Genetic design of antennas and electronic circuits

This dissertation explores the use of genetic optimization for both antenna and circuit designs. Chapters 2 and 3 focus on parametric optimization of Yagi-Uda antenna arrays. In Chapter 2, Yagi-Uda arrays optimized with traditional binary genetic algorithms were found to perform better than gradient optimized arrays previously published in the literature. In Chapter 3, several different types of genetic algorithms and one gradient method were used to optimize a Yagi-Uda array for operation over a band of frequencies. The most important result from this chapter demonstrated that combining gradient and genetic optimization methods within the same process produces excellent results in an efficient manner.; Chapters 4 and 5 introduce genetic design, which uses a formally defined design language in conjunction with genetic algorithms to optimize both the structure and the numerical parameters of a design. The design language includes a grammar that describes how building blocks are connected together to form a final design solution. This grammar can be either be very vague or very specific depending on the purpose of the designer. With a vague grammar, the genetic algorithm searches a very large design space, and occasionally finds an unexpected solution to a design problem. If the vague grammar fails to find a solution, incorporating knowledge about the problem into the grammar narrows the search to a region of the design space expected to yield good results. This results in more conventional design solutions that usually perform reasonably well.; The antenna design example in Chapter 4 compared three different design languages for the design a 40% bandwidth antenna array composed of up to 10 array elements. Two out of three languages produced arrays that performed better than a traditional 10-element log-periodic structure optimized for the same problem.; In Chapter 5, genetic design was applied to analog circuit design. The method outlined here was able to optimize a low-pass filter design in less than 10 minutes on a single CPU. This IS 4 orders of magnitude faster than genetic programming methods used on the same problem In the past. Design grammars are primarily responsible for this significant increase in efficiency. The simulation method, which integrated two-port network theory with the circuit chromosome and only had to recalculate small portions of a circuit as it evolved from generation to generation, also played a role.