Estimation of source directions by eigenvector rotation and combined direction vectors-a least-squares approach

Several previously published eigenvector rotation methods on multiple-source direction estimation using a sensor array (G. Vezzosi, 1982, D.R. Farrier, D.J. Jefferies, 1987) are further investigated in a correlated-source environment. It is demonstrated that one rotation method is basically limited to uncorrelated sources. To overcome this deficiency, an improved estimation procedure for dealing with correlated sources based on rotation plus a least-squares fit is proposed. It is shown that a hybrid approach gives substantially improved estimates of source direction over those obtained by rotation only.<>