Coded Wavefront Sensing: Methods & Applications

• Coded-WFS is a quantitative phase imaging technique that employs known optical masks to encode phase into intensity patterns for computational inversion.
• It leverages diverse optical architectures—from random phase masks to asymmetric pupils—to convert phase gradients into measurable displacement fields.
• Recent implementations in microscopy, adaptive optics, and laser diagnostics highlight its balance of photon efficiency, sensitivity, and reconstruction complexity.

Coded wavefront sensing (Coded-WFS) denotes wavefront and phase-imaging methods in which a known optical modification encodes phase information into measured intensity patterns that are then inverted computationally. In one explicit usage, Coded-WFS is a snapshot quantitative phase imaging technique that places a random phase mask close to the image sensor and leverages the memory effect so that local speckle displacement is related to the gradient of the specimen phase; this formulation was experimentally benchmarked against digital holographic microscopy on static silica beads and dynamic HEK cells, with comparisons of simultaneous bright-field intensity and optical path delay (Kazim et al., 23 Aug 2025). The surveyed literature also suggests a broader Coded-WFS viewpoint in which asymmetric pupils, Fourier-plane filters, coded subapertures, coronagraph-integrated reference beams, propagation diversity, and programmable amplitude masks are all treated as engineered phase-to-intensity encoders (Martinache, 2013, Denk et al., 11 Feb 2026).

1. Definition and conceptual scope

In the narrow sense used for quantitative phase imaging, Coded-WFS is a single-shot, non-interferometric wavefront sensing / phase imaging technique that estimates specimen phase delay from the perturbation of a coded speckle pattern produced by a random phase mask near the sensor. The acquisition is “snapshot” only after calibration, because the method still requires two measurements overall: a reference image in the absence of the specimen and an object image once the specimen is inserted (Kazim et al., 23 Aug 2025).

Related literature uses different names but shares the same organizing principle: phase information is deliberately coded into intensity by an optical transformation that is known or calibrated. This principle is stated explicitly in the pyramid-sensor literature as “The aim of a Wave Front Sensor (WFS) is to code the phase information using an incoming photon flux” (Fauvarque et al., 2015). In Fourier-filtering sensors, the code is the focal-plane mask; in coronagraph-integrated sensing, the code is the structured reference field created by the coronagraph; in image-domain sensing, the code is a non-centrosymmetric pupil; and in programmable computational sensing, the code is a sequence of binary amplitude masks (Chambouleyron et al., 2022, Haffert et al., 2023).

A useful conceptual distinction is between static optical coding and sequential coding. Static coding includes random phase masks, asymmetric pupils, flattened pyramids, Zernike-like phase masks, and coronagraphic reference holes. Sequential coding includes DMD-driven amplitude masks and, in a different sense, multi-plane propagation diversity, where the diversity is created by measurements at several axial planes rather than by a fabricated mask. This suggests that Coded-WFS is less a single instrument than a design pattern: recover the wavefront by inverting a known phase-to-intensity encoder.

2. Physical principles and forward models

The canonical Coded-WFS model for quantitative phase imaging represents the specimen by an exit wave

$e^{j\phi(r)},$

with $\phi(r)$ the phase delay map. The key memory-effect approximation is

$I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$

where $I_0(r)$ is the reference coded image, $I(r)$ is the object image, $z$ is the distance between the phase mask and camera sensor, and $k=2\pi/\lambda$. Equivalently, the local displacement field is

$u(r)=\frac{z}{k}\nabla\phi(r),$

so phase retrieval becomes optical-flow estimation followed by gradient integration (Kazim et al., 23 Aug 2025). The physical interpretation is directly analogous to Shack-Hartmann sensing, except that the motion field is continuous rather than lenslet-sampled.

