Adaptive Optics (AO) aims at compensating the quickly varying aberrations induced by the earth's atmosphere, and restoring images at the diffraction limit of current large ground based telescopes. Most current AO systems, however, require a bright and/or close reference natural source. A new generation of AO systems, a.k.a. Wide Field AO (WFAO), is addressing this limitation by significantly increasing the field of view of the AO-corrected images, and/or the fraction of the sky that can benefit from such correction.

In the era of giant telescopes it becomes harder and harder to have a comprehensive view of the performance of the AO system and its impact on the final performance of the instruments. Classical laws, models and rules no longer apply to the new generation of ground-based instrumentation. Indeed, on top of well-known phenomena (classical turbulence effects), the dramatic increase of the telescope size (up to 39m diameter) induces new deleterious effects that have not yet been fully characterized nor modeled (due to telescope spiders, segmentations, windshake residual, cophasing errors etc …). Moreover, the AO system have to use several deformable mirrors in cascade, many laser guide stars and many (often non linear) sensors, inducing a complexification of the control strategies and of the post-processing of the scientific data.

For this latter aspect, the knowledge of the instrument response (a.k.a. PSF for Point Spread Function) is fundamental. Most of the key science cases in today’s, and tomorrow’s astronomy call for extremely accurate astrometry (below the tenth of milli-arcsecond) and photometry (better than 0.1%) which are only achievable with a complete knowledge of the spatial, spectral and temporal aspects of the PSF.

The primary goal of the thesis will be to develop new concepts and methods to break the current barrier of the system and atmospheric parameter estimation using AO telemetry combined with exogenous information provided by the telescope and the instrument auxiliary sensors.
The use of Machine or Deep Learning strategies, as well as data fusion, will be considered. A strong collaboration with data processing experts has been initiated to that end. Furthermore access to dedicated Hardware such as powerful GPUs should be possible.
More importantly, the PhD student will have access to several years of data recorded on the VLT AOF+MUSE instrument, as a benchmark for testing, demonstrating and validating the proposed approaches. These real field datasets will allow exploring innovative methods based on machine learning algorithms (e.g. PyTorch, Keras), and compare their performance with classical methods like PCA and simple correlations (e.g. scikit-learn).

The thesis will be done in collaboration with the Laboratoire d'Astrophysique de Marseille (B. Neichel, T. Fusco)