Coordinate Descent Method for AVS Linear Array DOA Estimation in Non-Uniform Noise

• Published in: IEEE sensors journal Vol. 24; no. 7; pp. 10742 - 10754
• Main Authors: Wang, Mingguang, Zhu, Zhongrui, Chen, Feng, Wang, Huan
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.04.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Acoustic vector sensor arrays; Algorithms; coordinate descent; Covariance matrices; Covariance matrix; Direction of arrival; direction-of-arrival (DOA) estimation; Direction-of-arrival estimation; Estimation; Estimators; Factor analysis; Linear arrays; Maximum likelihood estimation; non-uniform noise; Nonuniformity; Orthogonality; Performance degradation; Sensor arrays; Sensors; Subspaces; Underwater acoustics; White noise
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2024.3363030

Direction-of-arrival (DOA) estimators are typically derived assuming uniform white noise, which is characterized by a scaled identity covariance matrix. However, in practice, the noise can be non-uniform, with an unknown diagonal covariance matrix. Ignoring this non-uniformity can greatly degrade the performance of DOA estimators. To address this problem, we select the Factor Analysis in the case of Anisotropic Noise algorithm, which is a coordinate descent method, to mitigate the impact of non-uniformity and improve convergence properties. Additionally, we consider the issue of virtual sources that arise from unequal noise powers in pressure and velocity components within an underwater acoustic vector sensor array. To address the impact of virtual sources, we increase the number of virtual sources included in the signal subspace and derive a corresponding coordinate decent method to obtain more accurate estimated subspaces. Then, the DOAs are determined using traditional DOA estimators with the subspace estimation and/or noise covariance matrix estimation. The proposed algorithm improves the accuracy of estimating the signal subspace and noise covariance matrix while enhancing the orthogonality between the signal and noise subspaces for more accurate DOA estimation. Simulation and experimental results are included to demonstrate the superiority of the proposed approach.