Related Experiment Video
Updated: Mar 1, 2026
10:35
Bringing the Visible Universe into Focus with Robo-AO
Published on: February 12, 2013
20.1K
Comparative Study of Neural Network Frameworks for the Next Generation of Adaptive Optics Systems
Carlos González-Gutiérrez1, Jesús Daniel Santos2, Mario Martínez-Zarzuela3
1Mining Exploitation and Prospecting Department, University of Oviedo, 33004 Oviedo, Spain. gonzalezgcarlos@uniovi.es.
Sensors (Basel, Switzerland)
|June 3, 2017
Summary
This study presents CARMEN, a machine learning-based tomographic reconstructor for adaptive optics systems. Native CUDA code offers the best performance for correcting atmospheric turbulence in large telescopes.
Area of Science:
- Astronomy and Astrophysics
- Optical Engineering
Background:
- Next-generation telescopes require advanced adaptive optics (AO) for atmospheric turbulence correction.
- Tomographic techniques are essential for wide-field AO, enabling multi-object adaptive optics (MOAO).
Purpose of the Study:
- To present and evaluate different implementations of a tomographic reconstructor using the CARMEN machine learning architecture.
- To compare the performance of various neural network frameworks and CUDA code for AO reconstructors.
Main Methods:
- Introduction to adaptive optics concepts and control systems.
- Detailed explanation of the CARMEN reconstructor operation and its neural network frameworks.
- Development and testing of CUDA code for reconstructor implementation.
Main Results:
- The size of the reconstructor impacts neural network training and execution times.
- Native CUDA code demonstrated superior performance across all tested systems.
- Certain frameworks showed good performance under specific conditions.
Conclusions:
- Native CUDA implementation is the optimal choice for CARMEN-based tomographic reconstructors in AO systems.
- The study provides valuable insights into optimizing AO performance for large astronomical facilities.
