Direction finding using uniform arrays and the constrained total least squares method

Direction finding using snapshots collected from a uniform array is equivalent to estimating the two-dimensional frequencies of sinusoids from a set of double indexed samples. The authors present a novel technique for estimating the two-dimensional frequencies which is based on a generalization of the constrained total least squares technique, to nonlinear equations whose coefficients have been perturbed by additive noise. Multilinear prediction on the data is used to arrive at these equations, and the frequencies are extracted directly from the solution of these equations. Simulation results are used to compare the resolution and accuracy of this technique with the corresponding Cramer-Rao bound for a variety of signal scenarios