Adaptive Aperture: Techniques & Applications

• Adaptive Aperture (AA) techniques are methods that dynamically adjust the collection or delivery geometry based on context, reducing errors and optimizing performance.
• In astrophotometry, AA photometry uses a dual-aperture strategy to capture total flux and color, effectively mitigating biases from bright star contamination.
• AA methods extend to fields like radiation therapy, computational imaging, and radio astronomy, where adaptive control enhances efficiency, suppresses interference, and adapts to system constraints.

Searching arXiv for the provided topic and key papers. {"query":"Adaptive Aperture (Daugevičius et al., 10 Jun 2025) star clusters adaptive aperture photometry", "max_results": 5} Adaptive Aperture (AA) denotes a family of context-dependent techniques in which the aperture, or the effective aperture, is adjusted to the object, scene, instrumental state, or observing environment. In recent astrophysical usage, AA photometry is a two-aperture strategy developed to obtain integrated colours of unresolved or semi-resolved star clusters that are less biased by the light of projected bright stars (Daugevičius et al., 10 Jun 2025). In other literatures, the same term refers to updating multileaf collimator shapes in online adaptive radiation therapy (Men et al., 2010), changing coded pupil patterns over time in computational imaging (Wang et al., 2015), constructing shape-preserving apertures for irregular galaxies (Huang et al., 2016), frame-by-frame adaptive elliptical apertures for high-cadence photometry (Bowman et al., 2019), active remapping of segmented or obstructed pupils for coronagraphy (Pueyo et al., 2012), dynamically weighted aperture arrays in radio astronomy (Gunst et al., 2020), active-surface maintenance of a millimeter/submillimeter telescope aperture (Tamura et al., 2021), and segmentation of the European Solar Telescope pupil into six subapertures before MCAO is operational (Scharmer et al., 4 May 2026). This suggests that AA is best understood as a technical principle—adaptive control of the collection or transmission geometry—rather than as a single algorithm.

1. Terminological scope and unifying principle

In the cited literature, AA is used in multiple technical domains, but the operational idea is consistent: a fixed aperture is replaced by an aperture whose geometry, weighting, or selection rule is matched to the current measurement problem. In star-cluster photometry, the aperture is changed to suppress bright-star contamination; in radiation therapy, the aperture is a deliverable multileaf collimator opening; in coded imaging, the pupil transmission pattern is changed across frames; in aperture arrays, the effective aperture is synthesized by software-defined weights and beam allocation (Daugevičius et al., 10 Jun 2025).

Domain	Adapted quantity	Representative paper
Star-cluster photometry	Two co-centred circular apertures, $T$ and $C$	(Daugevičius et al., 10 Jun 2025)
Online adaptive radiation therapy	Deliverable MLC apertures and their intensities	(Men et al., 2010)
Scene-adaptive coded imaging	Aperture transmission pattern over time	(Wang et al., 2015)
High-contrast coronagraphy	Effective pupil via two sequential deformable mirrors	(Pueyo et al., 2012)
Mid-frequency aperture arrays	Per-element weights, beam counts, bandwidth allocation, logical stations	(Gunst et al., 2020)
Multi-aperture solar-telescope operation	Six $1.4$ m subapertures from a $4.2$ m pupil	(Scharmer et al., 4 May 2026)

The shared rationale is also similar. Fixed apertures are simple and stable, but they can be suboptimal when contamination, crowding, PSF anisotropy, high-altitude seeing, RFI, or current anatomy changes the mapping between the aperture and the target quantity. A plausible implication is that AA methods arise when the dominant error term is geometric or spatial rather than purely radiometric.

2. Adaptive aperture photometry for unresolved star clusters

In the PHAT-based star-cluster literature, Adaptive Aperture photometry is a two-aperture strategy developed to obtain integrated colours of unresolved or semi-resolved star clusters that are less biased by the light of projected bright stars. AA photometry uses two co-centred circular apertures per cluster: $T$ (“total”), a larger aperture chosen to capture the cluster’s total light, and $C$ (“colour”), a smaller aperture centred on the cluster that encloses its central part and deliberately excludes contaminating bright stars (Daugevičius et al., 10 Jun 2025). The total-flux scale is transferred from $T$ to $C$ by an aperture correction measured in F475W,

$$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$

and then applied in other passbands,

$$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$

Colour indices remain those measured inside the $C$0 aperture,

$C$1

The PHAT-context implementation uses six HST passbands—F275W, F336W, F475W, F814W, F110W, F160W—spanning $C$2–$C$3. Artificial 3D clusters were generated across ages $C$4, masses $C$5, and EFF profiles with $C$6 arcsec and $C$7. The stellar content used a Kroupa IMF $C$8–$C$9, fully stochastic sampling, PARSEC-COLIBRI isochrones, solar metallicity, zero extinction, and $1.4$0. Each model cluster was observed from 100 viewing angles, and images were generated star-by-star with TinyTim PSFs (Daugevičius et al., 10 Jun 2025).

