Adaptive sampling for optimized sensor placement in sound field reconstruction.

Area of Science:

• Acoustics and Signal Processing
• Computational Physics
• Machine Learning

Background:

• Sound field reconstruction aims to create a continuous acoustic map from discrete measurements.
• Traditional sensor placement methods are often non-adaptive, suitable for static sound fields but inefficient for dynamic ones.

Purpose of the Study:

• To develop and evaluate an adaptive sampling (AS) strategy for efficient sensor placement in sound field reconstruction.
• To improve sensor efficiency, particularly for nonstationary acoustic environments.

Main Methods:

• Analysis of non-adaptive sampling criteria within a Bayesian/Gaussian process framework.
• Proposal of an adaptive sampling (AS) strategy combining leave-one-out cross-validation for exploitation and wavelength-based spacing for exploration.
• Simulation-based comparison of AS with non-adaptive methods on stationary and nonstationary sound fields.

Main Results:

• Adaptive sampling (AS) matches non-adaptive methods on stationary fields.
• AS demonstrates significantly improved efficiency on nonstationary fields, using approximately half the number of sensors for equivalent accuracy.
• The AS strategy effectively balances targeted data acquisition (exploitation) with broad spatial coverage (exploration).

Conclusions:

• Adaptive sampling (AS) offers a substantial improvement in sensor placement efficiency for sound field reconstruction, especially in dynamic scenarios.
• AS provides a practical and efficient solution for sequential measurement workflows in acoustics.
• The findings suggest a paradigm shift towards adaptive strategies for optimal sensor network design in complex acoustic environments.