Time-varying constraint optimization for sensorless adaptive optics systems driven by deep PSF-Zernike estimation

Area of Science:

• Optical engineering
• Computational imaging
• Control systems

Background:

• Sensorless adaptive optics (SAO) corrects wavefront aberrations without a dedicated sensor.
• Closed-loop SAO faces challenges inferring aberrations from noisy point-spread functions (PSFs) due to latency and deformable mirror (DM) constraints.
• Time-varying turbulence exacerbates these issues, leading to outdated correction commands.

Purpose of the Study:

• To develop a robust closed-loop sensorless adaptive optics framework.
• To address challenges posed by system latency and DM constraints in dynamic environments.
• To improve the stability and accuracy of wavefront correction in SAO systems.

Main Methods:

• A hybrid Convolutional Neural Network (CNN)-Transformer model estimates 28 residual Zernike coefficients from single-frame PSFs.
• An autoregressive predictor with online parameter adaptation and fractional-step forecasting aligns estimates with the effective actuation instant.
• A constrained model predictive controller (MPC) incorporates delay-aligned residuals and enforces DM amplitude and slew-rate limits.

Main Results:

• The learned estimator provides informative single-frame modal estimates from noisy PSFs.
• The proposed Neural Network (NN)+Autoregressive (AR)+MPC scheme enhances loop stability and reduces residual phase Root Mean Square (RMS).
• The system maintains a higher Strehl ratio compared to direct control and non-predictive MPC, especially in strong turbulence.

Conclusions:

• Temporal misalignment is a primary limitation in high-turbulence regimes for SAO.
• Delay-aware constrained optimization is crucial for robust, high-speed sensorless adaptive optics.
• The developed framework offers a practical solution for improving SAO performance in dynamic conditions.