Sparse Bayesian Learning Based Algorithm for DOA Estimation of Closely Spaced Signals

Off-grid Direction Of Arrival (DOA) estimation aims to handle the mismatch between the actual DOA and the presumed grid points. For DOAs of closely spaced signals, sparse grid points leads to degradation of accuracy and resolution, although dense grid points can improve the estimation accuracy, it significantly increases the computational burden. To solve this problem, this paper proposes a Sparse Bayesian Learning (SBL) based algorithm for DOA estimation of closely spaced signals, which consists of three steps. Firstly, a novel fixed point iterative method for signal of Laplace priori is derived to pre-estimate the hyper-parameters by maximizing the array's marginal likelihood function, which results in faster convergence speed compared to other classical SBL algorithms. Secondly, a new grid interpolation method is implemented to optimize a set of grid points, and signal power and noise variance are estimated again to resolve closely spaced DOAs. Finally, an expression of maximum likelihood function with respect to angle is derived to improve the search of the off-grid DOA. Simulation results show that the proposed algorithm has higher accuracy and resolution for closely spaced DOAs with higher computational efficiency compared with other classical algorithms based on SBL.