Optimizing well-depth sequences for Schroeder diffusers using machine learning

Abstract

Around mid-1970, Schroeder introduced a mathematical theory to generate sequences, which can compute the “depth of the Schroeder diffuser”. A Schroeder diffuser is a design consisting of a number of wells of different “depths,” which propagates sound energy in different directions. The Schroeder diffuser provides the ability to create an optimal scattering of the same sound energy in different diffraction lobes. This paper aims to exploit this property and optimize well-depth sequences from Schroeder diffuser using machine learning techniques. With the advent of machine learning and parallel computing, the optimization of sequences and the iterative process to select the optimal pattern of sequence can be done more efficiently. Hence, the primary focus of this paper is to come up with a machine learning model that predicts the scattered reflections from a random well-depth sequence surface and evaluates the quality of the scattered reflections using a diffusion coefficient. The model tries to achieve optimum diffuser properties by adjusting the sequence. The boundary element method can be used for computing the diffusion coefficient. The diffusion coefficient parameter determines the quality of the diffuser and can be calculated at ⅓ Octave frequency bands.