An Accuracy Controlled Combined Adaption-Optimization Scheme for Improving the Performance of Three-Dimensional Microwave Devices over a Frequency Band

The design of 3D microwave devices can be improved by using computational optimization techniques combined with numerical simulations of the electromagnetic field. However, high accuracy field analysis is often computationally expensive and time consuming. One way to cut costs is to vary the accuracy level of the analysis at different stages of the optimization. This idea is based on the premise that the accuracy need not be constant throughout the optimization, and so the numerical analysis can be run more cheaply without compromising design quality.;The combined adaption-optimization scheme was applied to 3D rectangular waveguide problems for validation: an E-plane miter bend, a U-bend, an impedance transformer and a compensated magic-T. For comparison, all the problems were also optimized using high-order finite elements at every step. Test results prove the computational efficiency of the new combined scheme at various stages of the optimization. In the early stages, when the element orders are low, the scheme is able to attain similar cost function reductions as the high-order analysis, with computational savings up to a factor of 25. Even in the late stages, when the accuracy is more stringent, the scheme manages a reduction in cumulative computation time of at least a factor of 4.;This thesis presents a software system that minimizes the return loss of 3D microwave devices over a frequency band efficiently through accuracy control. It combines a custom gradient-based optimizer with a p-adaptive frequency-domain finite element solver. The solver computes the cost function and its gradient to a specified accuracy in a cost efficient manner. The p-adaptive solver comprises of two original components: an a-posteriori error estimator to evaluate the error in the cost function gradient, and an error indicator to identify the high error regions in the mesh. The optimizer controls the accuracy of the cost function evaluation through a link with the solver, specifying the required relative error for the gradient at each optimization step.