Model-based reinforcement learning and inverse problems in extreme adaptive optics control
Nousiainen, Jalo (2023-05-26)
Väitöskirja
Nousiainen, Jalo
26.05.2023
Lappeenranta-Lahti University of Technology LUT
Acta Universitatis Lappeenrantaensis
School of Engineering Science
School of Engineering Science, Laskennallinen tekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:ISBN:978-952-335-940-6
https://urn.fi/URN:ISBN:978-952-335-940-6
Tiivistelmä
The field of exoplanet research is one of the most rapidly expanding research fields in modern astrophysics. In recent decades, astronomers have found most exoplanets via indirect techniques such as the transit and radial velocity method. The direct imaging technique called high contrast imaging (HCI) enables new ways to expand our knowledge of these exoplanets and exoplanetary systems. However, direct imaging of exoplanets is challenging due to the high contrast ratio and small angular separation from the host star. Thus, HCI detections, so far, are mostly limited to a few dozen young and luminous giant exoplanets.
The new generation of HCI instruments, under development, will push direct imaging into increasingly challenging areas, discovering and characterizing exoplanets dimmer and closer to their host start. The ultimate goal is direct imaging and characterization of potentially habitable exoplanets. On ground-based telescopes, HCI instruments are equipped with eXtreme Adaptive Optics (XAO) that correct the phase fluctuations caused by the atmosphere. With an optimized instrument design, the residuals left by XAO correction set the limitation of sensitivity; thus, minimizing the XAO residuals is a crucial objective for ground-based HCI. Further, most habitable exoplanets are located at small angular separations from their host stars, where current XAO control algorithms leave strong residuals of stellar light that could be suppressed with more advanced algorithms. This thesis explores novel data-driven control methods for XAO control that cope with crucial limitations of traditional control laws, such as temporal delay and calibration errors. Improvement in these potentially reduces the residual flux of stellar light in the coronagraphic point spread function and thus enables fainter observations closer to the host star.
We show that model-based RL is a promising XAO control approach that produces consistent results in numeric simulations and lab setups. The proposed methods suppress the temporal error, and photon noise compensates for misregistration and optical gain. It can also adapt to changing wind conditions in time scales of several seconds. Moreover, model-based RL manages the extreme time constraint of XAO control and, if well formulated, scales to ELT scale XAO.
The new generation of HCI instruments, under development, will push direct imaging into increasingly challenging areas, discovering and characterizing exoplanets dimmer and closer to their host start. The ultimate goal is direct imaging and characterization of potentially habitable exoplanets. On ground-based telescopes, HCI instruments are equipped with eXtreme Adaptive Optics (XAO) that correct the phase fluctuations caused by the atmosphere. With an optimized instrument design, the residuals left by XAO correction set the limitation of sensitivity; thus, minimizing the XAO residuals is a crucial objective for ground-based HCI. Further, most habitable exoplanets are located at small angular separations from their host stars, where current XAO control algorithms leave strong residuals of stellar light that could be suppressed with more advanced algorithms. This thesis explores novel data-driven control methods for XAO control that cope with crucial limitations of traditional control laws, such as temporal delay and calibration errors. Improvement in these potentially reduces the residual flux of stellar light in the coronagraphic point spread function and thus enables fainter observations closer to the host star.
We show that model-based RL is a promising XAO control approach that produces consistent results in numeric simulations and lab setups. The proposed methods suppress the temporal error, and photon noise compensates for misregistration and optical gain. It can also adapt to changing wind conditions in time scales of several seconds. Moreover, model-based RL manages the extreme time constraint of XAO control and, if well formulated, scales to ELT scale XAO.
Kokoelmat
- Väitöskirjat [1038]