Wideband DOA estimation by joint sparse representation under Bayesian learning framework

Wideband direction of arrival (DOA) estimation is a practical problem frequently occurring in sonar application. Compared to the entire angular domain, targets only occupy a few directions and the received signals are considered to be sparse in the angular domain. It is further noted that signals in different spectrum bands show a strong joint sparsity due to the fact that targets from different directions share the spectrum. This paper exploits the joint sparsity of the signals and reformulates the DOA estimation problem under the Bayesian learning framework. The resulted method is a data-driven learning process and does not need the tedious parameter tuning. Comparing to the conventional delay-sum beamformer, the proposed method has the advantages of reduced number of sensors, reduced spatial aliasing and increased resolution. The improved performance is validated by real sonar data experiments.