Metagrating-based acoustic wavelength division multiplexing enabled by deterministic and probabilistic deep learning models

Wavelength division multiplexing (WDM), the technology that utilizes different wavelengths of carrier signal as independent information channels and helps to realize enhanced information density and functional diversity, is desired in various acoustic applications for the growing demand of higher transmission efficiency and lower cost. In this paper, we marry the rising concept of acoustic metagrating with the technology of WDM and demonstrate that interesting and distinct wavelength-dependent functionalities with unitary efficiency in prespecified frequency ranges can be realized by a simple grating structure. To this end, we integrate the particle swarm optimization with the deep learning technique to resolve the uncertainty issue and multi-objective problem and develop both deterministic and probabilistic neural networks for the inverse design of a metagrating-based WDM device. As a result, the computational cost is greatly reduced, and the probabilistic model reveals the sensitivity and flexibility of the parameters of the metagrating with respect to the desired functionality. In this paper, we not only provide an intelligent inverse design paradigm of high-performance WDM devices for multiple manipulation purposes but also present a feasible solution for the development of integrated acoustic devices for wavelength-dependent applications.

Automated summary

This paper proposes the integration of acoustic metagratings with wavelength division multiplexing (WDM) technology to achieve wavelength-dependent functionalities. The study combines deep learning and particle swarm optimization to develop deterministic and probabilistic neural networks for inverse design. The results show reduced computational cost and reveal the flexibility and sensitivity of metagrating parameters.