Speaker
Description
Most available engineering tools for outdoor acoustic simulation rely on ray-based methods, or sometimes on even simpler energy-based formulas. These simplifications have long been necessary because of the limited computational resources and time frames available to engineers. However, modern numerical methods and the increasing computational density of hardware are bringing fully 3D wave-based simulations within reach.In this presentation, I will summarize the research behind SoundSim360, a new simulation tool for large-scale outdoor noise propagation. I will focus on its computational engine, which is based on a high-performance implementation of summation-by-parts finite differences (SBP-FD). SBP-FD enables provably accurate and stable schemes while supporting complex boundary conditions and source models, and it benefits from the efficiency of finite differences on modern hardware.SoundSim360 approximates solutions to the second-order acoustic wave equation, directly computing pressure fluctuations in 3D space and time. It is implemented in custom CUDA C++ code in a matrix-free formulation, enabling highly efficient memory use. On a mid-range consumer-grade GPU, SoundSim360 can compute a noise map from a point source over a domain spanning 120 wavelengths in each direction in approximately 20 minutes.The presentation will give an overview of the numerical method and CUDA implementation. I will also show several application examples where SoundSim360 has been used, including wind turbine, traffic, and underwater simulations.