Fabien Baron, Georgia State University; Douglas Hope, Georgia Tech Research Institute; Stuart Jefferies, Georgia State University; Daniel Johns, Georgia Sate University
Keywords: Imaging, Machine Learning, Wave front sensing
Abstract:
Detection, Imaging, and characterizing resident space objects require unprecedented high-resolution and high-contrast Imaging. While large aperture telescopes can offer improved sensitivity, mitigating the effects of turbulence on resolution brings another set of challenges. On the other hand, while providing extremely high resolution, optical interferometers do not offer the necessary sensitivity.
A new alternative to conventional interferometry and large full aperture telescopes (>10m) is the novel hybrid optical telescope (HOT) design, which consists of phase-distributed aperture arrays. The HOT architecture is a Fizeau-type telescope with an interferometric setup using apertures constructed from lightweight optics on a ring configuration. The interferometric design of HOT allows one to leverage PSF engineering techniques that can locally create contrast levels up to 1e-7 in the image.
While the hybrid optical telescope offers a new path toward high-resolution and high-contrast Imaging, a significant challenge will be achieving and maintaining wavefront coherence on the distributed apertures to enable successful beam combination at the prime focus.
The support structure for the HOT relies on tensegrity to support the weight of the distributed apertures. So, as the telescope slews to track objects, actuators on the support structure will have to adjust the position of the mirrors to maintain the coherence of the wavefront at the prime focus. Because of the nature of tensegrity, achieving and maintaining coherence will be a non-linear problem. This non-linearity will pose a problem for current wavefront sensing and adaptive optics system that rely on the linearity of the problem to rapidly close the phase loop at the coherence time of the atmosphere.
To overcome this limitation, we are developing a non-linear WFS based on machine learning. While Machine learning (ML) algorithms, in principle, can find solutions to non-linear problems based on vast amounts of training data, they can face challenges if training data is limited or does not fully represent physical reality. To overcome this limitation, we will focus on physics-guided ML (PGML) algorithms. The layered Taylor frozen flow model of turbulence represents one of the numerous physical constraints, including seeing conditions, scintillation effects, etc., we will use to train a PGML algorithm for wavefront sensing and eventual adaptive optics correction as the HOT aperture slews and locks on a target.
To ensure optimal operation of the PGML algorithm, we will use the GSU ARES benchtop atmospheric turbulence simulator to generate limited yet physically realistic training datasets. We will demonstrate the performance of our PGML, which will be evaluated based on wavefront correction level and how it might enable quantum super-resolution techniques based on wavefront projection methods.
This research is supported by AFOSR award FA9550-23-1-0536
Date of Conference: September 17-20, 2024
Track: SDA Systems & Instrumentation