The central empirical result is that reliable colour work requires a colour aperture larger than the half-light radius. The study reports that median $1.4$1 is $1.4$2 across cluster ages, masses, and profiles, validating AA’s core assumption, but the scatter decreases as $1.4$3 increases, with a strong drop near $1.4$4. The recommended constraint is $1.4$5, and in practice $1.4$6 further suppresses stochasticity-driven colour scatter (Daugevičius et al., 10 Jun 2025). This requirement is stronger than merely avoiding contaminants: it is needed to average over central stochasticity.

The dominant failure mode is bright evolved stars. With PMS stars included, $1.4$7 can reach $1.4$8 mag, especially for F475W−F814W at $1.4$9 Myr. Subtracting PMS stars reduces $4.2$0 to typically $4.2$1 mag. The simulations also isolate a projection term: for typical M31 clusters measured with $4.2$2, $4.2$3 can be as large as $4.2$4 mag, while for $4.2$5, $4.2$6 typically falls below $4.2$7 mag. The recommended colour error budget is

$4.2$8

The method has explicit limits. Young clusters $4.2$9 Myr) remain problematic regardless of the aperture size used. Even with large apertures, stochastic IMF sampling of upper-MS stars yields broad parameter posteriors; the 16–84 percentile ranges of $T$0 are $T$1–$T$2 dex and of $T$3 are $T$4–$T$5 dex depending on aperture size. The recommended practice is therefore to use CMD fitting for resolved or semi-resolved clusters $T$6 Myr when possible (Daugevičius et al., 10 Jun 2025). Earlier PHAT catalog work had already implemented the two-aperture procedure interactively for 1477 additional star clusters, producing a homogeneous AA photometry set of 2658 M31 clusters when combined with Paper I, and showed that $T$7-aperture colours follow the locus of stochastic PARSEC-COLIBRI cluster models more closely than $T$8-aperture photometry (Krisciunas et al., 2023).

3. Other adaptive-aperture photometry algorithms in astronomy

Outside the PHAT star-cluster context, AA photometry appears in several distinct astronomical forms. In merging-galaxy photometry, the “adaptive homomorphic aperture” method derives an irregular aperture from the source’s binary shape and expands it by morphological dilation so that the aperture is “quasi-homomorphic” to the source morphology (Huang et al., 2016). The pipeline uses local background estimation on $T$9 pixel grids, thresholding at $C$0 with default $C$1, morphological Opening, connected-component labeling, and a dataset-level choice of the number of dilations. For SDSS, the adopted setting is $C$2 with $C$3; for CFHT/RCS2, $C$4 with the same threshold. Tests on 100,000 simulated galaxies report that with $C$5, the method detects $C$6 of total flux for most galaxies, and on CFHT/RCS2 the photometry was successfully obtained for $C$7 mergers $C$8 completeness) (Huang et al., 2016). Here the aperture adapts to source morphology rather than to bright-star contamination.

A second form is adaptive elliptical aperture photometry in TEA-Phot. In this case the aperture used to sum source flux is an ellipse whose size and orientation are re-determined for every frame from the measured PSF shape in that frame, rather than a fixed circular aperture. TEA-Phot uses SEP to measure centroid $C$9, semi-major axis $T$0, semi-minor axis $T$1, and position angle $T$2, then applies a fixed scale factor $T$3 to the frame-specific ellipse. Background can be estimated with an elliptical sky annulus whose inner and outer bounds are $T$4 and $T$5, or with a “sky star” method (Bowman et al., 2019). The published case study on the roAp star J1640 reports a reduction in differential light-curve RMS from $T$6 mmag with fixed circular apertures to $T$7 mmag with adaptive elliptical apertures, while the amplitude SNR at $T$8 increased from $T$9 to $C$0, and at $C$1 from $C$2 to $C$3 (Bowman et al., 2019). In this usage, adaptivity is frame-by-frame PSF matching.

