Broadband acoustic absorbing metamaterial via deep learning approach

Sound absorption is important for room acoustics and remediation of noise. Acoustic metamaterials have recently emerged as one of the most promising platforms for sound absorption. However, the working bandwidth is severely limited because of the strong dispersion in the spectrum caused by local resonance. Utilizing the coupling effect among resonators can improve the absorbers' performance, but the requirement of collecting coupling effects among all resonators, not only the nearest-neighbor coupling, makes the system too complex to explore analytically. This Letter describes deep learning based acoustic metamaterials for achieving broadband sound absorption with no visible oscillation in a targeted frequency band. We numerically and experimentally achieve an average absorption coefficient larger than 97% within the ultra-broadband extending from 860 to 8000 Hz, proving the validity of the deep learning based acoustic metamaterials. The excellent ultra-broadband and near-perfect absorption performance allows the absorber for versatile applications in noise-control engineering and room acoustics. Our work also reveals the significance of modulating coupling effects among resonators, and the deep learning approach may blaze a trail in the design strategy of acoustic functional devices. Published under an exclusive license by AIP Publishing.

Automated summary

Sound absorption is crucial for room acoustics and noise control. Acoustic metamaterials, especially those based on deep learning, show promise in achieving broadband sound absorption. This letter presents a deep learning-based acoustic metamaterial approach that achieves ultra-broadband sound absorption without visible oscillation. Results from numerical simulations and experiments demonstrate the effectiveness and versatility of this approach.