A second major formulation is the small-aberration image-domain model used by the asymmetric pupil Fourier wavefront sensor. For a point source and a non-centrosymmetric pupil, the phase of the Fourier transform of a single direct image is linearized as

$\Phi=\mathbf{A}\,\varphi,$

or more generally

$\Phi=\Phi_0+\mathbf{A}\cdot\varphi,$

where $\phi(r)$0 is the discrete pupil phase vector, $\phi(r)$1 the sampled Fourier-phase vector, and $\phi(r)$2 the phase transfer matrix (Martinache, 2013). The decisive point is observability: with a symmetric pupil only odd modes appear in the measurable subspace, whereas a physically asymmetric pupil lifts that degeneracy.

A third formulation is coded diffraction phase retrieval. In the DMD-based architecture, the unknown field is

$\phi(r)$3

the $\phi(r)$4-th binary amplitude mask is $\phi(r)$5, and the measurements satisfy

$\phi(r)$6

Reconstruction is posed as the amplitude-flow objective

$\phi(r)$7

and solved with Reweighted Amplitude Flow with Optimal Spectral Initialization (Denk et al., 11 Feb 2026).

Propagation-diversity sensing fits the same template even though it uses no special coding mask. In multi-plane phase retrieval, the pupil field

$\phi(r)$8

is propagated to several defocused planes,

$\phi(r)$9

and tip/tilt appears as plane-dependent lateral shifts

$I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$0

This makes low-order pointing information directly observable in the same intensity data used for higher-order phase retrieval (Abbott et al., 12 Aug 2025).

3. Optical architectures and coding modalities

Approach	Optical code or diversity	Measurement domain
Coded-WFS QPI	Random phase mask close to the image sensor	Reference/object coded images; optical flow and phase integration
FPWFS	Reduced pyramid angle with 90% overlap rate	One recombined pupil intensity map
Coded SHWS	Array of phase-coded masks combined with lens transmittances	Correlation-peak displacement per subaperture
APF-WFS	Minor asymmetric obscuration of the pupil	Fourier phase of a single direct image
Optimized FFWFS	Numerically optimized focal-plane mask $I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$1	Re-imaged pupil intensity
PIAACMC-integrated sensing	$I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$2 phase mask or Lyot-stop reference pinhole	Pupil-plane ZWFS intensity or focal-plane fringes
Multi-plane phase retrieval	Propagation to planes at $I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$3 cm and $I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$4 cm	Defocused intensities and centroid shifts
APUCAM	Sequential binary amplitude masks on a DMD	Multiple far-field intensity patterns

The random-mask QPI implementation is optically close to a standard laboratory microscope except for the addition of a random phase mask near the image sensor. The phase object changes the incident wavefront, the mask converts that change into a local apparent displacement of a speckle-like coded intensity pattern, and the displacement field is related to the specimen phase gradient (Kazim et al., 23 Aug 2025).

The flattened pyramid wavefront sensor changes the coding geometry rather than the detector or reconstruction class. By reducing the pyramid angle, the four pupil images are forced to overlap into a unique intensity, so phase information is recombined optically before detection instead of being recovered from four separated channels by differential combinations (Fauvarque et al., 2015). The coded Shack-Hartmann variant makes a different substitution: each lenslet is replaced by a phase-coded mask plus lens transmittance, so local tilt is estimated from the displacement of a nonlinear-correlation peak rather than from the centroid of a focused spot (Dubey et al., 2021).

Image-domain sensing with APF-WFS moves the code to the pupil itself. A minor asymmetric obscuration makes the science image recoverably sensitive to high-order aberrations, enabling wavefront sensing and segmented-mirror phasing from a single aberrated PSF acquired with the science camera (Pope et al., 2014). Fourier-filtering WFS instead treats the focal-plane mask as the design variable and optimizes it numerically to maximize phase-to-intensity conversion efficiency, so classical ZWFS and PWFS become particular points inside a larger coded-mask design space (Chambouleyron et al., 2022).

Coronagraph-integrated sensing embeds the code into the coronagraphic reference field. In the PIAACMC architecture, a $I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$5-shifted PSF core yields a Zernike wavefront sensor, while an off-axis pinhole in the Lyot stop yields Self-Coherent Camera fringes; the FAST-PIAACMC variant further engineers the focal-plane phase pattern

$I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$6

to boost reference throughput for short-exposure sensing (Haffert et al., 2023). Sequential computational sensing with APUCAM places the code in a programmable DMD, enabling binary amplitude modulation and reference-free phase retrieval across multiple wavelengths (Denk et al., 11 Feb 2026).

