In the last years, a great amount of radio integrated circuits have appeared in the market. The availability of this kind of components has enabled the apparition of a new sort of applications: wireless sensor networks. In these systems, sensors are not located in the very same circuit, but they work as isolated complete autonomous systems. Thus, they can monitor and control physical magnitudes, which are distributed in a large geographical area. However, in order to keep these systems economically feasible, the price of each node (both in for its design and its production) must be kept to a minimum. The most important advantage of these systems is redundancy: if a large amount of sensors take part in the network, the absence of a particular node (e.g. due to some hardware failure) does not have a great impact in the overall performance, because a new node will take charge of its duties, until the problem is solved.

Nevertheless, the cooperation among such a huge amount of nodes is not a simple task: as all sensors in the network must have the same hardware and software and they are all limited in terms of energy consumption and computing power, the protocols used in this kind of networks must be carefully designed for the strong characteristics of this environment. In this case, the cooperation tasks may be simplified greatly if the location of each node and the overall network topology are known. The large amount of nodes makes the manual specification of the position impossible, requiring an automatically location system discovering algorithm, so that each node can calculate its own position.

In this thesis a new algorithm for finding a node's position is proposed and developed. In order to calculate this location, the distances between the node and some beacons are used. Since low cost nodes are used, the quality of these measurements is supposed not to be very high and, therefore, errors will appear. The algorithm described here tries to find the best position for the node, even in the presence of errors in the measurements.

Last, in order to examine the results of this algorithm, a simulation platform has been designed. By using this platform, the algorithm proposed in this thesis has been compared with other algorithms used in the recent years, to analyze what kind of computational load it imposes to the node and how it behaves in the presence of measurement errors. In addition, the algorithm has been implemented in a real node, showing that it is perfectly feasible in the environment, for which it has been designed.