2017, 39(6): 1340-1347.
doi: 10.11999/JEIT160921
Abstract:
In order to improve underdetermined wideband signals DOA estimation accuracy under low Signal to Noise Ratio (SNR) condition, an off-grid sparse learning via iterative minimization algorithm is proposed. Firstly, the novel algorithm vectorizes the covariance matrix in frequency domain to realize visual array extension, as a result, underdetermined wideband signals are transformed into overdetermined signals. Then linear transform is used to eliminate the noise contained virtual array elements, whitening process is utilized to the estimation error of covariance matrix, as a result, the interference in signals is suppressed. Finally, a Bayesian structure containing the joint sparsity parameter of different frequencies and off-grid parameter is built, the minimization sparse expressions of joint sparsity parameter and off-grid parameter are deduced and corresponding parameters are learned iteratively. Compared with other methods, the proposed method does not rely on any prior information, suppresses the inference in virtual array elements more efficiently, reduces the effects of off-grid problem, and gets higher DOA estimation accuracy and resolution under low SNR condition. Simulation experiments verify the validity of the novel algorithm.