Speaker
Description
For computational room acoustic models, a suitable characterization of boundaries is required. Yet, boundary conditions are often not accurately known.One possibility is to infer boundary impedance models from in-situ measurements using a comparatively simple measurement setup. Since measurement noise and modelling errors cannot be avoided, we formulate the estimation problem in a Bayesian framework. By prescribing a suitable noise model, we infer statistical information on the impedance parameters of interest. This provides not only a single best-fit estimate, but also uncertainty information in the form of posterior expectations and variances.If several impedance parameters are unknown, the sampling procedure may require a large number of finite element solutions, which can make the overall computation expensive. To reduce this cost, we employ Krylov-subspace-based model order reduction techniques that allow for faster evaluations of the forward model. In this talk, we present numerical results for the inferred impedance statistics and compare the computational performance of the full-order and reduced-order models.