Efficient design and optimisation techniques for planar microwave filters

The topic of this thesis is the efficient and accurate synthesis of arbitrarily shaped microwave filters. Currently, filter circuits are synthesised using an iterative optimisation in conjunction with an electromagnetic (EM) simulator. This so-called direct EM optimisation has two major drawbacks. It does not generally attain the optimal solution, but rather a sub-optimal solution. In addition, direct EM optimisation is computationally very expensive. In this thesis, we propose four novel techniques to overcome these drawbacks.; First, we propose the application of a genetic algorithm (GA) to guide the optimiser towards the optimal solution. GA's identify the optimal filter layout even where gradient-based methods fail. We enhance a generic GA in two ways. Our algorithm allows the user to impose arbitrary constraints. Our GA also provides the user with vital data about the correlation and the sensitivity of the parameters. With these enhancements, our improved GA generally attains better-performing circuits than gradient search methods at similar computational expenses.; Second, we extend the Cauchy method for fast frequency sweeps to a multi-dimensional Cauchy method, with respect to both frequency and physical dimensions. Hence, our new algorithm can be used as a substitute for neural networks. The proposed multi-dimensional Cauchy method accurately and inexpensively computes circuit models from arbitrary topologies. It minimises the computational cost of the optimisation process. Examples demonstrate that---without sacrificing accuracy---the optimisation time is reduced by several orders of magnitude.; Third, for tackling large filter circuits, we introduce a hybrid optimisation scheme based on coupling matrix alteration. This new technique combines an accurate---but expensive---complete electromagnetic analysis with an inexpensive---but less accurate---decomposed analysis. First, we derive two coupling matrices of the filter: one from the response found by complete analysis, and one from the response found by decomposed analysis. We then compare the two coupling matrices and apply this knowledge to eliminate inaccuracies in the decomposed circuit model. Our hybrid technique attains the computational speed of the decomposed analysis, but the accuracy of the full analysis. The technique can optimise filter circuits for which direct EM optimisation would require excessive computer resources.; Fourth, we create a technique for optimising non-resonant filter structures, such as transversal filters. Transversal filters may include up to several hundred design parameters, making direct EM optimisation impossible. An optimisation of such filters becomes feasible using the time-domain response rather than the frequency-domain response. Our technique isolates the fault locations of the circuit one after the other by time-domain reflectometry. Hence, the optimal circuit parameters can be found one at a time. The time-domain optimisation presented in this thesis can optimise a transversal filter without a restrictive upper limit on the number of parameters.; Several examples throughout this thesis illustrate that the proposed techniques result in a considerable gain in design reliability and a significant reduction of computational cost when compared to conventional methods. One chapter is exclusively dedicated to the design and the measurement of various microwave devices for superconductive satellite systems using our new techniques. The measured results are in excellent agreement with the theory.