Covariance-based DOA estimation for wideband signals using joint sparse Bayesian learning

The DOA estimation problem for wideband signals has attracted much attention in the past years, and how to utilize and derive the common DOA information among frequency bins is the essential question. We address the wideband DOA estimation problem in this paper, and to solve this problem we propose a joint sparse Bayesian learning algorithm based on the sparse signal representation (SSR) of the covariance vectors at each frequency bins. As the SSR of the covariance vectors are used, the proposed algorithm does not need the prior knowledge of the number of sources, and it is immune to the possible coherence between sources. Furthermore, the joint sparse Bayesian model can efficiently exploit the common DOA information among frequency bins. We employ the expectation-maximization (EM) idea in solving this problem. The simulation results show that the proposed algorithm can directly give the DOA estimation without the prior knowledge of the number of sources, even when all the sources are coherent.