Making the unmodulated pyramid wavefront sensor smart

R. Landman, S. Y. Haffert, J. D. Long, J. R. Males, L. M. Close, W. B. Foster, K. Van Gorkom, O. Guyon, A. D. Hedglen, P. T. Johnson, M. Y. Kautz, J. K. Kueny, J. Li, J. Liberman, J. Lumbres, E. A. McEwen, A. McLeod, L. Schatz, E. Tonucci, K. Twitchell · 2025

Read Making the unmodulated pyramid wavefront sensor smart by R. Landman, S. Y. Haffert, J. D. Long, J. R. Males, L. M. Close, W. B. Foster, K. Van Gorkom, O. Guyon, A. D. Hedglen, P. T. Johnson, M. Y. Kautz, J. K. Kueny, J. Li, J. Liberman, J. Lumbres, E. A. McEwen, A. McLeod, L. Schatz, E. Tonucci, K. Twitchell (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-16690.txt.

Introduction

Telescope optics must sense warped starlight and correct it thousands of times a second, and pyramid sensors are the tool of choice. Landman and a large team of co-authors report the first sky test of such a sensor run without its usual dithering motion. Dithering widens the sensor's range but costs sharpness and blinds it to some errors. A neural network instead untangles the raw sensor output.

Running on a telescope in Chile, the smart sensor held a stable correction loop at over two thousand cycles a second. On bright stars it nearly matched the classic dithered setup. On a fainter star in strong wind it did better. The team plans faster loops, easier calibration, and tests on fainter stars, where the gain should grow.

Catalog record

CREATORR. Landman, S. Y. Haffert, J. D. Long, J. R. Males, L. M. Close, W. B. Foster, K. Van Gorkom, O. Guyon, A. D. Hedglen, P. T. Johnson, M. Y. Kautz, J. K. Kueny, J. Li, J. Liberman, J. Lumbres, E. A. McEwen, A. McLeod, L. Schatz, E. Tonucci, K. Twitchell
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16690

