Root sparse asymptotic minimum variance for off-grid direction-of-arrival estimation

Abstract

Sparsity-based direction-of-arrival (DOA) estimation algorithms have received much attention due to their good performances. However, most of these algorithms suffer from the so-called basis mismatch if the true DOAs deviate from the discrete grid points. This paper provides an off-grid DOA estimation method which iteratively updates the grid until some of the grid points coincide with the DOAs. Based on the first-order Taylor series expansion of the true steering vectors, the deviations of these grid points from the true DOAs, i.e., the so-called grid errors, are linked to the steering vectors defined on the grid by a linear relation approximately. Using this linear relation, the grid errors can be estimated under the asymptotic minimum variance criterion. The grid is modified by adding the grid errors to the grid points, leading to them closer to the true DOAs, and consequently mitigating the basis mismatch. It is shown by simulations that the proposed method achieves high performance both in terms of estimation accuracy and computational efficiency.