Deep-learning-aided metasurface design for megapixel acoustic hologram

Abstract

Unlike the holography technique using active sound source arrays, metasurface-based holography can avoid cumbersome circuitry and only needs a single transducer. However, a large number of individually designed elements with unique amplitude and phase modulation capabilities are often required to obtain a high-quality holographic image, which is a non-trivial task. In this paper, the deep-learning-aided inverse design of an acoustic metasurface-based hologram with millions of elements to reconstruct megapixel pictures is reported. To improve the imaging quality, an iterative compensation algorithm is proposed to remove the interference fringes and unclear details of the images. A megapixel image of Mona Lisa's portrait is reconstructed by a 2000 × 2000 metasurface-based hologram. Finally, the design is experimentally validated by a metasurface consisting 30 × 30 three-dimensional printed elements that can reproduce the eye part of Mona Lisa's portrait. It is shown that the sparse arrangement of the elements can produce high-quality images even when the metasurface has fewer elements than the targeted image pixels.