Speaker
Description
Sound field reconstruction from sparse microphone measurements is an ill-posed problem highly sensitive to modeling choices. Approaches based on elementary wave expansions, such as plane wave expansions (PWE) or the equivalent source method (ESM), are popular and widely used, but they require domain discretizations prone to model mismatch. More recent physics-informed neural networks (PINNs) are more flexible but only enforce physical constraints weakly. In this study, we propose a sound field reconstruction approach that retains the physical consistency of the ESM while replacing fixed source parameters with learned parameters. We evaluate the method on both simulated and real measurements and observe improved reconstruction accuracy compared with standard PWE, ESM, and PINN baselines.