Guidestar-Free Adaptive Optics with Asymmetric Apertures

TL;DR

This work tackles imaging through unknown obscurants without guidestars by marrying asymmetric apertures with machine learning in a closed-loop AO framework. A two-network pipeline first estimates the PSF from natural scene measurements and then reconstructs the underlying phase aberration, which is climaxed by applying the conjugate phase on an SLM for optical correction. The approach achieves high-quality imaging with drastically fewer measurements and orders of magnitude less computation than prior guidestar-free methods, and it is validated both in simulation and with real-world obscurants (nail polish, onion skin, diffuser) in tabletop experiments. By breaking aperture symmetry and leveraging feedforward nets, the method enables passive wavefront sensing in uncontrolled environments and offers a practical route to real-time AO without guidestars or specialized sensors. The demonstrated framework has significant implications for astronomy, biomedical imaging, and surveillance, where reliable AO under natural conditions is highly desirable.

Abstract

This work introduces the first closed-loop adaptive optics (AO) system capable of optically correcting aberrations in real-time without a guidestar or a wavefront sensor. Nearly 40 years ago, Cederquist et al. demonstrated that asymmetric apertures enable phase retrieval (PR) algorithms to perform fully computational wavefront sensing, albeit at a high computational cost. More recently, Chimitt et al. extended this approach with machine learning and demonstrated real-time wavefront sensing using only a single (guidestar-based) point-spread-function (PSF) measurement. Inspired by these works, we introduce a guidestar-free AO framework built around asymmetric apertures and machine learning. Our approach combines three key elements: (1) an asymmetric aperture placed in the optical path that enables PR-based wavefront sensing, (2) a pair of machine learning algorithms that estimate the PSF from natural scene measurements and reconstruct phase aberrations, and (3) a spatial light modulator that performs optical correction. We experimentally validate this framework on dense natural scenes imaged through unknown obscurants. Our method outperforms state-of-the-art guidestar-free wavefront shaping methods, using an order of magnitude fewer measurements and three orders of magnitude less computation.

Figures (12)

• Figure 1: Breaking symmetry with asymmetric triangular apertures enables unique phase retrieval from PSFs.For a circular aperture, the original phase $\phi(x,y)$ and its conjugate-flipped counterpart $\overline{\phi(-x, -y)}$ produce identical PSFs after the Fourier transform, leading to phase ambiguity in intensity-only measurements. In contrast, for a triangular aperture---representative of a broader class of asymmetric apertures such as polygons with an odd number of sides---the wavefront phases $\phi(x,y)$ and $\overline{\phi(-x, -y)}$ yield distinct PSFs. This asymmetry enables unique phase retrieval from intensity measurements, resolving the ambiguity present in symmetric apertures.
• Figure 2: Overview of proposed closed-loop guidestar-free AO system.Light from the scene passes through an obscurant (e.g., nail polish, onion skin, optical diffusers) which introduces an unknown wavefront error $\phi_o$. This aberrated wavefront passes through an asymmetric aperture* before reaching a spatial light modulator (SLM) that introduces a phase delay $\phi_{SLM}$. A lens forms a blurry and distorted image of the scene, the effective/residual wavefront error of this blurry scene is $\phi=\phi_o+\phi_{SLM}$. The PSF U-Net estimates the PSF from the captured image and the Phase U-Net then forms a prediction, $\hat{\phi}$, of the phase error. We add the conjugate of the estimated phase error ($-\hat{\phi}$) to the SLM optically correct the residual aberration. We repeat the process iteratively until the sensor image is sharp. The results in this manuscript were captured with three AO loops (4 measurements). *To realize an asymmetric aperture, we display a checkerboard pattern on a phase-only SLM at the regions where we desire zero amplitude mendoza2014encoding.
• Figure 3: Estimated and ground-truth PSFs.Each column presents a PSF example, arranged in order of increasing Strehl ratio from left to right. The top row shows the ground-truth PSFs, followed by estimates from our method, Robust Kernel pan2016robust, and UFPNet fang2023self. We use the official open-source implementation of UFPNet and the released code of Robust Kernel. Robust Kernel and UFPNet provide sub-optimal kernel estimates because they are designed primarily for image deblurring.
• Figure 4: Simulated results for guidestar-free wavefront shaping compared with NeuWS.This figure presents a qualitative comparison between our method and the state-of-the-art guidestar-free wavefront shaping method, NeuWS feng2023neuws. Each column represents a different method/measurement setting. NeuWS includes both digital and optical reconstruction results, while our method and our method combined with Chimitt et al. chimitt2025wavefront provide only optical correction results. Notably, our approach achieves high-quality image restoration with only four measurements, whereas the performance of NeuWS declines significantly as the number of measurements decreases from 16 to 4. These results demonstrate that our method enables high-quality optical reconstruction using only four measurements, substantially reducing the measurement requirements compared to NeuWS.
• Figure 5: Simulated comparisons against deblurring baselines.This figure compares the performance of Robust Kernel pan2016robust, a non-deep learning image deblurring method, EVSSM kong2025efficient, a state-of-the-art deep learning-based deblurring method, and our proposed approach, which implements optical correction. Each row presents a different scene, and each column displays results from a specific method, including cases before correction and those with no aberrations as references. While both Robust Kernel and EVSSM produce digital corrections, our method performs optical correction and is able to restore image quality in cases where image deblurring algorithms fail.