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Making the unmodulated pyramid wavefront sensor smart R. Landman, S. Y. Haffert, J. D. Long, J. R. Males, L. M. Close, W. B. Foster, K. Van Gorkom, O. Guyon, A. D. Hedglen, P. T. Johnson, M. Y. Kautz, J. K. Kueny, J. Li, J. Liberman, J. Lumbres, E. A. McEwen, A. McLeod, L. Schatz, E. Tonucci, K. Twitchell Abstract Pyramid wavefront sensors (PWFSs) are the preferred choice for current and future extreme adaptive optics (XAO) systems. Almost all instruments use the PWFS in its modulated form to mitigate its limited linearity range. However, this modulation comes at the cost of a reduction in sensitivity, a blindness to petal-piston modes, and a limit to the sensor’s ability to operate at high speeds. Therefore, there is strong interest to use the PWFS without modulation, which can be enabled with nonlinear reconstructors. Here, we present the first on-sky demonstration of XAO with an unmodulated PWFS using a nonlinear reconstructor based on convolutional neural networks. We discuss the real-time implementation on the Magellan Adaptive Optics eXtreme (MagAO-X) instrument using the optimized TensorRT framework and show that inference is fast enough to run the control loop at ¿2 kHz frequencies. Our on-sky results demonstrate a successful closed-loop operation using a model calibrated with internal source data that delivers stable and robust correction under varying conditions. Performance analysis reveals that our smart PWFS achieves nearly the same Strehl ratio as the highly optimized modulated PWFS under favorable conditions on bright stars. Notably, we observe an improvement in performance on a fainter star under the influence of strong winds. These findings confirm the feasibility of using the PWFS in its unmodulated form and highlight its potential for next-generation instruments. Future efforts will focus on achieving even higher control loop frequencies (¿3 kHz), optimizing the calibration procedures, and testing its performance on fainter stars, where more gain is expected for the unmodulated PWFS compared to its modulated counterpart. II. First on-sky demonstration of extreme adaptive optics with deep learning Table 1: Targets, conditions, and results of the on-sky tests. Date Target I-band magnitude Observing band/ wavelength Modulation radius Reconstruction method Seeing Loop frequency Estimated Strehl ratio 2024-11-16 \alpha Eri 0.6 CH4 - Neural Network \sim 0.43” 2 kHz 58.1% 2024-11-16 \alpha Eri 0.6 CH4 3 \lambda/D Linear \sim 0.43” 2 kHz 62.7% 2024-11-16 AF Lep \sim 5.7 z - Neural Network \sim 0.60” 2 kHz 40.1% 2024-11-16 AF Lep \sim 5.7 z 3 \lambda/D Linear \sim 0.60” 2 kHz 20.6% 2024-11-16 \pi Pup 0.55 CH4 - Neural Network \sim 0.48” 2 kHz 48.6% 2024-11-20 \pi Pup 0.55 CH4 - Neural Network \sim 0.63” 2 kHz 22.2% ∗ 2024-11-20 \pi Pup 0.55 CH4 - Neural Network \sim 0.63” 3.6 kHz 27.8% ∗,+ 2024-11-20 \pi Pup 0.55 CH4 3 \lambda/D Linear \sim 0.63” 2 kHz 33.0% ∗ • Notes. The CH4 narrowband filter has a central wavelength of 875 nm and a bandwidth of 26 nm. ∗ The Strehl ratio was estimated on the companion due to saturation of the PSF of the primary. + Only frames unaffected by the latency issue were selected. 1 Introduction Extreme adaptive optics (XAO; Guyon 2018 ) is essential for reaching the diffraction limit of both current 10-meter-class telescopes and next-generation extremely large telescopes (ELTs). Enhancing the performance of these XAO systems is critical for directly imaging Earth-like planets in reflected light using future high-contrast imaging instruments. The majority of current and planned XAO systems use a pyramid wavefront sensor ( Ragazzoni 1996 , PWFS;) as their primary wavefront sensor ( Pinna et al. 2016 ; Fitzsimmons et al. 2020 ; Kasper et al. 2021 ; Males et al. 2022 ; Males et al. 2024 ; Haffert et al. 2024 ; Bond et al. 2022 ; Perera et al. 2022 ; Boccaletti et al. 2022 , e.g.,) . The preference for the PWFS stems from its enhanced sensitivity compared to the Shack-Hartmann wavefront sensor, which enables faster speeds and better performance on fainter stars ( Ragazzoni & Farinato 1999 ; Correia et al. 2020 , e.g.,) . A major disadvantage of the PWFS is that it has a nonlinear response, limiting the dynamic range that can be obtained with conventional reconstruction techniques ( Deo et al. 2019 ; Chambouleyron et al. 2020 , e.g.,) . To mitigate this, most instruments use modulation to increase its linearity range. However, there are several disadvantages to this modulation. First, it reduces the PWFS’s sensitivity, especially to low-order modes, limiting its loop speed and degrading the performance on faint stars ( Chambouleyron et al. 2023 ; Agapito et al. 2023 ) . Second, it hinders its ability to sense petal-piston modes, which are critical for future segmented telescopes ( Bertrou-Cantou et al. 2022 ; Hedglen et al. 2022 ; Engler et al. 2022 ) . Finally, the modulator also puts a limit on the speed at which the control loop can be run on bright stars, as the beam needs to be modulated with at least the same speed as the control loop frequency. A promising approach to avoid the need for modulation is the use of nonlinear reconstructors. These algorithms can provide a software-based solution without a reduction in sensitivity. There have been many different attempts at developing nonlinear reconstructors for the PWFS, but their effectiveness has not yet been demonstrated on-sky. A first group of reconstruction algorithms use optical models of the PWFS to perform a nonlinear estimation of the wavefront ( Hutterer & Ramlau 2018 ; Frazin 2018 ; Hutterer et al. 2023 ; Chambouleyron et al. 2024 ; Haffert 2024 , e.g.,) . These methods rely on an accurate model of the optical system of the PWFS and often need multiple iterations to converge to an accurate reconstruction. A second group of algorithms use neural networks (NNs) as approximate arbitrary functions in an attempt to learn the inverse relation between PWFS measurements and the wavefront ( Swanson et al. 2018 ; Landman & Haffert 2020 ; Archinuk et al. 2023 ; Wong et al. 2023 ; Pou et al. 2024b ; Weinberger et al. 2024 ) . While many works have shown the capability of these data-driven algorithms for nonlinear reconstruction, there have been no on-sky demonstrations so far. The main issues for on-sky XAO tests are the real-time implementation and the large gap between simulations, lab experiments, and on-sky conditions. In Landman et al. 2024 , we presented closed-loop lab experiments with the Magellan Adaptive Optics eXtreme (MagAO-X) instrument using a convolutional neural network (CNN) reconstructor. In this Letter we present the first on-sky demonstration of a nonlinear reconstructor for the unmodulated PWFS of the MagAO-X system. Section 2 discusses the real-time implementation of the CNN reconstructor within the MagAO-X software infrastructure. Section 3 presents the on-sky results. Finally, Sect. 4 summarizes the results and lists our conclusions. Figure 1: Steps required to run the NN on-sky. First, training data is collected using the internal source by applying random shapes on the DM and recording the resulting PWFS images. After that, the NN is trained on these data and is converted to an optimized TensorRT engine. This engine is loaded in the MagAO-X control software and runs in real time, picking up the

Read the complete text