DOA Estimation Method for Vector Hydrophones Based on Sparse Bayesian Learning

Hongyan Wang, Yanping Bai, Jing Ren, Peng Wang, Ting Xu, Wendong Zhang, Guojun Zhang

TL;DR

This paper introduces a novel Direction of Arrival (DOA) estimation method, Vector-SBL, that leverages sparse Bayesian learning (SBL) with vector hydrophones. This approach significantly improves DOA estimation accuracy and resolution, especially in challenging underwater environments with low signal-to-noise ratios (SNR), limited snapshots, and coherent sources, outperforming traditional methods.

What Problem Does This Paper Address?

Accurate underwater sound source localization is critical for applications like military reconnaissance, marine resource exploration, and underwater communication. However, the complex and variable underwater environment, coupled with noise interference, significantly degrades the performance of traditional DOA estimation techniques. While vector hydrophones offer a richer sensing capability by capturing both sound pressure and particle velocity, fully exploiting this multidimensional information for high-precision DOA estimation remains a challenge. Existing methods often struggle with low SNR, limited data snapshots, and the presence of multiple or coherent sources, leading to reduced accuracy and resolution.

Methods

The core of this paper is the proposed Vector-SBL algorithm, which combines the strengths of sparse Bayesian learning (SBL) with the capabilities of vector hydrophones.

  • Vector Hydrophone Model: The paper starts by establishing a signal model for a uniform linear array of vector hydrophones. It details how the sound pressure and particle velocity components are related and how these are combined to form the received signal.
  • Sparse Bayesian Learning (SBL): SBL is employed to reconstruct the received vector signal. The fundamental idea is to assume that the signal sources are sparse in the angular domain. This transforms the DOA estimation problem into a sparse signal reconstruction problem. SBL incorporates sparsity constraints into the reconstruction process, which helps in reducing redundant information and improving accuracy, especially under challenging conditions.
  • Hyperparameter Estimation: A key aspect of the Vector-SBL algorithm is its adaptive hyperparameter estimation using a fast factor maximization approach. This allows the algorithm to optimize its parameters based on the specific signal environment, enhancing its robustness and performance across different scenarios.
  • DOA Estimation: After the sparse signal reconstruction, DOA estimation is achieved by performing a spectral peak search on the estimated hyperparameters.

The Vector-SBL method is designed to overcome limitations of traditional algorithms like OMP (Orthogonal Matching Pursuit), MUSIC (Multiple Signal Classification), and CBF (Conventional Beamforming), particularly in scenarios with low SNR, few snapshots, and coherent sources.

Datasets

Not discussed in the paper. The experiments were conducted using simulated data and a real-world lake test experiment.

Key Insights and Ideas

  • Synergy of Vector Hydrophones and SBL: The paper demonstrates that combining the multidimensional sensing capabilities of vector hydrophones with the sparsity-promoting and robust estimation properties of SBL is highly effective for underwater DOA estimation.
  • Robustness to Challenging Conditions: Vector-SBL shows remarkable resilience to low SNR, limited snapshots, and coherent sources, areas where traditional methods often falter. This suggests a significant advancement in practical underwater acoustic sensing.
  • Adaptive Hyperparameter Optimization: The adaptive estimation of hyperparameters is crucial for Vector-SBL's superior performance. It allows the algorithm to tailor itself to varying environmental conditions, leading to consistently high accuracy.
  • Improved Resolution for Closely Spaced Sources: The method exhibits superior resolution capabilities, enabling better discrimination of signals originating from nearby angles, which is a common challenge in acoustic sensing.

Key Results

The paper presents extensive simulation and experimental results comparing Vector-SBL with OMP, MUSIC, and CBF algorithms.

  • Low SNR and Small Snapshots: Under low SNR (e.g., -15 dB with five sources) and few snapshots, Vector-SBL achieved significantly lower RMSE values (e.g., 0.454 vs. 17.351 for OMP).
  • Resolution: For closely spaced sources, Vector-SBL maintained stable and low RMSE values even for small angular differences (e.g., >6°), outperforming other methods.
  • Coherent Sources: Vector-SBL demonstrated high stability and accuracy under coherent source conditions, with RMSE differences of up to 10⁻¹ compared to non-coherent conditions, while other algorithms showed larger variations.
  • Success Rate: In terms of success rate (estimated angle within ±0.25° of true angle), Vector-SBL achieved a 100% success rate at -9 dB SNR, significantly outperforming other algorithms.
  • Lake Test Experiment: A real-world experiment using MEMS vector hydrophones confirmed the high estimation accuracy of Vector-SBL, with a DOA estimated value of 90.1000° and a very small absolute error of 0.1000°.

Limitations and Future Work

  • Environmental Variations: The authors acknowledge that underwater environments are constantly changing and that further research is needed to optimize and enhance the method for even greater reliability and efficiency in practical applications.
  • Computational Complexity: While SBL offers advantages, it can sometimes be computationally intensive. The paper mentions MSBL (Multisnapshot SBL) as an extension that can adapt to complex data environments and enhance processing performance, hinting at potential avenues for further optimization.
  • Further Optimization: The paper suggests that continued research is necessary to further enhance and optimize the method, implying that there is still room for improvement in terms of performance and applicability.

Summary updated

BibTeX
@misc{wang2024doa,
  author = {Hongyan Wang and Yanping Bai and Jing Ren and Peng Wang and Ting Xu and Wendong Zhang and Guojun Zhang},
  title = {{DOA Estimation Method for Vector Hydrophones Based on Sparse Bayesian Learning}},
  year = {2024},
  doi = {10.3390/s24196439},
}

Loading related content...

Figure