The DOSSA (Decentralization of Space Situational Awareness) project, led by SpaceAble in collaboration with Unistellar and the Laboratoire d’Astrophysique de Marseille, aims to enhance space surveillance through a collaborative effort involving amateur astronomers, researchers, and industrial networks, leveraging advanced technology and data collection to improve space situational awareness in an era of escalating satellite numbers. This project aims to create a comprehensive sky map of objects transiting around Earth. It benefits from a large dataset collected by the Unistellar telescope network, a global network comprised primarily of robotically controllable evScopes 1 and 2 telescopes, each featuring a 11.4 centimeter aperture diameter. We develop dedicated deep learning algorithms to account for the relatively compact diameters of the telescopes and to extend detection thresholds. Firstly, we use traditional convolutional neural networks (CNNs) based classifiers to detect images including a satellite streak, and secondly, we use UNet to identify pixels affected by such streaks. Both neural networks are trained on realistic simulations generated using our optical Fourier simulation software, by combining observed sky backgrounds and synthetic satellite streaks spanning a wide range of orbital parameters. With this strategy, we reach excellent performance on the segmentation of images in test set, with a recall of 79.6% pixels belonging to the satellites masks, at a false positive rate of 0.001%. For 90% of streak pixels recovered by the neural networks, we obtain a precision of 96.3% on the predicted satellite masks. This exceeds the performance of non-ML algorithms and paves the way to measuring accurate satellite positions over broader magnitudes ranges and down to lower S/N, in order to increase the precision on their orbital parameters.
The reliability of Free Space Optical (FSO) communications between a ground station and celestial objects is significantly hampered by the variability in atmospheric conditions. Enhancing the system’s capabilities to recover the received signal can significantly increase the robustness and broaden the operational scope of this type of communication. One of the most promising avenues for improvement entails integrating Adaptive Optics systems with the latest Machine Learning techniques. We study different control laws based on a classical integrator, a LQG with a Kalman filter (with a second order autoregressive model) and a Reinforcement Learning approach: we evaluate the performance of the three control laws with the Strehl ratio.
Predictive control laws for Adaptive Optics (AO) using Artificial Intelligence has been recently explored as an alternative to the classic methods, such as the integrator law. Reinforcement Learning excels in predictive control tasks by enabling systems to learn optimal control strategies through continuous interaction with their environment, adapting to dynamic conditions and achieving effective decision-making in real-time. In our previous work, a Model-based Reinforcement Learning (MBRL) method called Policy Optimization for Adaptive Optics (PO4AO) was used in conjunction with the Object-Oriented Python Adaptive Optics (OOPAO) to simulate the Provence Adaptive Optics Pyramid Run System (PAPYRUS) optical bench. PO4AO demonstrated high adaptability to turbulence and rapid convergence, achieving optimal corrections after just 500 frames of interaction, outperforming a simulated integrator in different atmospheric conditions. Building upon this, our current study explored PO4AO’s capability to adapt to sudden atmospheric changes by worsening turbulence conditions during evaluation, notably the wind speed and the seeing. In the result’s section, we compare PO4AO’s performance in terms of Strehl Ratio (SR) to the integrator. Further description of the experiments are present in the paper.
As the Extremely Large Telescope (ELT) nears operational status, the focus is made on maximizing its capabilities to produce the best possible images. Among other influential factors, the differential pistons – caused by the joint presence of spiders breaking the spatial continuity of the wavefront and low-order aberrations from various origins – significantly impact the overall image quality and must therefore be addressed. Our study is centered on utilizing Neural Networks (NN) to accurately estimate the ELT’s six differential pistons in the presence of turbulence residuals using a 2x2 Shack-Hartmann wavefront sensor (SHWFS). The results of this work could be applied to instruments like HARMONI or MORFEO, which will be equipped with 2x2 SHWFS for their natural guide star wavefront sensing. In our earlier work, the use of ResNet networks fed with data covering all pupil/SHWFS rotation angles demonstrated success in piston estimation. In that study, a simplified atmospheric turbulence model was employed, consisting solely of turbulence residuals, implying perfect correction for frequencies below the deformable mirror’s cutoff frequency. In this paper, we propose to evaluate the network’s performance under more realistic atmospheric turbulence conditions relevant to the ELT. We show that the network can still extract differential piston information from single-frame images, while exhibiting increased uncertainty in its estimation. However, using 10-frame averaged images leads to a significant improvement in root mean square error (RMSE) performance. We conclude this work by addressing remaining open questions and outlining potential future research directions. These findings contribute to refining the NN models usage for cophasing applications, addressing alignment challenges, and enhancing ELT instrumentation performance.
