Sparse Bayesian Learning for DOA Estimation with Recursive Grid-Refining

The modeling error for off-grid direction-of-arrival (DOA) estimation can be alleviated by the technique of linear approximation, but cannot be fully eliminated, especially for coarse grids. In this paper, we not only adopt a linear approximation to cover the off-grid gap, but also try to combine it with the idea that all the sampled locations can be viewed as the adjustable parameters. Then, we utilize an expectation-maximization (EM) to recursively refine the grid points, rather than updating the coefficients in the linear approximation directly. In this way, the refined grid points will tend to the true DOAs after several iterations. Simulation results illustrate that the proposed method can give a fast DOA estimation while remain a reasonable accuracy.