4. Reconstruction pipelines and calibration

Reference-based Coded-WFS for QPI uses a compact but calibration-sensitive pipeline: record a reference coded image $I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$7, record an object coded image $I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$8, estimate apparent motion between them with optical flow, convert the motion field into the phase gradient, integrate the gradient to recover $I(r)=I_0\!\left(r-\frac{z}{k}\nabla\phi(r)\right),$9, and simultaneously recover a speckle-free bright-field amplitude for weakly absorbing specimens. The reference is meaningful only if the optical system remains identical except for specimen insertion or removal, so calibration stability is a primary operational constraint (Kazim et al., 23 Aug 2025).

Linearized adaptive-optics variants typically use calibrated interaction matrices. In FPWFS, the detector signal is converted to a meta-intensity

$I_0(r)$0

the interaction matrix is defined on the first 24 Zernike radial orders corresponding to the first 299 Zernike modes, and the outgoing meta-intensity is inverted via the pseudo-inverse of the interaction matrix (Fauvarque et al., 2015). This is a classical small-signal framework: calibration defines the forward operator, and reconstruction is matrix inversion.

Correlation-based and image-domain methods occupy an intermediate position between direct geometric sensing and global inverse problems. The coded SHWS calibrates a reference coded response, cross-correlates each measured subaperture pattern with that reference using nonlinear cross-correlation, localizes the peak by a center-of-mass method, converts the displacement to local slopes, and reconstructs the full wavefront by a zonal reconstruction technique (Dubey et al., 2021). APF-WFS loads a single image, bias-subtracts it, re-centers it, computes its Fourier transform, samples Fourier phases at the baselines generated by the pupil model, applies the transfer-matrix pseudoinverse, subtracts an overall tip-tilt, and then estimates piston, tip, and tilt on each segment from the reconstructed pupil phase (Pope et al., 2014).

Iterative phase-retrieval methods use the coded intensities more directly. Multi-plane phase retrieval uses a modified Gerchberg-Saxton algorithm with up to five iterations and benefits substantially from pre-compensating tip/tilt using centroid estimates from the outer planes (Abbott et al., 12 Aug 2025). APUCAM uses Optimal Spectral Initialization followed by Reweighted Amplitude Flow, then extracts the wrapped phase from the reconstructed complex field and unwraps it by solving the Poisson equation using a discrete cosine transform (Denk et al., 11 Feb 2026). In both cases, the forward model is simple but the inversion is computationally heavier than pseudo-inverse sensing.

5. Reported performance and application domains

Approach	Reported result	Domain
Coded-WFS QPI (Kazim et al., 23 Aug 2025)	Phase maps and bright-field intensity in agreement with DHM; compatibility with narrowband illumination and broadband white-light LED illumination	Static silica beads; dynamic HEK cells
FPWFS (Fauvarque et al., 2015)	98% used photons; 1.1 pixels versus 4 for PWFS at 90% overlap; mean linearity around 50 nm RMS; noise propagation almost as low as ZWFS for radial orders 10 to 20	High-contrast AO
Coded SHWS (Dubey et al., 2021)	Best full-field MSE 0.00075 versus 0.01089 for regular SHWS; best central-crop MSE 0.00045 versus 0.00705	Subaperture slope sensing
APF-WFS hardware (Pope et al., 2014)	Residual wavefront errors of order $I_0(r)$1 nm using 1600 nm light from a starting point of $I_0(r)$2 nm in piston and $I_0(r)$3 mrad in tip-tilt; $I_0(r)$4	Segmented-mirror fine phasing; non-common-path sensing
FAST-PIAACMC (Haffert et al., 2023)	Post-processed sensitivity of $I_0(r)$5 with only several seconds of exposure time	High-contrast coronagraphy
Multi-plane jitter sensing (Abbott et al., 12 Aug 2025)	Outer-plane weighted-average retrieval within approximately $I_0(r)$6 for an unaberrated beam and better than $I_0(r)$7 in aberrated conditions	Tip/tilt sensing for phase retrieval
APUCAM (Denk et al., 11 Feb 2026)	Initial RMSE $I_0(r)$8 to $I_0(r)$9 after five iterations and defocus removal; Strehl $I(r)$0 to $I(r)$1; 3–5 s reconstruction for 20 masks	Quasi-static laser-beam AO at 650 nm; sensing at 2116 nm

