Spatial reconstruction of sound fields using local and data-driven functions.

Area of Science:

• Acoustics
• Signal Processing
• Computational Physics

Background:

• Traditional sound field analysis uses analytical basis functions, which can be suboptimal for complex sound fields at mid-to-high frequencies.
• Model discrepancy arises in classical methods when sound fields exhibit high spatial complexity, particularly in enclosed spaces.

Purpose of the Study:

• To investigate the efficacy of local sound field representations for improving reconstruction accuracy.
• To explore data-driven approaches for obtaining optimal local models for sound field analysis.

Main Methods:

• Employed local representations to reconstruct sound fields over large spatial apertures.
• Compared the performance of local models against conventional plane wave reconstructions.
• Utilized dictionary learning and principal component analysis on spatial measurements to derive data-driven local functions.

Main Results:

• Local partitioning models demonstrated superior conformity to sound fields characterized by high spatial complexity.
• Data-driven local functions, particularly those from dictionary learning, showed generalization capabilities across different rooms and frequencies.
• The study confirmed the potential of local and statistical properties for modeling complex acoustic phenomena.

Conclusions:

• Local representations are a viable alternative to analytical models for complex sound field reconstruction.
• Data-driven methods, like dictionary learning, provide robust and generalizable models for acoustic analysis.
• This approach holds promise for accurate sound field characterization in challenging acoustic environments.