type to match the science target and avoid systematics such as differential charge migration.

3. Data Processing, Calibration, and Algorithmic Innovations

The AMI data reduction pipeline centers on the extraction, calibration, and modeling of Fourier-domain observables. Two main extraction pathways are used: image-plane model fitting (ImPlaneIA) and Fourier-plane analysis (AMICAL, SAMPip). Both methods correct for bad pixels, recenters images to sub-pixel precision, and apply super-Gaussian or Hanning windowing to optimize signal isolation (Soulain et al., 2022, Sivaramakrishnan et al., 2022).

Key observables include:

• Squared visibilities, $V_{ij}^2 = |C_{ij}|^2$, where $C_{ij}$ is the complex Fourier coefficient for baseline $ij$.
• Closure phases, $CP_{ijk} = \arg(C_{ij}C_{jk}C_{ik}^*)$.
• For full-pupil “kernel phase” imaging (KPI), kernel phases $\theta = K \cdot R \cdot \varphi$ are computed, leveraging the linear regime of high-Strehl space imaging (Sivaramakrishnan et al., 2022, Kammerer et al., 2022).

Calibration accounts for static and baseline-dependent errors: point-source calibrator stars are used to empirically correct visibility amplitudes and subtract residual closure phases. Recent advancements include Amigo (Desdoigts et al., 10 Oct 2025), an integrated data-driven, end-to-end forward model using JAX and dLux, which handles optical, detector, and readout physics, including non-linear charge migration (“brighter-fatter” effect) and mask metrology errors within a differentiable pipeline, enabling optimal extraction of robust kernel-phase and amplitude observables.

For extended sources, neural network-based deconvolution (supervised CNNs, unsupervised Deep Image Prior networks) has been shown to yield reliable images of complex objects such as Io, resolving spatial structures where traditional Fourier inversion fails due to low fringe contrast (Sanchez-Bermudez et al., 20 Aug 2025).

Self-calibration techniques imported from radio astronomy further enhance phase and amplitude robustness. Under this paradigm, complex visibility gains and baseline-based errors are estimated in tandem with the source structure through iterative least squares and CLEAN deconvolution. This framework enables “real-time” wavefront error sensing, extracting segment piston values with ∼10–15 nm precision (Carilli et al., 15 Oct 2025).

4. Performance Metrics, Limitations, and Error Budget

AMI reaches contrast levels of ∼7.5–10 magnitudes at separations as small as 70 mas (inner working angle) and up to several hundred milliarcseconds (outer working angle) at λ = 3.8–4.8 μm (Sivaramakrishnan et al., 2022, Doyon et al., 2023). The achievable contrast is fundamentally determined by photon noise, flat field and intra-pixel sensitivity errors, static wavefront aberrations, and, for very high SNR observations, by systematics such as charge migration effects in the H2RG detector.

The phase noise for each observable is given by

$$\sigma(\varphi) = \frac{N_h}{N_p V} \sqrt{0.5 (N_p + n_p \sigma_{\mathrm{pix}})}$$

and, for closure phases (triple product),

$$\sigma(\varphi_{CP}) = \frac{N_h}{N_p} \sqrt{0.5 (N_p + n_p \sigma_{\mathrm{pix}}) \sum_{i=1}^3 (1/V_i^2)}$$

where $N_h$ is the number of holes, $N_p$ the collected photons, $n_p$ the sampled pixels, $V$ the visibility, and $\sigma_{\mathrm{pix}}$ the per-pixel noise (Sallum et al., 2023).

In practice, non-idealities inflate the noise budget: early in-orbit results indicate that closure phase scatter can be ∼22 times greater than the pure photon noise estimate, primarily due to static detector effects and imperfect calibration of PSF features impacted by charge migration. Application of sophisticated empirical or data-driven calibration strategies (as in Amigo) has been shown to restore performance close to the photon noise limit, enabling robust detection limits even for challenging cases with faint inner companions or crowded systems (Desdoigts et al., 10 Oct 2025).

AMI’s inherent throughput loss (15% for the 7-hole mask) makes the filled-pupil KPI mode generally preferable for very faint targets and/or when broader uv-coverage is advantageous (Kammerer et al., 2022, Sallum et al., 2019), whereas AMI retains an edge for the highest spatial resolution and robust observables at the smallest angular separations and in regimes severely affected by photon noise from the core and first Airy ring.

