Sub-Millisecond Phase Retrieval for Phase-Diversity Wavefront Sensor.

Area of Science:

• Optical Engineering
• Artificial Intelligence
• Image Processing

Background:

• Traditional phase-diversity wavefront sensing algorithms are computationally intensive, limiting real-time applications.
• Phase-diversity techniques require capturing multiple images (e.g., in-focus and out-of-focus) to reconstruct wavefront aberrations.

Discussion:

• The proposed Phase Diversity Convolutional Neural Network (PD-CNN) offers a significant speed improvement over conventional methods.
• PD-CNN fuses information from focal and defocused intensity images for efficient wavefront sensing.
• This light-weight model avoids complex iterative transformations and optimization inherent in traditional algorithms.

Key Insights:

• PD-CNN achieves state-of-the-art inference speeds of approximately 0.5 ms.
• The method demonstrates high accuracy and speed, validated through experimental results.
• It presents a viable, accelerated alternative for phase-diversity wavefront sensing.

Outlook:

• Potential for integration into adaptive optics systems and real-time optical metrology.
• Further research could explore variations of CNN architectures for enhanced performance.
• The approach may be applicable to other image-based sensing and reconstruction problems.