Realization of Linear Wave-Propagation Models from HPC Simulations

Modeling of sound propagation in complex environments requires high performance computing (HPC) to simulate three-dimensional wave fields with realistic fidelity. This is especially true for urban areas, where sound waves reflect and diffract due to the built-up infrastructure. HPC can predict these wave fields with desired fidelity, but the computational investment would have greater return if reduced-size models that operate with considerably less computational resources could be produced from the results. The objective of this work is to develop such models. The work applies a modified version of the Eigensystem Realization Algorithm, using Markov parameters from HPC input-output response functions, to generate state-space models that simulate hundreds of thousands of output signals of the HPC wave field. The results include predicted acoustic signals and signatures from realized models, using a source with a different time series than the source used to generate the Markov parameters. We compare wave-field signals from reduced-order models with HPC model signals over a large urban domain, adjusting the model order and accuracy by singular-value cutoff. We conclude that the method produces efficient high-fidelity models of sound propagation in complex environments.