GA optimization of terminal antennas by the estimation of the population density of probability using dependency trees

The paper presents an application of a GA algorithm to the optimization of terminal antennas, where the number of variables to optimize and the complexity of the problem could make standard GA approaches fail. This GA algorithm is based on the estimation of the density of probability of the highest fitness chromosomes in the population. This estimation is achieved by the use of dependency trees whose structure varies dynamically along the optimization process. A general overview of Bayesian networks and probability theory is presented. The algorithm based on dependency trees (TREE) is presented with some examples and compared to standard GA with dual population (DUAL) and with linkage crossover operator based GA (GLINX). The structure optimised is an antenna covering three frequency bands (GSM, DCS and UMTS), with one feed port for the two lower bands and another for the upper band. Convergence curves are presented for the three algorithms.