Computation of monostatic RCS using adaptive sparsity pattern preconditioning with an MBF-MLFMM approach

Abstract

This work presents a numerical technique for the analysis of the Radar Cross Section (RCS) of large targets combining macro-basis functions, the multilevel fast multipole method and the generation of near-field preconditioners, using an approximate inverse matrix where the sparsity pattern is dynamic and computed considering an upper memory threshold. In order to improve the scalability, a group of rows is computed using the same least squares matrix minimising the Frobenius norm of the error, rendering each row group problem independent from the rest. This approach is applied to large and realistic problems in the test cases included. The presented preconditioner can be used to optimise the convergence of complex problems with respect to the hardware resources available in each case while being transparent to the user.