Machine learning approach to estimating optical-phase discontinuities using single-aperture irradiance patterns.

Area of Science:

• Optical physics
• Machine learning
• Wavefront sensing

Background:

• Optical-phase discontinuities cause aberrations in far-field irradiance patterns.
• Accurate estimation of these discontinuities is crucial for wavefront sensing applications.
• Traditional methods may struggle in dynamic environments with shock waves.

Purpose of the Study:

• To develop and validate a machine learning model for estimating optical-phase discontinuity (Δϕ) magnitude.
• To assess the model's performance with simulated and experimental data.
• To support the development of shock-wave-tolerant phase reconstruction algorithms.

Main Methods:

• An encoder-style convolutional neural network (CNN) was designed to process 32x32 normalized far-field irradiance patterns.
• The CNN was trained and validated using simulated data with varied Δϕ, discontinuity locations, and noise levels.
• Model robustness was tested by varying irradiance pattern spatial resolution and on experimental Shack-Hartmann wavefront sensor (SHWFS) data.

Main Results:

• The CNN accurately estimated Δϕ down to an irradiance pattern width of approximately 2 pixels when trained on varied resolutions.
• The model demonstrated successful application to experimentally collected SHWFS data.
• Accurate Δϕ estimation was achieved even with low-resolution irradiance patterns.

Conclusions:

• The proposed CNN model provides an effective method for estimating optical-phase discontinuities from far-field irradiance patterns.
• This approach is applicable to real-world experimental data from SHWFS.
• The research contributes to developing robust, shock-wave-tolerant wavefront sensing algorithms.