Directions-of-arrival estimation using a sparse spatial spectrum model with uncertainty

This paper is concerned with the estimation of the directions-of-arrival (DOA) of narrowband sources using a sparse spatial spectral model, when the model itself is not precise. When the model uncertainty is limited to DOAs not belonging to the search grid, off-grid error linearization given in [1], and constraint relaxation are used to effect a convex problem. This approach is useful either in reducing the dimension through the use of coarser candidate DOA grid, or improving the performance of estimators that use dense grids. In addition, a particular diagonal loading approach is proposed to reduce the sensitivity of the estimator to the choice of its regularization parameter.