Accurate and real-time acoustic holography using super-resolution and physics combined deep learning

Abstract

Acoustic holography is a promising technique for contactless manipulation, remote sensing, and energy harvesting. It involves retrieving holograms used to modulate acoustic sources for reconstructing target acoustic fields. The performance of reconstruction is primarily determined by two key criteria, including the spatial bandwidth product, which measures the pixel number representing information capacity, and the resolution, which quantifies the pixel size supporting detail gain. However, existing techniques face limitations in reconstructing high-fidelity, dynamic, and real-time acoustic fields with enhanced spatial bandwidth product and resolution across the entire aperture size. These challenges stem from the reliance on physically constrained holograms with static nature or relatively low spatial bandwidth product and resolution. Here, we introduce super-resolution acoustic holography, wherein the spatial bandwidth and resolution of the reconstructed target acoustic fields surpass those of the retrieved source holograms, especially within the same aperture size. We further develop a deep learning strategy that combines a classical neural network architecture with a linear accumulation based physical model, allowing for the customization of reconstructed acoustic planes with higher resolution while maintaining the same lateral coverages. Extensive algorithmic validations, numerical simulations, and practical experiments demonstrate the capability of our method to achieve high-fidelity, dynamic, real-time super-resolution acoustic holography, rendering its potential to advance practical applications in holographic acoustics.