Low Complexity Modeling of Cross-Spectral Matrix and Its Application in the Non-Synchronous Measurements of Microphones Array

Liang Yu, Yi Fan, Ning Chu, Qian Huang, Yue Ning, Peng HouIEEE Access 2021 · DOI

Introduction

High-resolution acoustic imaging is essential for identifying noise sources in complex environments like automotive testing and underwater sonar. However, achieving high resolution typically requires large, dense microphone arrays, which are often cost-prohibitive or physically impractical. A common workaround is non-synchronous measurement, where a small array is moved to multiple positions to simulate a larger "virtual" array.

While effective, this technique introduces a significant mathematical challenge: the phase relationship between different measurement positions is lost. This results in a Cross-Spectral Matrix (CSM) that is only partially filled (a block-diagonal structure). This paper proposes a novel low-complexity modeling approach using max-norm minimization to reconstruct the missing data and produce high-fidelity acoustic images.

Methodology: Non-Synchronous Measurement and Matrix Completion

The researchers frame the reconstruction of the sound field as a Matrix Completion (MC) problem. Traditional methods often rely on rank minimization or nuclear-norm minimization. Rank minimization assumes a limited number of sound sources, while nuclear-norm minimization focuses on limiting total signal energy.

Non-synchronous measurement setup and the resulting Cross-Spectral Matrix (CSM) completion problem.

This study introduces the Proximal-Point method for Max-norm based Spectral matrix completion (PPMSMC). The authors argue that the max-norm is a superior "convex relaxation" for acoustic data because it bounds the peak influence of sources rather than just total energy. This distinction is crucial for maintaining structural integrity in high-frequency or high-noise scenarios.

Experimental Setup and Simulations

To validate the PPMSMC algorithm, the authors conducted extensive numerical simulations. The setup involved a 41x41 grid source plane with point sources expanded in a B-spline basis to create realistic velocity distributions. To test robustness, complex Gaussian noise was added, creating Signal-to-Noise Ratios (SNR) ranging from a clean 60 dB to a challenging 3 dB.

Key Insights and Results

Breaking the Frequency Barrier

Standard arrays are limited by the Nyquist limit; if the sound frequency is too high for the microphone spacing, spatial aliasing occurs. The proposed max-norm completion method effectively "breaks through" this limit. At frequencies above 2100 Hz, the PPMSMC algorithm significantly outperformed nuclear-norm methods (like FISTA) in both matrix completion accuracy and final image resolution.

Robustness in High-Noise Environments

In extreme noise conditions (3 dB SNR), the max-norm approach demonstrated remarkable resilience. While nuclear-norm results became unusable and "very poor," the PPMSMC method maintained a low Acoustic Reconstruction Error (ARE), providing clear and usable acoustic images where other methods failed.

Trade-offs and Comparisons

While the max-norm approach excels in noise and high-frequency resolution, the study noted that the nuclear-norm remains more robust to "shift distance." This means if the array is moved in larger jumps with less overlap between positions, nuclear-norm methods might still hold an advantage.

Conclusion and Future Work

The introduction of max-norm minimization via the PPMSMC algorithm provides a powerful new tool for acoustic engineers. By focusing on peak signal bounds rather than total energy, it offers superior performance for high-frequency imaging and high-noise environments.

However, the method does require an initial estimate of the source rank, and the algorithm itself is more computationally sophisticated than simpler rank-based approaches. Future research should focus on validating these findings with real-world experimental data in wind tunnels or anechoic chambers to account for physical factors like hardware calibration and reflections.

Summary updated

BibTeX
@article{yu2021low,
  author = {Liang Yu and Yi Fan and Ning Chu and Qian Huang and Yue Ning and Peng Hou},
  title = {{Low Complexity Modeling of Cross-Spectral Matrix and Its Application in the Non-Synchronous Measurements of Microphones Array}},
  journal = {IEEE Access},
  year = {2021},
  doi = {10.1109/access.2021.3055728},
}
Figure