A third form is deep-learning-based adaptive aperture selection for time-domain stellar photometry. DeepAP defines AA as choosing, for each source and each image, the aperture radii that best balance signal inclusion against noise and contamination. The framework uses a Vision Transformer for feasibility assessment and a ResNet-18 for prediction of the inner aperture, annulus inner radius, and annulus outer radius. For feasibility assessment, the ViT model yields an ROC AUC value of $C$4, and achieves a precision of $C$5, a recall of $C$6, and an F1 score of $C$7 on the test set. For aperture size prediction, the combined framework achieves a median total processing time of $C$8 milliseconds for a batch of 10 images, representing a speed-up of approximately $C$9 times compared to exhaustive aperture size enumeration (Du et al., 5 Aug 2025). This version of AA is neither interactive nor morphology-driven; it is learned from cutouts and explicitly optimized for light-curve SNR.

Taken together, these photometric variants show that “adaptive aperture photometry” does not designate one canonical algorithm. It can mean contaminant-avoiding two-aperture colour measurement, quasi-homomorphic shape dilation, frame-by-frame elliptical matching, or learned feasibility and radius prediction. The common element is that the aperture is not globally fixed.

4. Adaptive pupils in computational imaging and coronagraphy

In computational imaging, Adaptive Aperture means actively changing the aperture’s spatial transmission pattern over time so the optics encode information that is best suited to the current scene and task (Wang et al., 2015). The scene-adaptive coded aperture method addresses a conflict between broadband deconvolution and depth discriminability. Under incoherent imaging, the observation is modeled as

$$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$0

or in the Fourier domain,

$$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$1

The method estimates dominant edge orientations from a pinhole or sharp preview, then sequentially applies oriented aperture codes with an LCoS microdisplay. In each frame, one principal direction is made depth-discriminative, while the orthogonal direction remains broadband. High-confidence per-frame depth and all-focus reconstructions are then propagated temporally across frames (Wang et al., 2015). In this setting, the aperture is adapted to scene orientation statistics.

High-contrast coronagraphy uses a different but related meaning. Active Compensation of Aperture Discontinuities (ACAD) treats AA as active reshaping of an “unfriendly” aperture with central obstruction, spiders, or segment gaps so that it behaves like a friendly, high-throughput pupil for coronagraphy (Pueyo et al., 2012). ACAD uses two sequential deformable mirrors separated by a distance $$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$2 and computes the mirror surfaces by solving the non-linear Monge–Ampère pupil-remapping equation. The input field is

$$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$3

and the remapping is designed so that amplitude discontinuities are smoothed and redistributed before linear fine tuning by standard wavefront-control methods. For geometries similar to JWST, the method can attain at least $$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$4 in contrast, and an order of magnitude higher for future Extremely Large Telescopes and on-axis architectures reminiscent of HST. The reported implementations use two $$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$5 cm DMs, $$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$6 actuators across, $$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$7 m, and maximum surface deformations of $$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$8 in the JWST-like case and $$F475W_{\mathrm{AC}} = F475W_{\mathrm{T}} - F475W_{\mathrm{CA}},$$9 in the TMT case (Pueyo et al., 2012).

These two optical usages are technically distant but conceptually aligned. The coded-aperture method changes the pupil transmission to improve inverse problems in depth and deblurring, while ACAD changes the effective pupil amplitude seen by the coronagraph to suppress diffraction from aperture discontinuities. In both cases the aperture is an active design variable rather than a fixed optical boundary.

5. Adaptive apertures in radiation therapy and radio aperture arrays

In online adaptive radiation therapy, Adaptive Aperture refers to updating multileaf collimator shapes so that the radiation field conforms to the patient’s current anatomy as seen on on-board imaging (Men et al., 2010). In intensity-modulated radiation therapy, an aperture is a physically deliverable MLC configuration at a given gantry angle. Direct aperture optimization formulates planning directly over deliverable apertures and their intensities:

$$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$0

with a convex objective built from one-sided quadratic penalties for underdose and overdose. Because the set of candidate apertures is extremely large, the method uses column generation: a master problem reweights a current working set of apertures, and a pricing problem identifies a new aperture with most negative reduced cost. The pricing step exploits the fact that, under consecutiveness per row, each MLC row reduces to a minimum-sum contiguous subsequence problem (Men et al., 2010).

