Flexible continuous-time predictive adaptive optics with a lightweight liquid network for high-order Zernike...

Area of Science:

• Optical Engineering
• Computational Optics
• Machine Learning for Optics

Background:

• Adaptive optics (AO) systems face limitations due to latency, which exacerbates errors under dynamic turbulence.
• Existing AO correction methods struggle with rapidly changing atmospheric conditions.

Purpose of the Study:

• To develop a continuous-time predictive framework, LZCP-Net, for pre-driving deformable mirrors in AO systems.
• To enhance AO system performance by predicting future Zernike coefficients for improved turbulence compensation.

Main Methods:

• LZCP-Net integrates a 1-D temporal convolutional front-end with liquid time-constant (LTC) units.
• The framework utilizes Ordinary Differential Equation (ODE)-based dynamics for stable, adaptive prediction.
• Predictive capabilities are evaluated using simulations and experimental validation.

Main Results:

• LZCP-Net achieved up to 57% lower residual RMS compared to conventional AO under strong turbulence.
• The model outperformed LSTM and Kalman predictors, particularly for mid and high-order Zernike modes.
• Experimental results showed a 34.7% RMS reduction and 32.3% smaller EE80 radius using simulated data.

Conclusions:

• LZCP-Net offers a lightweight and robust solution for high-precision, low-latency AO correction.
• The predictive framework demonstrates strong generalization across different wavefront sensor configurations.
• LZCP-Net shows significant potential for real-world applications in adaptive optics.