Speaker
Description
Reconstructing room-acoustic wavefields from sparse measurements remains challenging in the presence of diffraction and scattering, where multiple propagation components overlap in space–time. Classical representations, such as plane waves, provide global field expansions but require significant coefficients to capture localized interactions and destructive interference. Multi-scale approaches, such as wavelets and shearlets, improve spatial adaptivity but are not explicitly aligned with the physics of wave propagation. In this work, we propose a boostlet-based framework for space–time superresolution of acoustic fields (i.e., recovering missing microphone responses) that captures propagating wavefronts as sparse, local structures in joint space–time coordinates. The reconstruction is formulated as an ℓ1-regularized inverse problem and evaluated under various undersampling rates (half and 1/3 of the data) and severe noise levels (down to 5 dB signal-to-noise ratios). A room-impulse response dataset is used to assess the method against baselines. Our results highlight two main observations. First, the proposed approach improves the overall reconstruction of propagating structures compared to the baselines, particularly in regimes with multiple overlapping wave components. Second, the performance of the proposed approach is governed by a spectral filter that penalizes boostlet coefficients in a frequency-dependent manner: a higher/lower cut-off frequency in the early/late parts. The results demonstrate the potential of boostlet representations for sparse sound field reconstruction and motivate further investigation and extensions to higher spatial dimensions.