Continuous Approximation Based Dimension-Reduced Estimation for Arbitrary Sampling

Frequency/direction-of-arrival (DOA) estimation via grid searching or sparse representation is time-consuming in 2D cases. Few dimension-reduction methods exist for arbitrary temporal/spatial sampling. In this letter, we propose the Continuous Approximation based Dimension-Reduced Estimation (CADRE) framework to address this issue. By the linear approximation of vectors from a continuous space using only a few bases, dimension reduction is achieved. For some complicated manifolds or realistic scenarios with only a discrete set of steering vectors available, a discrete simplification is also effective. For parameter estimation, parameter-space multiple signal classification and a group-sparse based algorithm are proposed. Simulations verify the superiority of the proposed estimators in both speed and accuracy.