Sparse deconvolution beamforming with non-negative L1-αL2 regularization for acoustic source localization

Area of Science:

• Acoustics
• Signal Processing
• Computational Mathematics

Background:

• Sparse deconvolution beamforming methods are crucial for sound source localization.
• Existing methods may lack sufficiently accurate sparse descriptions, limiting localization precision.
• Enhancing the sparse representation is key to improving localization performance.

Purpose of the Study:

• To propose a novel sparse deconvolution localization method, FFT-L1ML2, utilizing non-convex L1-αL2 regularization.
• To enhance sound source localization accuracy by better approximating the L0 norm.
• To develop an efficient optimization solver for the proposed model.

Main Methods:

• Implementation of a sparse deconvolution localization method (FFT-L1ML2).
• Application of non-convex L1-αL2 regularization to approximate the L0 norm for improved sparse representation.
• Development of an optimization solver using forward gradient descent and backward proximal operator.

Main Results:

• The FFT-L1ML2 method demonstrates superior localization accuracy compared to existing techniques.
• Improved energy concentration and significant reduction in pseudo sound sources were observed.
• The method shows effectiveness in both simulated and experimental scenarios.
• Reduced computational cost was noted as an advantage.

Conclusions:

• The proposed FFT-L1ML2 method offers a significant advancement in sparse deconvolution-based sound source localization.
• The L1-αL2 regularization effectively captures the sparse structure of sound sources, leading to enhanced accuracy.
• FFT-L1ML2 provides a more effective and computationally efficient alternative for acoustic source identification.