SNR enhancement for low-SNR amplitude-modulated holographic data storage based on deep learning

Area of Science:

• Optical Engineering
• Image Processing
• Machine Learning

Background:

• Holographic data storage faces challenges with random noise in amplitude data pages, reducing image signal-to-noise ratio (SNR).
• Traditional U-Net models struggle to effectively enhance SNR, especially in low-SNR holographic data pages.

Purpose of the Study:

• To propose an improved U-Net architecture, DRAMCU-Net, for effective SNR enhancement in holographic data storage images.
• To address the limitations of traditional U-Net in handling noise and improving image quality for holographic data retrieval.

Main Methods:

• Developed DRAMCU-Net (dilated residual attention and multi-scale convolution U-Net) incorporating dilated residual attention blocks (DRABs) and multi-scale convolutional blocks (MS-Conv).
• Implemented a dropout layer to improve the global robustness of the model.
• Employed multi-level feature extraction and attention mechanisms for efficient feature capture at various scales.

Main Results:

• DRAMCU-Net demonstrated superior performance in enhancing SNR for low-SNR holographic data pages compared to the traditional U-Net.
• Experimental results showed an approximate 30% increase in SNR for reconstructed data pages using DRAMCU-Net.

Conclusions:

• DRAMCU-Net effectively enhances SNR in holographic data storage images, overcoming limitations of conventional U-Net models.
• The proposed architecture offers a promising solution for improving the reliability and quality of holographic data retrieval.