Evaluation of Techniques for Simulating Sound Propagation

Accurate simulation of outdoor sound propagation is essential for various applications in environmental acoustics, urban planning, and noise control. This thesis presents a numerical framework for solving the two-dimensional Linearized Euler Equations (LEE) under realistic meteorological conditions using a Finite Volume Method (FVM). The approach employs a semi-discrete formulation with high-order reconstructions, namely slope limiter and WENO schemes, to capture key wave phenomena such as diffraction, interference, and refraction due to temperature and wind gradients. An SSP-RK2 time integration method, combined with an appropriate choice of CFL parameters, provides robust performance over extended simulation times by maintaining numerical stability and reducing dispersive errors. To accommodate complex geometries, a rasterization strategy based on monochrome raster images is introduced, allowing obstacles to be mapped efficiently onto a structured Cartesian grid. Basic but effective absorbing boundary conditions are incorporated through sponge layers, mitigating reflections at domain limits. In addition, inclined or curved boundaries can be implemented to reduce stair-casing artifacts. The solver is further optimized for GPU platforms using NVIDIA’s CUDA, showing significant speedups over a serial CPU implementation. Validation and performance tests confirm that the proposed method accurately reproduces theoretical solutions and expected wave behavior. Detailed numerical experiments, including a realistic urban scenario, demonstrate its versatility and scalability. Finally, potential extensions - such as 3D reconstruction, integration of turbulent fields, multi-GPU parallelization with MPI, and advanced boundary modeling—are proposed to further enhance the solver’s accuracy and applicability to large-scale, real-world acoustic simulations.