Robust Beamforming Based on Weighted Vector Norm Regularization

Xiaoying Ren, Yingmin Wang, Lichen Zhang, Qi WangIEEE Access 2021 · DOI

Abstract

Steering vector mismatch rapidly degrades the performance of the minimum variance distortionless response beamformer. To solve this problem, a robust beamforming method based on weighted vector norm regularization is proposed. First, the factors affecting the robustness of the beamformer are analyzed. Second, by introducing the weighted vector norm, an optimization problem is constructed to increase the robustness of the beamformer. Furthermore, the regularization coefficient is provided to achieve a balance between the output power and the robustness of the beamformer. Meanwhile, a method of finding the appropriate regularization coefficient is provided. Then, simulations of an irregular arc array are carried out, showing that the proposed method is robust to the snapshot number. Finally, the results and data analysis indicate the effectiveness of the proposed method.

Introduction

In array signal processing, beamforming is a critical technique used to isolate a desired signal from a background of noise and interference. The Minimum Variance Distortionless Response (MVDR) beamformer is widely regarded as the gold standard because it theoretically provides the optimal signal-to-interference-plus-noise ratio (SINR). However, MVDR is notoriously sensitive to model mismatches. When the assumed steering vector differs from the actual one—due to element displacement, calibration errors, or direction-of-arrival (DOA) estimation errors—the beamformer mistakenly identifies the desired signal as interference and attempts to suppress it. This "signal self-cancellation" effect is particularly severe in high signal-to-noise ratio (SNR) environments or when data snapshots are limited.

Methodology: Weighted Vector Norm Regularization (WVNR)

To address these sensitivities, this paper introduces the Weighted Vector Norm Regularization (WVNR) approach. The method is built on a formal mathematical proof that reducing the norm of the weight vector $w$ directly enhances the beamformer's tolerance to physical and electrical errors in the array.

The proposed optimization framework modifies the standard MVDR problem by incorporating a regularization term:

$$\min_w w^H \hat{R} w + \gamma \|w\|_W^2 \quad \text{subject to} \quad w^H \bar{a} = 1$$

where $\hat{R}$ is the sample covariance matrix, $\bar{a}$ is the assumed steering vector, and $\gamma$ is the regularization coefficient. The use of a weighted norm allows for more nuanced control over the array's response. Because the objective function is convex, it can be solved efficiently and reliably using standard optimization tools.

Automated Tuning via the L-Curve Strategy

A critical challenge in regularized beamforming is choosing the optimal regularization parameter $\gamma$. The authors utilize the L-curve method, a graphical tool that identifies the best trade-off between the output power and the weighted vector norm. By plotting these two values on a log-log scale, a characteristic L-shaped curve is formed. The "corner" of this curve—the point of maximum curvature—represents the optimal balance where the beamformer is robust enough to handle errors without sacrificing its ability to suppress interference.

Simulation Environment

The researchers evaluated the WVNR method using computer simulations of a complex sensor arrangement. Unlike standard linear arrays, they utilized an irregular arc array to test the method's versatility.

The geometry of the 62-element irregular arc array used in the simulations to evaluate beamforming robustness.

The simulation environment involved plane waves in Gaussian white noise with an SNR of 10 dB and an Interference-to-Noise Ratio (INR) of 30 dB. The array configuration consisted of 62 omnidirectional elements arranged in the non-uniform arc shown above.

Key Results and Performance Analysis

The simulations compared WVNR against standard Conventional Beamforming (CBF) and MVDR across several scenarios:

  • Robustness to Mismatch: In tests with a $5^\circ$ steering vector mismatch, the standard MVDR's SINR plummeted by nearly 24 dB. In contrast, the proposed WVNR method maintained a stable output with only a 5.5 dB drop, demonstrating significantly higher reliability.
  • Snapshot Efficiency: MVDR performance often degrades as more snapshots are collected if a mismatch exists, as the algorithm becomes "better" at cancelling the desired signal. WVNR remained stable regardless of the number of snapshots, making it ideal for dynamic environments where data is scarce.
  • Spatial Resolution: Unlike some robust methods that widen the "main lobe" (resulting in a blurry spatial response), WVNR maintains sharp resolution, ensuring it can still distinguish between signals that are spatially close together.

Conclusion

The Weighted Vector Norm Regularization method provides a robust alternative to traditional MVDR beamforming. By linking weight vector norms to physical robustness and providing an automated way to tune regularization via the L-curve method, the authors offer a practical solution for real-world array processing where steering vector mismatches are inevitable. Future research will explore the application of different norm types, such as $l_1$ or $l_\infty$, and validation on physical hardware in underwater or aerial environments.

Summary updated

Abstract

Steering vector mismatch rapidly degrades the performance of the minimum variance distortionless response beamformer. To solve this problem, a robust beamforming method based on weighted vector norm regularization is proposed. First, the factors affecting the robustness of the beamformer are analyzed. Second, by introducing the weighted vector norm, an optimization problem is constructed to increase the robustness of the beamformer. Furthermore, the regularization coefficient is provided to achieve a balance between the output power and the robustness of the beamformer. Meanwhile, a method of finding the appropriate regularization coefficient is provided. Then, simulations of an irregular arc array are carried out, showing that the proposed method is robust to the snapshot number. Finally, the results and data analysis indicate the effectiveness of the proposed method.
BibTeX
@inproceedings{ren2021robust,
  author = {Xiaoying Ren and Yingmin Wang and Lichen Zhang and Qi Wang},
  title = {{Robust Beamforming Based on Weighted Vector Norm Regularization}},
  booktitle = {IEEE Access},
  year = {2021},
  doi = {10.1109/access.2021.3090104},
}
Figure