Achieving both high angular resolution and frequent revisit times for Earth (or planet) observation from low Earth orbit poses numerous challenges. The trade-off between increasing aperture size and the associated costs necessitates a novel approach. AZIMOV is a payload prototype project of a 6U CubeSat segmented deployable telescope with an aperture diameter of 30 cm currently in design phase. The large primary mirror enables a 1 m ground sampling distance in the visible. Optimal telescope performance requires precise phasing of the primary mirror, but Cubesat limitations (volume, power, computing) preclude conventional dedicated wavefront sensing methods. Only focal plane sensing appears feasible on small platforms. However, classical methods are iterative and computationally heavy due to the non-linearity between phase and image intensity. In this paper, we investigate deep learning for correcting piston, tip, and tilt aberrations across the primary mirror's four segments from a single focal plane image. We demonstrate diffraction-limited performance on a point source. This method, based on Convolutional Neural Network (CNN), is robust to noise and higher order aberrations, and outperforms classical iterative methods in terms of speed, accuracy and robustness. Finally, when imaging an unknown extended object on Earth’s surface, we demonstrate that our methods can consistently meet diffraction limited performance.
DEEPLOOP is a Python toolbox, originally dedicated to the estimation of the parameters of an Adaptive Optics (AO) Point Spread Function (PSF), describing the atmospheric turbulence and the static modes of a telescope. This toolbox is using the Tensorflow/Keras deep learning API and a Graphical Processor Unit (GPU) computing framework. DEEPLOOP is based on a small set of Python scripts dedicated to the data loading, to the Neural Network (NN) models architectures and their compiling, to the training methods, to the learning curves display and to the performances evaluation on the test sets. This toolbox has a great flexibility: it enables to make simulations on a specific parameters grid (for searching the best hyperparameters configuration), to parallelize the calculations on several GPUs (synchronous data parallelism on the same node), and to use some specific ’on-the-fly’ images loading for each batch, in order to use very few Random Access Memory (RAM). In this paper, we will first explain the main characteristics of this toolbox. Then, the first results with data simulations on Keck II telescope will be presented.
For space-based Earth Observations and solar system observations, obtaining both high revisit rates (using a constellation of small platforms) and high angular resolution (using large optics and therefore a large platform) is an asset for many applications. Unfortunately, they prevent the occurrence of each other. A deployable satellite concept has been suggested that could grant both assets by producing jointly high revisit rates and high angular resolution of roughly 1 meter on the ground. This concept relies however on the capacity to maintain the phasing of the segments at a sufficient precision (a few tens of nanometers at visible wavelengths), while undergoing strong and dynamic thermal gradients. In the constrained volume environment of a CubeSat, the system must reuse the scientific images to measure the phasing errors. We address in this paper the key issue of focal-plane wave-front sensing for a segmented pupil using a single image with deep learning. We show a first demonstration of measurement on a point source. The neural network is able to identify properly the phase piston-tip-tilt coefficients below the limit of 15nm per petal.
The main objective of the COMPASS project is to provide a full scale end-to-end AO development platform, able to address the E-ELT scale and designed as a free, open source numerical tool with a long term maintenance plan. The development of this platform is based on a full integration of software with hardware and relies on an optimized implementation on heterogeneous hardware using GPUs as accelerators. In this paper, we present the overall platform, the various work packages of this project, the milestones to be reached, the results already obtained and the first output of the ongoing collaborations.
