We derive an exact solution to the problem of estimating the rotation of a rigid body from noisy 3D image data. Our approach is based on total least squares (TLS), but unlike previous work involving TLS, we include the constraint that the transformation matrix should be orthonormal. It turns out that the solution to the estimation problem has the same form as if the data are not noisy, and thus the solution to the standard Procrustes problem can be applied.

Competitive feedback network solutions for narrowband direction finding are developed. The main interest lies in estimating the directions of arrival for closely spaced sources. It is shown that if there is a priori information about the number of sources a conditional maximum likelihood solution can be obtained by the network. A suboptimal estimator that requires no a priori knowledge about the number of signals and a network that performs a covariance fit is also presented. Results are presented using both synthetic and real array data.>

The authors have investigated the narrowband direction-finding problem with neural networks, and propose taking into consideration several snapshots of array data when programming the network. This yields an average version of the interconnection strengths and bias terms that program the network. Simulation results show that using these averaged interconnection strengths and bias terms enhances the performance of the network compared to the method used by R. Rastogi et al. (1987). The authors extend the direction-finding problem to the wideband case in which the sources have broad temporal frequency spectrum. They utilize the information present at different frequencies in the source spectrum to obtain a suitable cost function that can be minimized by a neural network.>