Reperage de la direction d'arrivee d'un faisceau par rapport a un reseau d'antennes disposees sur deux axes et a l'aide d'un reseau de neurones

This thesis presents a study that aims to create a system to locate the direction of arrival of a beam for angles of arrival contained in a cone whose vertex angle is 90°. We have conceived this original and simple system that gives a response with an error of less than 4° in order to embed it on a Micro satellite of the Canadian Space Agency. This system can be divided into three parts: an antenna array which receives the incidental beam,a radio frequency combiner which combines the powers received by the various antennas and a neural network which finds out the direction of arrival of the beam from the outputs of the combiner.; This report explains how we have optimized both the architecture of the antenna array and the characteristics of the combiner in order to create signals which can be interpreted by the neural network in the best conditions. Nevertheless, the core of this work resides in the design and the optimization of the neural networks.; Our methodology was always leaded by the curse of simplicity as well as performance. Indeed, an antenna array was finally designed under qualitative criteria. The principal contributions of our work are as follows. (1) We proposed a daring method but still efficient to create the architecture of the antenna array. (2) We defined the criteria to take into account to choose and create a filter which increases the signal to noise ratio. (3) We have optimized and trained neural networks to give the angle of arrival and to assure that the signal was really in the cone of study. (4) We tested the system under real conditions with experimental dataset coming from our real antenna array. (Abstract shortened by UMI.)...