In practical applications of holographic data storage (HDS), which offers both high capacity and high transfer rates, reduction of the error rate is a major issue. In this study, we have applied a convolutional neural network (CNN)-based demodulation scheme for constant-weight codes to HDS and have demonstrated that the error rate can be reduced by approximately one order of magnitude when compared with that of the conventional ranking method. We have also evaluated the HDS error characteristics, which were not clarified fully in previous studies. The results of this evaluation showed that symbols with certain specific shapes are prone to errors. This study is expected to provide useful knowledge for use in practical applications of HDS.

在提供高容量和高传输率的全息数据存储（HDS）的实际应用中，降低错误率是一个主要问题。在这项研究中，我们将基于卷积神经网络 (CNN) 的恒权码解调方案应用于 HDS，并证明与传统排序方法相比，错误率可以降低大约一个数量级. 我们还评估了 HDS 错误特征，这些特征在以前的研究中没有完全阐明。该评估的结果表明，具有某些特定形状的符号容易出错。该研究有望为HDS的实际应用提供有用的知识。