The GPU implementation is notable because it makes per-fraction AA feasible in practice. On an NVIDIA Tesla C1060 GPU card, optimal treatment plans using 50 MLC apertures were generated in $$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$1–$$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$2 seconds for five 9-field prostate and five 5-field head-and-neck IMRT cases, versus $$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$3–$$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$4 minutes on an Intel Xeon $$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$5 GHz CPU for the same problems. Plan quality agreed to within $$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$6–$$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$7 in objective values, and DVHs and dose distributions showed no discernible differences (Men et al., 2010). Here AA means near-real-time redefinition of a set of deliverable beam openings.

In radio astronomy, the Mid-Frequency Aperture Array uses AA in a software-defined sense. An Aperture Array is a large number of fixed, electronically connected antennas whose signals are combined by analog and/or digital beamformers to produce one or more station beams without mechanical pointing. The telescope can dynamically shape, steer, and optimize its effective aperture on the sky by adjusting per-element, per-tile, or per-station weights, beam counts, bandwidth allocation, and even the logical definition of stations (Gunst et al., 2020). The MFAA document targets $$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$8–$$F?W_{\mathrm{C}} = F?W_{\mathrm{CA}} + F475W_{\mathrm{AC}}.$$9 MHz, best effort down to $C$00 MHz and up to $C$01 MHz, with instantaneous bandwidth up to $C$02 GHz. It states an essential sensitivity requirement $C$03, a desirable value $C$04, and a beam-bandwidth product $C$05 at $C$06 GHz (Gunst et al., 2020). Adaptive beamforming can also use MVDR weights,

$C$07

for RFI suppression and sidelobe control.

The contrast between these two usages is instructive. In radiation therapy, the aperture is a discrete, mechanically deliverable opening optimized by convex programming. In aperture arrays, the aperture is effectively synthesized in software by weights and delays. Both are AA because the collection or delivery geometry is recalculated from the current state of the system.

6. Adaptive aperture control in large telescopes

Large-telescope AA appears in two related forms: active maintenance of surface quality in the millimeter/submillimeter regime, and intentional segmentation of the optical pupil when full-aperture operation is counterproductive. For millimetric adaptive optics, the wavefront sensor based on aperture-plane interferometry measures changes in excess path lengths from characteristic positions on the primary mirror surface to the focal plane by correlating a broadband reference noise injected at those positions with a direct reference path (Tamura et al., 2021). The basic relation is

$C$08

so the slope $C$09 gives EPL directly, and for reflective surfaces near normal incidence $C$10. The prototype uses broadband noise $C$11–$C$12 GHz, an OCTAD-M FX correlator, and time-multiplexed feed horns. The requirement is $C$13 RMS in surface error at $C$14 ms time resolution; the laboratory test obtained residual $C$15 mean $C$16 and $C$17 (Tamura et al., 2021). In this usage, AA is the effective, high-quality aperture maintained by continuously adjusting the primary or secondary shape.

The European Solar Telescope paper defines a different adaptive strategy for the years before MCAO is operational. EST is proposed to operate in multi-aperture mode by optically segmenting the $C$18 m aperture into six $C$19 m subapertures, implemented in the reimaging optics of each Fabry–Pérot system at the second pupil stop $C$20, with no changes to the primary or secondary optical systems or to the actual FPI systems (Scharmer et al., 4 May 2026). The motivation is quantitative. For high-altitude seeing characterized by Fried’s parameter $C$21, the short-exposure phase variance is

$C$22

With $C$23 m, the paper gives $C$24 m $C$25 rad, $C$26, whereas $C$27 m $C$28 rad, $C$29 (Scharmer et al., 4 May 2026). The diffraction limit is worse for the subapertures, but the short-exposure Strehl and the stability of MFBD/MOMFBD reconstructions are much better.

The EST multi-aperture mode also improves the telecentric FPI behavior because each subaperture samples a narrower range of incidence angles on the etalons than the full $C$30 m pupil. For EST-V at $C$31 nm, the full-aperture mode gives FWHM $C$32 pm and peak transmission $C$33, while multi-aperture, HR tilt $C$34, gives FWHM $C$35 pm and peak $C$36 for the usable subapertures (Scharmer et al., 4 May 2026). Switching between full-aperture and multi-aperture is fast—within about an hour with suitable mechanics—and the added cost is only $C$37–$C$38 of the total cost of an FPI system (Scharmer et al., 4 May 2026).

These telescope examples show two complementary meanings of AA. One preserves a large aperture by active metrology and control; the other deliberately reduces the effective aperture because a smaller $C$39 gives better science output under uncompensated high-altitude seeing. The shared criterion is performance of the effective aperture under current conditions, not nominal geometric diameter.