Deep learning-based phase retrieval with embedded data for holographic data storage.

Area of Science:

• Digital holography
• Computational imaging
• Deep learning applications

Background:

• Phase retrieval is crucial for reconstructing 3D information from holograms.
• Traditional methods often require large datasets and significant recording areas.
• Multi-level phase encoding presents challenges in data arrangement and unique solution identification.

Purpose of the Study:

• To develop a deep learning-based (DL-based) phase retrieval method utilizing embedded data as constraints.
• To simplify the coding strategy for multi-level phase encoding.
• To enhance phase retrieval accuracy and reduce data requirements, especially in limited recording areas.

Main Methods:

• A deep learning framework is proposed for phase retrieval incorporating known embedded data.
• An innovative coding strategy reduces signal data arrangements from 4^5 to 4 types.
• The embedded data acts as a robust constraint on diffraction characteristics between adjacent data.

Main Results:

• The method achieves precise phase retrieval even with a reduced hologram recording area (Nyquist size).
• Required training samples were reduced by a factor of 50 compared to traditional approaches.
• The bit error rate (BER) was decreased by approximately 9 times.

Conclusions:

• The DL-based phase retrieval with embedded data offers a significant advancement in holographic data processing.
• The proposed method demonstrates superior performance in terms of data efficiency and accuracy.
• This technique is particularly beneficial for applications with constraints on recording area and computational resources.