Abstract
All coronagraphic instruments for exoplanet high-contrast imaging need wavefront correction systems to reject optical aberrations and create sufficiently dark holes. Since the most efficient wavefront correction algorithms (controllers and estimators) are usually model-based, the modeling accuracy of the system influences the ultimate wavefront correction performance. Currently, wavefront correction systems are typically approximated as linear systems using Fourier optics. However, the Fourier optics model is usually biased due to inaccuracies in the layout measurements, the imperfect diagnoses of inherent optical aberrations, and a lack of knowledge of the deformable mirrors (actuator gains and influence functions). Moreover, the telescope optical system varies over time because of instrument instabilities and environmental effects. In this paper, we present an expectation-maximization (E-M) approach for identifying and real-time adapting the linear telescope model from data. By iterating between the E-step (a Kalman filter and a Rauch smoother) and the M-step (analytical or gradient-based optimization), the algorithm is able to recover the system even if the model depends on the electric fields, which are unmeasurable hidden variables. Simulations and experiments in Princeton's High Contrast Imaging Lab demonstrate that this algorithm improves the model accuracy and increases the efficiency and speed of the wavefront correction.
Introduction
To image an Earth-like exoplanet, astronomers must suppress the blinding glare of its host star—a feat comparable to spotting a firefly next to a lighthouse from miles away. Instruments called coronagraphs achieve this by creating "dark holes" in the starlight’s point spread function. However, these dark holes are incredibly fragile. Small optical imperfections (aberrations) and thermal drifts create "speckles" of light that drown out the planet.
Current correction systems rely on model-based controllers, but the underlying physics models (Fourier optics) are often biased. Inaccuracies in actuator gains on deformable mirrors (DMs) and unmeasured "non-common-path errors" deep inside the instrument often lead to slow correction processes and performance floors. This paper addresses these challenges by introducing a data-driven framework to calibrate and adapt wavefront correction systems in real-time.
System Architecture and the E-M Framework
The Focal Plane Wavefront Control (FPWC) system is a closed-loop architecture that bridges optical hardware with computational estimation and control.
As illustrated in Figure 1, the optical system consists of a telescope, deformable mirrors, and a coronagraph, with the science camera providing intensity measurements ($I_k$). The control loop utilizes an Estimator to determine the state of the complex electric field and a Controller to issue commands ($u_k$) to the DMs.
The primary challenge is that the complex electric field ($E_k$) is a "hidden variable"; it cannot be measured directly, only its intensity. The authors reformulate FPWC as a stochastic optimization problem using an Expectation-Maximization (E-M) algorithm:
- The E-Step (Expectation): The algorithm uses a Kalman filter and a Rauch smoother to estimate the most likely state of the hidden electric fields based on noisy camera images.
- The M-Step (Maximization): Using these estimates, the algorithm updates the system's Jacobian matrix (the map of how DM movements affect the light) and noise covariance matrices. This can be done analytically or via stochastic gradient ascent.
This framework allows for "active learning," where the model is refined while the telescope is operational, effectively treating wavefront control as a reinforcement learning problem. Notably, the authors show that standard controller tuning parameters (like Tikhonov regularization) are mathematically equivalent to the magnitude of Jacobian uncertainties.
Experimental Validation and Results
The researchers validated their approach using numerical simulations and real-world testing at Princeton’s High Contrast Imaging Lab (HCIL). The HCIL testbed uses a Shaped Pupil coronagraph and two MEMS DMs with 952 actuators each.
Key Findings:
- Model Accuracy: The E-M algorithm successfully reduced Jacobian errors in simulation from approximately 9% to 2%.
- Speed and Contrast: In lab experiments, the identified model reached a contrast of better than $3 \times 10^{-7}$ in just 4 to 5 iterations, significantly faster than the biased physics-only model.
- Robustness: While analytical updates are precise in simulation, stochastic gradient ascent proved more robust in real-world experiments by avoiding overfitting to camera noise.
- Self-Correction: The adaptive framework successfully corrected for initially underestimated noise levels, allowing the Kalman filter to better balance model predictions against new measurements.
Limitations and Future Work
The current model assumes a linear relationship between DM commands and electric field changes. While this holds for small corrections (dark hole maintenance), it fails when large DM strokes are needed to correct major aberrations. Additionally, the current math assumes independent actuator errors (scalar regularization), whereas real-world actuators are often coupled.
Future work will focus on integrating Extended Kalman Filters (EKF) and Neural Networks to handle system nonlinearities and complex inter-actuator couplings. Such advancements are essential for developing "self-maintaining" telescopes for future space missions like HabEx or LUVOIR.
Summary updated