Large-sample analysis of MUSIC and min-norm direction estimators in the presence of model errors (Q1922610)

The authors analyse the mean square error of some direction-of-arrival (DOA) estimators and provide effective means for evaluation of the DOA's variance and bias. The analysis is made under fairly general and realistic conditions -- there may be errors in the noise covariance, the assumed number of sources, and in sensor positions. Furthermore, no model for the noise covariance matrix is assumed. This results in leaving out requirements for large signal to noise ratio or small errors which are necessary in other studies. An examination of the MUSIC and Min-Norm estimators in the presence of different model errors is reported. It confirms the optimal property of the MUSIC estimator for small errors but at the same time gives some advantage to the Min-Norm estimator for larger errors.