Abstract

This manuscript introduces a robust and efficient sampling technique for optimizing data retrieval from large-dimension covariance matrix (CM). The proposed approach focuses on selecting the least-correlated and most-independent columns from the CM to construct the so-called projection matrix (PM) applied in angle of arrival (AOA) estimation. In contrast to the developed sampling techniques in literature, the recommended method excludes the signal variances located in the main diagonal of the CM during the column selection process. That is to say, the proposed technique focuses entirely on the off-diagonal elements of the CM that captures the covariance (correlations) between signals collected from different array elements. The proposed methodology is therefore named variance-omitted sampling technique (VoST). Applying this principle, we are able to extract the columns with minimal signal correlations corresponding to the off-diagonal entries of the CM and decrease the computational burden involved in the PM formulation process. To validate the theoretical claims and demonstrate the advantages of the proposed technique, a numerical example is provided, followed by extensive Monte Carlo simulations across several scenarios in which the performance of the proposed method is systematically compared with the existing techniques. The results demonstrate that VoST consistently surpasses previous algorithms in estimation resolution, root mean square error (RMSE), successful detection rate, ability to detect correlated signals, and computational speed.

Key words: AOA estimation, Covariance matrix sampling, off-diagonal covariance matrix, computational complexity, VoST.