These results span several distinct application regimes. In microscopy and QPI, the emphasis is snapshot operation, non-interferometric phase recovery, and compatibility with broadband biological illumination. In astronomical adaptive optics, the emphasis is photon efficiency, detector efficiency, non-common-path sensing, and segmented-aperture phasing. In laser-beam diagnostics, the emphasis is high spatial resolution, broad wavelength adaptability, and operation without wavelength-specific optical elements.

At the design level, optimized Fourier-filtering WFS pushes the coding problem toward a fundamental photon-efficiency viewpoint. The reported strategy designs focal-plane masks so that sensitivity approaches the upper bound $I(r)$2, which the paper treats as the fundamental limit, although the same work also states that the optimized masks inherit the very small dynamic range and strong chromaticity of ZWFS-like sensors (Chambouleyron et al., 2022). This establishes an important theme across Coded-WFS research: coding can improve sensitivity and photon use, but it does not remove all system-level tradeoffs.

6. Limitations, misconceptions, and research directions

A recurring misconception is that “coded” implies either single-shot or reference-free operation. The literature shows that neither implication is universal. Random-mask Coded-WFS for QPI is snapshot only after a reference image has been acquired; APF-WFS uses a single conventional direct image but only in the high-Strehl / small-aberration regime and with an unresolved source; APUCAM is reference-free but requires 20–30 sequential coded measurements and 3–5 s reconstruction time (Kazim et al., 23 Aug 2025, Martinache, 2013, Denk et al., 11 Feb 2026).

A second misconception is that observability can be created purely in software. In APF-WFS the asymmetry must be physically present in the pupil: with a symmetric pupil, only odd modes are sensed, and using an asymmetric computational model without a real asymmetric obscuration showed no sign of improvement at all in the PSF (Pope et al., 2014). The same general lesson applies to other coded architectures: the code must alter the optical transfer behavior, not merely the estimator.

A third misconception is that better coding automatically yields broad capture range or broadband robustness. The summarized literature consistently limits the strongest claims to specific regimes. APF-WFS is linearized around wavefront errors of the order of one radian or less; optimized Fourier-filtering WFS focuses on sensitivity in the linear regime and explicitly retains very small dynamic range and strong chromaticity; the multi-plane jitter-sensing paper demonstrates sensing and image-domain compensation but does not report closed-loop bandwidth or measured real-time residual jitter; the DMD-based sensor is particularly suited to slowly varying or quasi-static laser fields, where computational reconstruction speed is not of the primary concern (Martinache, 2013, Chambouleyron et al., 2022, Abbott et al., 12 Aug 2025, Denk et al., 11 Feb 2026).

Current research directions in the surveyed work are correspondingly architectural. One direction is pupil-specific or PSF-specific code optimization in Fourier-filtering sensors. Another is integrated coronagraphy, where ZWFS and SCC telemetry are combined so that pupil-plane sensing and focal-plane truth sensing coexist within the same PIAACMC platform. A third is self-sensed low-order control, in which tip/tilt is estimated from the same propagation-diversity data used for higher-order retrieval. A fourth is wavelength-flexible programmable coding, where binary amplitude modulation and computational phase retrieval replace wavelength-specific wavefront-sensor optics (Chambouleyron et al., 2022, Haffert et al., 2023, Abbott et al., 12 Aug 2025, Denk et al., 11 Feb 2026). Taken together, these directions suggest that Coded-WFS is increasingly defined by co-design of optical encoder, calibration model, and inverse algorithm rather than by any single canonical sensor geometry.