Decentralized Sparse Regression for Super-Resolution DoA Estimation

Decentralized array processing is of increasing importance in recent years as the number of antennas continue to increase to meet the demands of high resolution direction of arrival (DoA) estimation for sensing and localization applications. This work presents a decentralized method for super-resolution DoA estimation with co-located antennas but decentralized processing architecture. In contrast to prior works on high resolution DoA estimation based on subspace of covariance matrix, our proposed approach exploits spatial sparsity. Specifically, multiple source DoAs are estimated by solving a regularized least-square regression with multiple-measurement vector through a coordinate descent framework. Compared with conventional decentralized subspace method, the proposed algorithm has several benefits. It provides estimate with fewer samples than the number of elements in the array with modest information exchanged per interconnect. These benefits translate to a reduction of interconnect data rate requirements. Numerical experiments showed that the proposed solution needs a small number samples and power iterations to deliver comparable performances predicted by Cramér-Rao bound.