Low-frequency source imaging in an acoustic waveguide

Josselin GarnierarXiv 2020 · arXiv

Abstract

Locating sound sources in underwater waveguides often requires large sensor arrays. However, at low frequencies where wavelengths are comparable to waveguide depth, arrays are frequently sub-wavelength in size. This paper provides a rigorous mathematical framework for imaging in these regimes, analyzing how different array geometries—vertical, horizontal, and planar—extract information. The study reveals that while planar arrays offer superior noise robustness, their information capacity is fundamentally constrained by the wave dispersion relation, leading to significant data redundancy.

Introduction

In underwater acoustics, environments like shallow oceans act as waveguides where sound is trapped between the surface and the seafloor. Traditional imaging techniques rely on arrays spanning a significant portion of the waveguide to achieve high resolution. In low-frequency scenarios, the wavelength $\lambda$ is large, and the available sensor array is often much smaller than $\lambda$.

Schematic of a 2D acoustic waveguide with a sensor array (antenna) and a sound source.

The core challenge in these "sub-wavelength" regimes is extracting the amplitudes of guided modes—the specific vibrational patterns within the channel—from limited and noisy measurements. This paper quantifies the information capacity of various array configurations and identifies the physical laws limiting their performance.

Methodology

The imaging problem is approached through a combination of wave physics and linear algebra. The acoustic field is first decomposed into a finite number of guided modes and an infinite set of evanescent (decaying) modes. The primary goal is to estimate the amplitudes of these guided modes.

Because the sensor arrays are small, the resulting sensing matrix is highly ill-conditioned, making the inversion process sensitive to noise. To stabilize the estimation, the author employs Tikhonov regularization and hard thresholding. The "information capacity" of an array is measured by the effective rank of the sensing matrix, determined via Singular Value Decomposition (SVD). This metric indicates how many independent signal components the array can distinguish above the noise floor. Finally, Reverse-Time Migration is used to back-propagate the estimated amplitudes and localize the source.

Array Performance and the Redundancy Paradox

The study compares four geometries: discrete point arrays, dense planar squares, vertical lines, and horizontal lines. Numerical simulations in both homogeneous and parabolic waveguides reveal a clear hierarchy of performance: Planar > Vertical Linear > Horizontal Linear.

Vertical vs. Horizontal Arrays

For a fixed diameter, vertical arrays outperform horizontal ones because they better sample the mode profiles $\phi_j(z)$. Horizontal arrays rely on phase differences that are less distinct at low frequencies. In the high-frequency limit, a horizontal array must be twice as long as a vertical array to provide equivalent information.

The Planar Advantage and Its Limits

Moving to a 2D planar array provides a massive boost in stability, allowing for source localization at noise levels several orders of magnitude higher than linear arrays. However, a "redundancy paradox" emerges: while a planar array's geometric capacity might be expected to grow quadratically ($Q^2$), its effective rank is analytically derived as only $2Q-1$.

This limitation is intrinsic to scalar waves. The horizontal and vertical wavenumbers are coupled by the dispersion relation ($\alpha_j^2 + \beta_j^2 = k_o^2$), meaning the data captured by a small planar array is physically redundant regardless of the medium's complexity.

Results and Conclusion

Numerical results demonstrate that even with an array as small as 4% of the wavelength ($0.04\lambda_o$), sub-wavelength localization accuracy is achievable given a sufficiently high signal-to-noise ratio. The derived effective rank formula ($2Q-1$) provides a concrete metric for predicting array performance in engineering applications.

While this analysis is limited to 2D time-harmonic sources in idealized waveguides, it highlights a critical barrier in PDE-constrained inverse problems: adding more sensors in a small area yields diminishing returns due to the inherent redundancy of the wave field. Future work may explore 3D waveguides with azimuthal modes and the potential for broadband signals to mitigate these redundancy effects.

Summary updated

Abstract

Time-harmonic far-field source array imaging in a two-dimensional waveguide is analyzed. A low-frequency situation is considered in which the diameter of the waveguide is slightly larger than the wavelength, so that the waveguide supports a limited number of guided modes, and the diameter of the antenna array is smaller than the wavelength, so that the standard resolution formulas in open media predict very poor imaging resolution. A general framework to analyze the resolution and stability performances of such antenna arrays is introduced. It is shown that planar antenna arrays perform better (in terms of resolution and stability with respect to measurement noise) than linear (horizontal or vertical) arrays and that vertical linear arrays perform better than horizontal arrays, for a given diameter. However a fundamental limitation to imaging in waveguides is identified that is due to the form of the dispersion relation. It is intrinsic to scalar waves, whatever the complexity of the medium and the array geometry.
BibTeX
@misc{garnier2020low,
  author = {Josselin Garnier},
  title = {{Low-frequency source imaging in an acoustic waveguide}},
  year = {2020},
  eprint = {2008.02022},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2008.02022},
}
Figure