Deep learning-empowered moving cascaded acoustic holography for high-fidelity and high-capacity acoustic holographic reconstruction

Abstract

Acoustic holograms offer a promising platform for reconstructing complex acoustic fields with high resolution in three dimensions. However, traditional design algorithms based on single acoustic holograms have limitations in terms of their information capacity. To address this challenge, our study proposes a novel strategy called deep learning-empowered moving cascaded acoustic holography. This strategy enables high-capacity target patterns to be encoded onto two moving cascaded acoustic phase plates at a set of movable distances along the forward diffractive path of the incident acoustic wave. To achieve the high-fidelity and high-capacity acoustic holographic reconstruction, we develop a physics-enhanced moving cascaded acoustic hologram deep neural network (PhysNet_MCAH) method, which employs a physics-driven neural network to inversely design the phase offset distributions of two phase plates. Our results demonstrate that the PhysNet_MCAH method effectively generates high-fidelity and high-capacity moving cascaded acoustic holograms that are capable of rendering different complex target patterns at different design movable distances. Additionally, we show that moving cascaded acoustic holograms significantly improve the holographic storage capacity and enhance acoustic field reconstruction accuracy, surpassing the capabilities of moving single acoustic holograms. Remarkably, we further demonstrate that the proposed PhysNet_MCAH method achieves a superior performance in terms of the trade-off between reconstruction quality and holographic storage capacity when compared to existing unsupervised deep learning methods and multiplexing strategies for designing the multi-plane or cascaded holograms. Our method offers a new paradigm for high-capacity acoustic volumetric display, dynamic particle manipulation, and other applications.