Physics-informed localization of sound sources in reflective environments including hidden sources

Sound source localization in the presence of rigid boundaries remains challenging because reflections and diffractions substantially distort the measured signals and are difficult to account for with conventional localization techniques. To address this issue, this paper extends the adjoint-based method introduced in Sroka, Sarradj, and Lemke [(2025). J. Acoust. Soc. Am. 158, 419-432] by incorporating a volume penalization approach to represent rigid boundaries. The resulting formulation enables the adjoint framework to account for wave reflections at rigid surfaces while maintaining high numerical accuracy. For suitable penalization parameters, the boundary representation exhibits an observed error convergence rate of approximately O(3). The method is demonstrated in two scenarios: first, a fully hidden sound source is identified in both space and time without additional assumptions; second, experimental microphone measurements of a moving vehicle are analyzed, yielding multiple physically plausible source locations and highlighting the strong influence of the rigid ground surface on localization results.