Síntesis de antenas mediante Reconstrucción de Fuentes y Redes Neuronales Artificiales

Abstract

This Ph.D. Thesis studies and develops radiation pattern synthesis methods for antenna arrays making use of artificial neural networks. The work is divided in two related blocks: the use of neural networks as an alternative to array factor synthesis methods or to synthesis methods based on reconstruction of the current distribution in the plane of the antenna; and the use of neural networks to develop more realistic models taking into account interaction between elements and the radiating properties of the basic antennas in the array in order to calculate the voltages that must be applied to the feeding ports. The first part presents neural networks as a competitive and efficient synthesis tool able to transfer the computational cost of the process to a previous training scheme, working then in real time with a minimal complexity. In the same way, different methodologies for different pattern synthesis-related problems are developed making use of neural networks. Equivalent source reconstruction and near-field to far-field transformation are also considered. All these tools are presented with significant results. This first part is extended to present a set of synthesis-tools, also based on the use of neural networks, that make use of their abilities to model the physical effects of coupling between elements, taking into account their radiating properties in order to calculate the voltages that must be applied to the ports of the array. These tools contrast with the traditional synthesis methods, based on an approximation of equal radiating characteristics of the elements assuming a certain error in the results. Some numerical results of significant examples illustrate the proposed techniques. Measurement results of a prototype designed with those techniques are also presented. Finally, real radiating systems modelling is extended including the presence of conducting obstacles in a near environment, in such way that it is posible to perform a pattern synthesis taking into account their effect in a realistic way.