CNN-based bit error rate reduction in holographic data storage systems using simulated training datasets
Year: 2026
Type: article
Abstract: In a holographic data storage system (HDSS) experimental setup, several noise sources can increase the bit error rate (BER) when we recover the data from the reconstructed data page. While deep learning algorithms, particularly convolutional neural networks (CNNs), have emerged as powerful tools for data recovery and phase retrieval, training such networks requires a large volume of paired ground-truth and degraded data pages. In the early development phases of an HDSS setup, experimentally collecting thousands of data pages is labor-intensive and impractical due to manual alignment constraints, sample translation limitations, and material degradation over repetitive reads. To address this fundamental bottleneck, we present a novel, fast physical-mimicking synthetic degradation model based on digital image processing techniques to synthesize thousands of paired training images in seconds. The proposed framework models contrast dispersion via randomized Gaussian distributions, spatial defocusing via low-pass filtering, localized spatial intensity non-uniformity due to material thickness variations, and geometric alignment errors (translation and rotational tilt). A simplified U-Net architecture without normalization layers and with reduced channel width is trained exclusively on synthetically degraded data pages and evaluated on both synthetic tests and real experimentally reconstructed data pages. The experimental setup incorporates a parallel-aligned liquid crystal on silicon (PA-LCoS, HOLOEYE PLUTO) spatial light modulator and self-made polyvinyl alcohol/acrylamide (PVA/AA) photopolymer media under an angular multiplexing scheme. The results demonstrate a systematic reduction of the raw BER down to 1.5 × 10 –2 , in our case 60 errors in a data page with 4096 bits of information, significantly outperforming conventional global and adaptive local thresholding methods. Furthermore, the CNN output establishes a broad, stable operational plateau centered around the ideal decision threshold, ensuring robust and reliable data retrieval during early-stage HDSS prototyping.
Source: Optics and Lasers in Engineering
Institutions: University of Alicante, Tsinghua University
Language: English
Subfield: Atomic and Molecular Physics, and Optics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,020
OpenAlex ID: W7168358834