Holographic Reconstruction with All-Acoustic Diffractive Network

Acoustic hologram has shown its great potential for controlling complex pressure fields with minimal hardware, making it attractive for various applications in particle manipulation, cellular assembly, and ultrasound therapy, etc. However, conventional inline reconstruction methods suffer from twin-image artifacts. Computer-based iterative retrieval or deep learning methods have been used to eliminate artifacts in acoustic holograms. Nevertheless, once a hologram is calculated by these methods and fabricated, it can only generate a pressure field with a particular pattern at a particular position. Here, we report an all-acoustic diffractive neural network (ADNN-AH) and perform a proof-of-principle numerical demonstration of high-quality pressure field reconstruction from a corresponding inline hologram. The proposed ADNN-AH comprises a series of spatially-engineered physical transmissive diffractive layers/surfaces. When the transmissive layers are designed using a deep learning strategy, they process the diffracted sound wave from an arbitrary input hologram through acoustic-matter interactions to reconstruct a high-resolution pressure field with the suppression of twin-image artifacts. A series of numerical experiments demonstrate that as an all-acoustic hologram reconstruction method, ADNN-AH has great potential in various ultrafast acoustic hologram-based applications that benefit from high-quality reconstruction and application flexibility.