Sparse Bayesian Learning for DOA Estimation in MIMO Radar with Unknown Nonuniform Noise

In this paper, a sparse Bayesian learning framework for DOA estimation in multiple input multiple output (MIMO) radar is proposed with unknown nonuniform noise. In the proposed method, the redundant elements of MIMO radar can be eliminated by using the reduced dimensional (RD) transformation. Then a sparse Bayesian model of covariance vector is formulated by assuming that the prior source power is independent zero-mean Gaussian distributed with hyperparameters for its unknown variance. The hyperparameters and nonuniform noise variances are estimated by utilizing the expectation-maximization (EM) algorithm and least squares (LS) criterion, respectively. Finally, the spectrum of hyperparameters is used to estimate the coarse DOA, and a high-precision DOA estimation is achieved by using a refined 1-D searching procedure based on the reconstruction result. Simulation results have demonstrated that the proposed method can work well with different nonuniform noise and achieve better performance.