Géolocalisation d'émetteurs en une étape : Algorithmes et performances

The context of the thesis is the transmitters geolocalization (position estimation in the space) of radiocommunication from several widely spaced multi-sensor stations. The conventional geolocalization methods as the tirnagulation are in 2 staps (the first step estimates intermediate parameters and the second step "merges" these mesures from the stations in order to obtain the transmitters positions). The 1 stap methods use the observertions from all the stations to directly and optimally estimate the sources positions. Directly handling the signals on the global array (composed of all the local stations) leads to a broadband effect on the signals between the stations.The thesis proposes to study the residual broadband effect on the global array of the 1 step methods. Then we propose improvements on some 1 step geolocalization methods, especially based on the random matrix theory in large dimension and on the introduction of a new method named LOST-FIND. Finally, a new approach differently tackling the braodband problem has been introduced and leads to TARGET algorithm.