Real-time phase-retrieval and wavefront sensing enabled by an artificial neural network

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning Applications

Background:

• Wavefront reconstruction is crucial for optical system alignment and aberration correction.
• Traditional iterative phase retrieval methods are time-consuming, limiting real-time applications.
• Diffraction pattern analysis is a key technique in optical metrology.

Purpose of the Study:

• To develop a real-time wavefront reconstruction method using neural networks.
• To overcome the speed limitations of iterative phase retrieval techniques.
• To enable live adjustments and adaptive optics correction in complex optical systems.

Main Methods:

• A neural network was trained using simulated diffraction pattern data.
• The trained neural network was experimentally validated.
• Real-time wavefront reconstruction was achieved by processing diffraction patterns from a diffracting mask.

Main Results:

• The neural network achieved live wavefront reconstructions in milliseconds, a significant speed improvement over iterative methods (seconds).
• The method demonstrated superior performance compared to iterative phase retrieval, especially with high-noise diffraction patterns.
• Real-time capability allows for dynamic adjustments and corrections in optical setups.

Conclusions:

• Neural network-based wavefront reconstruction offers a fast and robust solution for optical metrology.
• This technique facilitates real-time control and optimization of complex optical systems.
• The method shows promise for applications requiring rapid optical feedback and correction.