Content area
Full Text
Received Nov 25, 2017; Accepted Feb 21, 2018
This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1. Introduction
Development of large telescopes is one of the biggest challenges in nowadays astronomy and astrophysics. Future construction of the Thirty-Meter Telescope (TMT) [1] and the European Extremely Large Telescope (E-ELT) [2], the two largest telescopes in the world, have originated enormous challenges for engineers and researchers [3]. One of the key elements of these telescopes is the development of the adaptive optics (AO) [4] system that helps to improve the quality of the received image.
There are several tomographic techniques employed in the image reconstruction for AO systems, like Single Conjugate Adaptive Optics (SCAO), Multiconjugate Adaptive Optics (MCAO), or Multiobject Adaptive Optics (MOAO) [5] to be used in the future E-ELT [3]. MOAO uses several reference guide stars to obtain information to reconstruct the atmosphere turbulence profile [6]. To combine this information, it is necessary to use tomographic reconstruction algorithms. Some of the most popular are based on a matrix vector multiplication, with the control matrix being defined by either least squares (LS) [7, 8] or minimum variance techniques [9]. However, during recent years most complex solutions have been developed, like Learn and Apply (L & A) [10], or the intelligent system known as Complex Atmospheric Reconstructor based on Machine Learning (CARMEN) [11, 12]. Due to the increasing complexity and amount of data used by these algorithms [13], some of previous algorithms have been implemented in Graphics Processing Units (GPUs) [14, 15], speeding up substantially their execution and development [16, 17].
CARMEN is a tomographic reconstructor for MOAO systems created at the University of Oviedo. It was initially developed using nonparametric estimation techniques [18] and Multivariate Adaptive Regression Splines (MARS) [19], but its development using Artificial Neural Networks (ANN) achieves great results at on-sky testing [20, 21]. ANN and deep learning have become very popular in recent years [22], and several frameworks have been developed to help researchers in their projects [23]. Most of these frameworks provide GPU acceleration, and some of them have shown good results speeding up CARMEN training and execution [24–26]. However,...