Variational Bayesian Compressive Sensing with Equivalent Source Modeling for Sound Field Reconstruction.

Area of Science:

• Acoustics
• Signal Processing
• Computational Physics

Background:

• Conventional Bayesian compressive sensing for sound field reconstruction faces challenges with computational complexity and slow convergence.
• Accurate reconstruction from under-sampled measurements is crucial in various acoustic applications.

Purpose of the Study:

• To develop a computationally efficient and accurate sound field reconstruction method using variational Bayesian compressive sensing and equivalent source modeling.
• To address the limitations of existing methods in terms of speed and robustness.

Main Methods:

• A sparse representation of the sound field is established using the equivalent source method.
• Hierarchical prior distributions are assigned to equivalent source strengths and noise precision.
• Mean-field variational inference is employed for efficient estimation of source strengths and sound field reconstruction.

Main Results:

• The proposed method achieves superior reconstruction accuracy compared to conventional Bayesian compressive sensing and orthogonal matching pursuit.
• Significant reduction in computational burden and enhanced robustness in low signal-to-noise ratio scenarios were observed.
• The method retains the statistical advantages of Bayesian modeling while improving computational efficiency.

Conclusions:

• The variational Bayesian compressive sensing framework with equivalent source modeling offers an effective solution for sound field reconstruction.
• This approach provides a balance between statistical rigor, computational efficiency, and reconstruction accuracy.
• The method demonstrates significant improvements for practical applications requiring fast and reliable acoustic analysis.