Overview / Usage

The biggest challenge faced in directly imaging exoplanets is the huge contrast in brightness between the host star and the exoplanet. Current state-of-the-art exoplanet imaging systems on ground based telescopes, make use of adaptive optics (AO) subsystems with linear controllers, that aim to flatten the wavefront of the light collected by the telescope, and coronagraphs - optical components used to create dark regions in the image, where exoplanets can be imaged.

However, there are still issues that need to be solved, because the coronagraph and adaptive optics systems are not perfect - some aberrations are invisible to the wavefront sensor, some aberrations require a much faster deformable mirror, or several more modes than the DM can correct for.

This research combines the functionality of the AO subsystem and the coronagraph by implementing a non-linear controller to exclusively optimize the contrast, achieving an "adaptive coronagraph" in the process. By doing so, we can fine tune the imaging system to produce a designated dark region in the image plane, in which the exoplanet will stand out and can easily be imaged. This method of active control of the AO subsystem will also provide resilience to wavefront aberrations, since the goal is not to flatten the wavefront, but exclusively to optimize for the contrast at the dark region in the image plane.

Methodology / Approach

The goal of this research is to design an AO controller that attempts to minimize diffracted starlight in a small, designated region at the image plane of the exoplanet imaging instrument. This is fundamentally a non-linear problem, and therefore requires more complex control strategies.

The first step in tackling this problem is developing a relationship between the various wavefront aberrations that occur in the optical path, the actions of the deformable mirror of the AO subsystem, and the intensity of starlight in the designated region of the image plane.

Next, I will investigate different methods of non-linear control of the AO subsystem to optimize the contrast at the image plane. I am currently implementing a neural network based controller, treating the control strategy as a reinforcement learning problem. This involves experimenting with different neural network architectures to combine information from the wavefront sensor and imaging camera, and use this information to drive the deformable mirror.

I will run simulations of these experiments using open source libraries for optical testbenches and neural networks. Following this, I will design an optical test setup in the lab and run the experiments here. Finally, I will build a prototype instrument and test this on sky, at a telescope.

Technologies Used

Open source Python libraries for optical testbench simulation. Keras and TensorFlow for neural networks. Various optical components, including deformable mirrors, spatial light modulators, and mirrors and lenses for laboratory verification.

Repository

https://github.com/VikramRadhakrishnan/DeepContrast