Applications of model-based parameter estimation in electromagnetic computations

A widely used computer model in electromagnetics is the method of moments (MM) whereby an integral equation is discretized and approximated as a matrix whose solution yields a sampled representation of the physical problem of interest. The computer time T required to evaluate a MM model at a single frequency depends on the number of unknowns or equations N as T~AN²+BN³, where A and B are computer- and algorithm-dependent constants. Many frequency samples F may be required to establish a response over a prescribed frequency range, leading to a computer time proportional to TF. One way of reducing this time is by minimizing the number of frequencies at which the MM calculation must be done, i.e. by reducing F using prior knowledge of the expected frequency behavior in the form of a model, a pole series, for example. Another is to reduce T, which for some problems is dominated by the matrix fill time, or the A term. Both possibilities are candidates for model-based parameter estimation (MBPE) as demonstrated here, the first using a predetermined, pole-series model and latter a multinomial, self-organizing model