On DOA estimation based on higher order statistics

The authors present a DOA estimation procedure which is based on the assumption of highly correlated Gaussian noise contaminating nonGaussian sources, and which jointly employs second order statistics and higher order cumulants statistics. From second order statistics, they identify a set of candidate angles in which both true signal DOA´s and spourious noise induced DOA´s are present. Then, the signal DOA´s are extracted by resorting to higher order statistics and to the nonGaussianity of the sources. Even though the estimates are biased when the noise is not fully correlated, simulation results show that, for SNR values below a certain threshold, this bias does not (significantly) affect the estimation accuracy and that the proposed approach outperforms the straight-forward application of Root-MUSIC to the matrix of fourth order cumulants.