Reinforcement Learning Algorithm and FDTD-Based Simulation Applied to Schroeder Diffuser Design Optimization

This paper aims to propose a novel approach the algorithmic design of Schroeder acoustic diffusers by employing deep learning optimization algorithm and fitness function, which are based on computer simulation propagation waves. The method employed for research consists policy gradient algorithm. It is used as tool carrying out sequential process, seeks maximize function parameters characterizing autocorrelation diffusion coefficient designed diffuser. As coefficients calculated polar response diffuser, finite-difference time-domain (FDTD) obtain set impulse responses, necessary calculate responses optimized diffusers. results obtained from derived were compared with outcomes similar genetic random selection diffuser well-depth pattern. We found that best result was achieved gradient, it produced that, in terms provided coefficient, statistically better than properties designs supplied two other baseline approaches.