For the last few years, Laboratory of Astrophysics of Marseille has been carrying out several R and D activities in Adaptive Optics (AO) instrumentation for Extremely Large Telescopes (ELTs). In the European ELT (D = 40 m) framework, both theoretical and experimental studies are jointly led. A new theoretical approach for AO control command law with large degrees of freedom is being developed: it is based on the use of Local Ensemble Transform Kalman Filter (Local ETKF). In parallel, an experimental multi-purpose AO bench is mounted to allow the validation of new wave-front sensing and correction concepts dedicated to the next generation of ELTs. All the main AO components, with a large number of spatial (up to thousand) and/or temporal (up to 1.5 kHz) frequencies, are available. From different combinations of these AO elements, several correction and sensing (low order and high order frequencies) studies are possible. Our AO bench is combining different corrector mirrors (MEMS deformable mirror from Boston Micromachines and Spatial Light Modulator from Holoeye) which can be used with Shack-Hartmann and Pyramid Wave Front Sensors (respectively, SHWFS and PWFS). For this last type of sensor (PWFS), we will use the world’s fastest and most sensitive camera system OCAM2 (developed at LAM), to demonstrate the concept of a fast and hyper-sensitive PWFS (up to 100x100 sub-pupils) dedicated to the first generation instruments for ELTs.
Freeform optics offer additional degrees of freedom that can lead to a simplification of instrument optical designs with
compact solutions. In this context, we propose a new mathematical description of freeform surfaces. This new
mathematical formalism, based on the "eigen-modes" of Bernstein polynomials was developed for off-axis highly
aspherical surfaces modelling. It allows to take into account different kinds of deformations of the optical surface with
local influence capabilities. We present the mathematical formalism developed and then we focus on the optical analysis
of an innovative instrument design. The advantages provided by this new modelling are examined.
Optimal control laws for new Adaptive Optics (AO) concepts in astronomy require the implementation of techniques
intended for real time identification of the atmospheric turbulence. Contrary to the Optimized Modal
Gain Integrator (OMGI), it has been proved that the Kalman Filter (KF) based optimal control law enables
estimation and prediction of the turbulent phase from the measurements and corrects efficiently the different
modes of this phase in the case of a wide field tomographic AO system. But using such kind of processes, for
any Extremely Large Telescope (ELT), will be extremely difficult because of the numerical complexity of the
computations involved in the matrices calculations as well as the recording of large covariance matrices. A new
control law is proposed, based on the Ensemble Transform Kalman Filter (ETKF) and its efficient variation,
Local ETKF (recently developed for geophysics applications), allowing to dramatically reduce the computation
burden for an ELT implementation and also to deal with non stationary behaviors of the turbulence.
Off-axis highly aspheric optical surfaces modeling needs a new mathematical formalism in order to implement it into
ray-tracing optimization codes. This new description must be able to take into account different kinds of deformations:
from low order to medium and high order deformations. This paper presents a new basis of polynomials (based on
Bernstein polynomials) for a new analytical definition of such optical surfaces. A general definition of Bernstein
polynomials and some of the mathematical properties will be first introduced. Then, we will briefly review some
straightforward tools which can be very useful for optical design and optimization in the use of this specific basis.
We developed a new mathematical formalism to model highly aspherical optical surfaces opening the possibility to
explore innovative optical designs. This formalism is based on Bernstein polynomials allowing to describe from low to
high order deformations of the optical surface. It has been implemented into Zemax making use of the User-Defined
Surface (UDS-DLL) Zemax capability. In this case, the mathematical definition of the surface is imported into Zemax
then allowing to apply classical optimization and analysis functionalities. This paper presents the UDS-DLL tool based
on Bernstein polynomials together with an initial optical analysis performed to evaluate the gain obtained in using such a
new formalism.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.