5. Scientific Applications and Key Results

NIRISS AMI has provided a new window into parameter spaces previously unavailable to coronagraphic or adaptive optics imaging. Examples include:

• Direct detection and high-precision astrometry of exoplanets at separations close to λ/2D, such as PDS 70 b and c at 4.8 μm, with SNR ∼15 and ∼7, enabling studies of circumplanetary disk emission and placing deep upper limits ($>7$ magnitudes) for additional companions within 110–250 mas (Blakely et al., 19 Apr 2024).
• Robust imaging of circumbinary disks and colliding-wind phenomena (WR 137), resolving fine-scale (200–300 mas) linear dust filaments and constraining wind-driven dust formation mechanisms through matched geometric modeling (Lau et al., 2023).
• Detailed mapping and flux calibration of volcanic hot spots on Io, including dynamical characterization of eruption centers with neural deconvolution yielding flux and positional measurements at the ∼10% and sub-degree level, respectively (Sanchez-Bermudez et al., 20 Aug 2025).
• Discovery and confirmation of tight binary systems (e.g., AB Dor C at Δm ≈ 4.5, 326.8 ± 0.5 mas from AB Dor A) with precision surpassing prior ground-based capabilities (Doyon et al., 2023).
• Non-detection and strong exclusion of companions more massive than 10–12 M_Jup at 10–20 AU in high-contrast studies (e.g., HIP 65426) (Ray et al., 2023).

AMI’s ability to extract both closure phases and direct complex visibilities (in space, without atmospheric corruption) uniquely enables absolute astrometry and “model-free” imaging of AGNs (Ford et al., 2014), exoplanetary systems, and compact circumstellar environments. The small inner working angle (∼70 mas) is uniquely positioned to complement JWST coronagraphy (IWA ∼400 mas), particularly for inner planetary system architectures.

6. Future Directions and Methodological Developments

Several methodological advances are rapidly improving the fidelity and scientific reach of NIRISS AMI:

• End-to-end calibration frameworks (e.g., Amigo (Desdoigts et al., 10 Oct 2025)) now model the full optical, detector, and electronic train within a differentiable, data-driven pipeline, addressing systematics at the level required for 10-magnitude contrasts and robust kernel-phase recovery. These frameworks are openly available to the community for extension and integration.
• Interferometric self-calibration and CLEAN imaging, adapted from radio astronomy, have been demonstrated to yield dynamic range improvements of ∼23% and provide real-time wavefront error sensing with 10–15 nm accuracy on segment pistons (Carilli et al., 15 Oct 2025).
• Hybrid approaches such as kernel phase imaging (KPI) are enabling fainter target observations, using full-pupil images and high-throughput path, with performance that matches or exceeds AMI beyond ∼325 mas (Kammerer et al., 2022).
• Ongoing commissioning and calibration efforts focus on curated calibrator libraries, optimal observing schemes (matching brightness, maximizing groups per integration), and advanced destriping or pattern noise corrections to further mitigate residuals and approach the fundamental noise floor (Sallum et al., 2023).
• Bayesian and neural inference tools (e.g., fouriever toolkit, neural deconvolution) support robust uncertainty propagation and parameter estimation even for extended and complex targets (Sanchez-Bermudez et al., 20 Aug 2025).

7. Scientific and Technical Legacy

The scientific impact of NIRISS AMI is evidenced across diverse domains:

• Opening access to parameter spaces—separations and contrast regimes—previously impenetrable from the ground (e.g., inner working angles < λ/D, planetary companions within 0.1″).
• Advancing spatially resolved studies of circumstellar and circumbinary environments, exoplanet demographics, AGN fueling structures, and Solar System bodies.
• Enabling robust, high-precision astrometry and photometry critical for binary orbit determination, atmospheric retrievals, and constraining planetary formation models.
• Demonstrating the power and necessity of integrative calibration models and data-driven, machine-learning-enabled inference for extracting maximal scientific value from next-generation space-based interferometric data.

NIRISS AMI, through its innovative design, combined with continual algorithmic and calibration improvements, remains a central tool for high angular resolution, high-contrast astrophysics in the mid-infrared, serving both as a pathfinder for future mission concepts and as a platform for community-driven innovation in observational methodology.