A nyström-based method for incoherently distributed source localization

Subspace-based methods are attractive solutions to localization problems due to their satisfactory performance and super-resolution property. In large-scale MIMO systems, the prohibitive computational complexity induced by direct eigenvalue decomposition of the high-dimensional covariance matrix severely limits their practical application. In this paper, a Nyström-based method is proposed to solve the complexity problem. A randomized SVD procedure embedded with orthogonal iteration is introduced into the proposed method which releases the computational burden to a big extent. To address the degradation problem of the proposed method in low SNR scenario, an approximate noiseless covariance matrix is devised based on Nyström approximation. Numerical experiments indicate that the proposed method can obtain adequate performance compared with the standard Nyström method as well as the classical subspace-based method, while the complexity of the proposed method is further reduced which makes it a more practical option in large-scale MIMO systems.