ON THE USE OF SVD-IMPROVED POINT MATCHING IN THE CURRENT-MODEL METHOD

Fictitious-current models have been applied extensively in recent year s to a variety of scattering problems in computational electromagnetic s. This paper introduces an approach which uses the singular value dec omposition (SVD) to improve the accuracy of the numerical solution. In this approach, the SVD is essentially facilitating a systematic way t o optimally reduce the generalized inverse matrix used in the solution to a submatrix of smaller rank. This reduction strikes a balance betw een the fulfillment of the boundary conditions at the matching points and that between them. Clearly, the boundary condition errors at the m atching points are no longer strictly zero. However, the previously di scernible errors between the matching points are markedly suppressed. The suggested approach is efficacious not only when the impedance matr ix is inherently singular or highly ill conditioned, but also when thi s matrix is entirely well conditioned. It can be generalized and imple mented in any method of moments code which uses point matching for tes ting. The approach has been incorporated into an existing solution bas ed on the current-model method for the problem of scattering from peri odic sinusoidal surfaces, and is shown to render the solution more acc urate.