Optimization of RF/microwave filters and diplexers using hybrid evolutionary programming and model calibration technique

This thesis presents efficient and robust optimization techniques for computer aided design and diagnosis of complex RF/microwave filters and diplexers used in modern communication systems. We specifically focus on two areas in the design/verification cycle of microwave filters and diplexers, namely the numerical optimization methods and the methodologies for improving the speed of electromagnetic (EM) optimization.; It is shown that the available numerical optimization methods used by many filter designers and implemented in RF and microwave software tools are not suitable for several complex multimodal optimization problems. In order to address this issue, an efficient global optimization algorithm is developed based on the hybrid evolutionary programming (EP). The formulation of conventional EP is reviewed and a methodology for enhancing the robustness of the algorithm using the clustering technique is presented. The modified EP is then combined with the quasi-Newton search method in a novel fashion to improve the speed of optimization process. Furthermore, several specific enhancements for RF and microwave filters are presented. The outline of the developed numerical software based on the hybrid EP is also presented. Performance of the new algorithm was tested using several benchmark problems as well as three practical problems including the synthesis of cross-coupled resonator filters and diplexers, and computer aided diagnosis of a complex microwave filter. In all cases, superior performance was observed in terms of robustness and speed of optimization in comparison to other available optimization techniques.; In area of EM optimization of microwave filters, we developed a linear model calibration approach in the framework of space mapping technique. Our approach provides solutions to several bottlenecks in the current space mapping formulations including the failure of the method due to the inaccuracy of the approximate model, the nonuniqueness of the parameter extraction procedure and the complicated numerical implementation of the algorithm. It is shown that by using the reflection function alignment in the process of approximate model calibration, the requirement for use of a very accurate approximate model of the filter is eliminated. Since we use a linear formulation of the model calibration, the implementation of the approach in the available EM simulators is very easy as well. Several guidelines are also given for reducing the nonuniqueness of the parameter extraction step. The developed method was successfully used for EM optimization of four microwave filters. The method has performed equally well in all cases including the lossless and lossy filters as well as filters with cascade or cross-coupled topology. In all tested problems, only a few number of EM simulations were required for achieving the optimal solutions while using the simplest type of approximate models.