Topology recognition in Wireless Sensor Networks

Using self-organizing wireless sensor networks for object tracking requires ordering events with regard to time and location. In labyrinth-shaped topologies, one-dimensional ordering suffices within the different parts of the network. We present an algorithm to derive this ordering without the use of location information. Local order knowledge in the nodes can then be additionally used to detect junctions. INTRODUCTION An important aspect of wireless sensor networks is the logical partitioning into sub-networks. This aspect is fundamental for a lot of location-aware applications. Furthermore, applications often require an ordering of the nodes with respect to their environment, e.g. for object tracking. Computation of such location information can either be done at a central point or by the nodes themselves. The latter is beneficial in order to reduce the amount of transmitted data. Since the energy consumption decreases with decreasing amount of data transmissions, a localized algorithm helps to increase the overall lifetime of the network [1, 5]. We assume that the wireless sensor network consists of a topological structure resembling road networks or corridors in buildings and that the devices are equipped with motion detectors. In [4] a method for topology recognition is described, particularly with regard to border and junction detection, but while this paper deals with extreme high densities, we assume rather sparse networks. The problem requires detecting the one-dimensional spatial ordering of neighbors within the communication range. Since passing objects cause ordered chains of motion detection events, movement directions can be inferred by mapping object detection events to this spatial order of the nodes. OUR APPROACH This section describes our localized algorithm to detect the one-dimensional ordering of a node and its neighbors. Let N be the node computing its local topology, i.e. the ordering of its neighbors. Our algorithm comprises 3 steps: 1. Detection of border nodes, i.e. far away nodes near the maximum communication distance 2. Computation of chains from border nodes to N 3. Mapping the remaining nodes onto the path nodes Note that N only needs information of its direct neighbors, which minimizes the communication costs. Figure 1 shows exemplarily the three steps of the algorithm.