Distance-adaptive unsupervised CNN model for computer-generated holography

Convolutional neural networks (CNN) are useful for overcoming the trade-off between generation speed and accuracy in the process of synthesizing computer-generated holograms (CGH). However, methods using a CNN cannot specify the propagation distance when reproducing a hologram, thereby limiting their practical usage across various contexts. Therefore, in this study, we developed a CNN model that can generate CGH by specifying the target image and propagation distance. The proposed method demonstrates performance comparable to traditional methods with a fixed distance and achieves the generation accuracy and speed necessary for practical use even when the propagation distance is changed.