Wide-angle and high-efficiency acoustic retroreflectors enabled by many-objective optimization algorithm and deep learning models

By redirecting the incident wave energy back toward the source or sensor, retroreflectors mitigate signal losses, render a higher signal-to-noise transmission with improved sensing performance, and are highly desirable in acoustic applications such as remote sensing and detection, medical ultrasound imaging, and underwater communications. In this paper, we propose a different and efficient approach to realize a wide-angle high-efficiency retroreflector that is based on the design frame of acoustic metagrating. By integrating a many-objective optimization algorithm with a deep learning neural network, we construct a comprehensive and efficient optimization framework to intelligently design the metagrating so that its seven different diffraction orders work synergistically to realize the intended high-efficiency retroreflection functionality over a broad and continuous range of incident angles. Compared with the existing approaches based on corner cubes, Luneburg lenses, or dual-layer metasurfaces, the single-layer configuration of the retroreflector has a relatively simple unit-cell structure and can be easily extended to higher working frequencies. Both the proposed design paradigm itself and the high-performance retroreflector configuration may find applications in various scenarios including ultrasonic detection and imaging, music performance monitoring, and underwater communications.