An Improved Reduced-Dimension Robust Capon Beamforming Method Using Krylov Subspace Techniques

Xiaolin Wang, Xihai Jiang, Yaowu ChenSensors 2024 · DOI

TL;DR

This paper introduces an improved robust Capon beamforming (IRDRCB) method that addresses a critical limitation in existing reduced-dimension robust Capon beamforming (RDRCB) techniques: the potential for an invalid diagonal loading level. By establishing upper and lower bounds for this crucial parameter, the IRDRCB achieves more robust performance, better interference suppression, and significantly reduced computational complexity compared to previous methods, making it well-suited for real-time applications.

What Problem Does This Paper Address?

In passive sonar and radar applications, beamforming is essential for tasks like source localization and signal detection. The Capon beamformer (also known as MVDR) offers high resolution and interference rejection but is sensitive to errors in the array steering vector (ASV). Robust Capon Beamforming (RCB) techniques aim to mitigate this sensitivity. Reduced-dimension RCB (RDRCB) methods, particularly those using Krylov subspace techniques, offer computational efficiency. However, a key challenge with RDRCB is its reliance on a diagonal loading level determined by a Newton iteration method. If the initial guess for this iteration is incorrect or the Hessian matrix is not positive definite, the resulting diagonal loading level can become invalid, leading to a sharp decline in signal-to-noise ratio (SNR) gain and overall performance degradation. This paper tackles the problem of ensuring a valid and robust diagonal loading level within the RDRCB framework.

Methods

The core innovation of this paper is the development of an Improved Reduced-Dimension Robust Capon Beamforming (IRDRCB) method. The authors build upon existing RDRCB techniques that utilize Krylov subspace methods (specifically, the conjugate gradient algorithm) for dimension reduction. The key advancements are:

  • Derivation of Diagonal Loading Boundaries: The paper analyzes how the diagonal loading factor varies with the eigenvalues of the reduced-dimensional covariance matrix. Based on this analysis, they derive explicit upper and lower boundaries for the diagonal loading level.
  • Robust Diagonal Loading Determination: The IRDRCB method ensures that the determined diagonal loading level stays within these derived boundaries. If the Newton search for the loading level yields a value outside these bounds, the algorithm employs a bisection method to find a valid level. This approach prevents the invalid loading level issue that plagues standard RDRCB.
  • Stopping Criteria: The IRDRCB incorporates specific stopping criteria to manage the iterative process of finding the optimal diagonal loading level.

The paper also details the underlying RDRCB framework, which involves:

  • Calculating the sample covariance matrix.
  • Evaluating the Reduced Dimension Transform (DRT) using the conjugate gradient algorithm.
  • Forming intermediate matrices and performing eigendecomposition.
  • Using a Lagrangian function and Newton search to find the Lagrange multiplier, which is then used to determine the optimal diagonal loading level.

Datasets

Not discussed in the paper. The evaluation relies on simulated data and experimental data from a real-world scenario.

Key Insights and Ideas

  • The Achilles' Heel of RDRCB: The paper clearly identifies the invalid diagonal loading level as a critical vulnerability in existing RDRCB methods, directly impacting their robustness and performance.
  • Eigenvalue Analysis for Robustness: By analyzing the relationship between eigenvalues and the diagonal loading factor, the authors unlock a principled way to bound this parameter, leading to a more stable algorithm.
  • Bisection Method as a Safety Net: The integration of the bisection method provides a reliable fallback mechanism to guarantee a valid diagonal loading level, even if the initial Newton search is problematic.
  • Trade-off Between Rank and Complexity: The paper highlights that while reduced rank (N) is key to computational efficiency, there's a point (around N/M = 0.8) where increasing rank can lead to higher computational demands, suggesting an optimal operating range for N.

Key Results

  • Improved Robustness: The IRDRCB effectively solves the problem of sharp declines in SNR gain and invalid diagonal loading levels.
  • Enhanced Interference Suppression: Experimental results show that the IRDRCB provides 3~5 dB higher interference noise suppression compared to conventional adaptive beamforming.
  • Significant Computational Complexity Reduction: The IRDRCB reduces computational complexity by 15% to 20% compared to the standard Robust Capon Beamforming (RCB) method.
  • Performance Parity with RCB: In simulations, the IRDRCB's performance is consistently close to, or even better than, that of the RCB methods, especially under various SNR conditions and with steering vector errors.
  • Real-time Applicability: The substantial reduction in computational complexity makes the IRDRCB highly suitable for real-time implementation in sonar and radar systems.

Limitations and Future Work

  • Limitations:
    • The paper notes that when the rank reduction ratio (N/M) exceeds 0.8, the IRDRCB's computational complexity increases.
  • While the IRDRCB performs well, the paper mentions that for very small numbers of snapshots, the standard RCB might offer a slightly higher SNR gain.
  • Future Work:
    • The authors suggest that the IRDRCB's efficiency and robustness make it an ideal candidate for real-time sonar and radar systems.
  • They anticipate that continued advancements in hardware acceleration will further enhance the potential of the IRDRCB in real-time processing.

Summary updated

Abstract

A reduced-dimension robust Capon beamforming method using Krylov subspace techniques (RDRCB) is a diagonal loading algorithm with low complexity, fast convergence and strong anti-interference ability. The diagonal loading level of RDRCB is known to become invalid if the initial value of the Newton iteration method is incorrect and the Hessel matrix is non-positive definite. To improve the robustness of RDRCB, an improved RDRCB (IRDRCB) was proposed in this study. We analyzed the variation in the loading factor with the eigenvalues of the reduced-dimensional covariance matrix and derived the upper and lower boundaries of the diagonal loading level; the diagonal loading level of the IRDRCB was kept within the bounds mentioned above. The computer simulation results show that the IRDRCB can effectively solve the problems of a sharp decline in the signal-to-noise ratio gain and an invalid diagonal loading level. The experimental results demonstrate that the interference noise of the IRDRCB is 3~5 dB higher than that of conventional adaptive beamforming, and the computational complexity is reduced by 15% to 20% compared with that of the RCB method.
BibTeX
@inproceedings{wang2024improved,
  author = {Xiaolin Wang and Xihai Jiang and Yaowu Chen},
  title = {{An Improved Reduced-Dimension Robust Capon Beamforming Method Using Krylov Subspace Techniques}},
  booktitle = {Sensors},
  year = {2024},
  doi = {10.3390/s24227152},
}

Loading related content...

Figure