Highly Stable Spatio-Temporal Prediction Network of Wavefront Sensor Slopes in Adaptive Optics

Area of Science:

• Optics
• Machine Learning
• Astronomy

Background:

• Adaptive Optics (AO) systems compensate for wavefront distortion but suffer from inherent delay errors.
• This delay limits correction performance, especially at high wavefront change frequencies exceeding the sensor's sampling rate.
• Traditional AO methods struggle with rapid wavefront fluctuations due to multi-frame delays.

Purpose of the Study:

• To develop a stable deep learning-based AO prediction network.
• To overcome the limitations of inherent delay errors in AO systems.
• To achieve high-precision wavefront prediction for improved correction accuracy.

Main Methods:

• A deep learning network was designed to predict future wavefront slopes.
• The network utilizes 10 frames of prior wavefront data to predict the next six frames.
• Simulations were conducted under various distortion intensities.

Main Results:

• The proposed network achieved high-stability and high-precision open-loop predicted slopes.
• Prediction accuracy for six frames decreased by no more than 15% across different distortion levels.
• Experimental results demonstrated superior open-loop correction accuracy at 500 Hz compared to traditional methods at 1000 Hz.

Conclusions:

• The deep learning AO prediction network offers a stable and accurate solution for wavefront distortion compensation.
• The method effectively mitigates the impact of system delays, enhancing AO performance.
• This approach enables better correction accuracy at lower sampling frequencies than conventional techniques.