Evolutionary Optimization of Microwave Filters
Maria José Pereira Dantas, S. Adson, Ciro J. A. Macedo, Leonardo da Cunha Brito, P.C.M. Machado, Paulo H. P. de Carvalho · InTech eBooks · 2011
The optimization of circuits with the aim of improving performance, lowering costs, and more recently, to reduce the size and weight of electronic devices, among other objectives, has become a research field in the areas of mathematics and engineering around the world (Antoniou & Lu, 2007). The problem can be interpreted as: among all possible models for the circuit, considering topology and values of components, find a model with minimum size and appropriate parameters, able to meet a set of typically conflicting hard specifications. The deterministic methods are not effective in the design of new circuits, since it does not always have available accurate mathematical models of the processes to be optimized (Levy et al., 2001). However, a number of new stochastic algorithms have allowed the development of nearly optimal circuits, which are quite acceptable from the standpoint of their implementations. The techniques currently used require much prior knowledge about the circuit to be optimized and this knowledge is not always available. The challenge, then, is the development of robust techniques that rely less in specialist knowledge or even incorporating this knowledge, and show efficiency equal or superior to the traditional methods. In general, due to the complexity of the problems associated with different types of circuit designs, sub-optimal solutions are looked for in order to mitigate the mathematical complexity of the analytical models, generating approximate solutions, but with acceptable results (Levy et al., 2001). Evolutionary methods, so called because they use principles of evolution found in nature, are natural candidates to this task, since they are able to find optimal solutions or solutions near the optimum, using mechanisms of selection, crossover and mutation (Holland, 1970). Most methods in the literature consider an initial topology for the circuit and only optimize its parameters, such as, for example, in (Hsu & Huang, 2005). However, it is desirable that the topological structure also goes under optimization, allowing, among other things, the search for new topologies, which is a requirement of current applications. Other methods, using the traditional Genetic Programming as in the work of Koza (Koza et al., 1996), optimize the topology, without considering any prior knowledge, but demand a high computational cost and also have some methodological drawbacks such as premature convergence and stagnation, among others (Hu & Rosemberg, 2004).