Adaptive Diagonal Loading using Krylov Subspaces for Robust Beamforming

TL;DR

This paper presents an efficient Lanczos-based method to estimate extreme eigenvalues of the spatial correlation matrix for robust adaptive beamforming, reducing computational complexity while maintaining performance.

Contribution

We introduce a Krylov subspace technique using Lanczos iterations to accurately estimate eigenvalues, enabling faster adaptive diagonal loading for large microphone arrays.

Findings

Lanczos method achieves eigenvalue estimates comparable to exact EVD.

The approach maintains target signal integrity and interference suppression.

Computational cost is reduced from $ ext{O}(M^3)$ to $ ext{O}(kM^2)$.

Abstract

Reliable adaptive beamforming is critical for large microphone arrays operating in highly dynamic acoustic environments. In scenarios characterized by fast-moving talkers and interferers, the available sample support for estimating the spatial correlation matrix is often snapshot-deficient. This deficiency degrades the White Noise Gain (WNG), leading to severe target signal cancellation. To ensure stable and robust beamforming, we previously proposed an adaptive diagonal loading method that leverages the Kantorovich inequality to guarantee the WNG remains strictly within specified bounds. However, accurately determining the smallest necessary loading level requires calculating the extreme eigenvalues of the spatial correlation matrix, a computationally expensive $\mathcal{O}(M^3)$ operation for large arrays. In this paper, we introduce a highly efficient $\mathcal{O}(kM^2)$ estimation technique using Lanczos iterations to build a small Krylov subspace. By projecting the correlation matrix onto a tridiagonal matrix of dimension $k \ll M$, we extract Ritz values that rapidly converge to the exact extreme eigenvalues. Our evaluations demonstrate that this Lanczos-accelerated approach achieves performance identical to exact Eigenvalue Decomposition (EVD), ensuring optimal interference suppression and strict WNG adherence at a fraction of the computational cost.