Angle-dependent Phononic Dynamics for Deep Learning and Source Localization

Abstract

In this work, a parameterized eigenvalue problem is analyzed for a phononic array in a 2D stress wave scattering setup, and a corresponding sensing application of this system is proposed to achieve source angle localization. The phononic domain consists of a periodic micro-structured medium, of which the eigen-wavevector band structure and the eigen-modes are exploited. The eigen-modes are naturally angle dependent due to changes in phases and periodic mode shapes determined by the incident angle. Intriguingly, the band exhibits angle-dependent transitions at the exceptional points (EPs) and critical angles (CAs), where the eigenvalues coincide or vanish. Coupled with these transitions, it is found that the eigen-modes switch their energy characteristics and symmetry patterns at these branch points, leading to enhanced angle dependence. Moreover, these eigen-modes also serve as the basis functions of the scattered waves. Therefore, the scattering response of the medium inherently possesses the angle-dependent properties, making this system naturally suitable for sensing applications. An artificial neural network (ANN) is trained with randomly weighted eigen-modes to achieve deep learning of the eigen features and angle dependence. The training data is derived only based on the eigen-modes of the unit cells. Nevertheless, the trained ANN can accurately identify the incident angle of an unknown scattering signal, with minimal side lobe levels and suppressed main lobe width. The ANN shows superior performance in comparison with standard delay-and-sum technique of estimating angle of arrival. The proposed application of ANN and micro-structured media highlights the physical importance of band structure topology and eigen-modes to a technological application, adds extra strength to the existing localization methods, and can be easily enhanced with the fast-growing data-driven techniques.