[Paper Review] A Unified SVD-Modal Solution for Sparse Sound Field Reconstruction with Hybrid Spherical-Linear Microphone Arrays

Shunxi Xu, Thushara D. Abhayapala+1 more|arXiv (Cornell University)|Feb 3, 2026

Hearing Loss and Rehabilitation0 citations

TL;DR

The paper presents a data-driven SVD-modal framework for sparse sound-field recovery using a hybrid spherical–linear microphone array, unifying SH processing with complementary LMA modes to improve robustness and resolution in reverberant environments.

ABSTRACT

We propose a data-driven sparse recovery framework for hybrid spherical linear microphone arrays using singular value decomposition (SVD) of the transfer operator. The SVD yields orthogonal microphone and field modes, reducing to spherical harmonics (SH) in the SMA-only case, while incorporating LMAs introduces complementary modes beyond SH. Modal analysis reveals consistent divergence from SH across frequency, confirming the improved spatial selectivity. Experiments in reverberant conditions show reduced energy-map mismatch and angular error across frequency, distance, and source count, outperforming SMA-only and direct concatenation. The results demonstrate that SVD-modal processing provides a principled and unified treatment of hybrid arrays for robust sparse sound-field reconstruction.

Motivation & Objective

• Motivate robust sparse recovery of 3D sound fields using hybrid SMA–LMA arrays.
• Develop a unified, data-driven modal dictionary via SVD of the transfer operator.
• Generalize spherical harmonics processing and exploit LMA information for enhanced spatial selectivity.
• Demonstrate improved robustness in reverberant conditions compared with SMA-only and concatenation approaches.

Proposed method

• Model the transfer operator between plane-wave directions and microphone observations with a discretized, hybrid SMA–LMA array.
• Apply singular value decomposition to obtain an orthogonal modal basis that diagonalizes the transfer operator.
• Truncate to the dominant K singular values to form a reduced-rank, well-conditioned dictionary.
• Whiten and project observations using the SVD factors to obtain a stable sparse-recovery problem.
• Solve for sparse plane-wave amplitudes using iterative reweighted least squares with an l2,p-norm (p=0.7) regularization, initialized by l1 minimization.
• Relate SMA-only processing to spherical-harmonic processing and show how LMAs extend the basis beyond SH.

Experimental results

Research questions

• RQ1 Can a unified SVD-based modal dictionary effectively fuse SMA and LMA measurements for sparse sound-field reconstruction?
• RQ2 How does the SVD-modal approach compare to SMA-only and joint-SR in terms of energy-map fidelity and localization accuracy under reverberation?
• RQ3 What is the impact of the number of modes (K) on reconstruction quality and robustness across frequencies and distances?
• RQ4 Do principal angles between SH and SVD subspaces quantify the divergence and benefits of the hybrid modal basis?

Key findings

• Modal SVD basis yields orthogonal, frequency-dependent modes that improve reconstruction stability.
• Hybrid SMA–LMA processing reduces energy-map mismatch and angular error across frequency, distance, and source count compared with SMA-only and direct concatenation.
• Increasing the number of modes improves angular localization accuracy but may slightly raise energy-map mismatch due to weaker, noise-sensitive modes.
• The approach performs on par with a residue-refinement baseline while providing a principled, unified framework.
• Modal analysis shows frequency-dependent divergence from SH and enhanced spatial resolution in the hybrid array.
• Under reverberant conditions, the proposed method consistently outperforms SMA-only and concatenation strategies.