TASC: topology adaptive spatial clustering for sensor networks

The ability to extract topological regularity out of large randomly deployed sensor networks holds the promise to maximally leverage correlation for data aggregation and also to assist with sensor localization and hierarchy creation. This paper focuses on extracting such regular structures from physical topology through the development of a distributed clustering scheme. The topology adaptive spatial clustering (TASC) algorithm presented here is a distributed algorithm that partitions the network into a set of locally isotropic, non-overlapping clusters without prior knowledge of the number of clusters, cluster size and node coordinates. This is achieved by deriving a set of weights that encode distance measurements, connectivity and density information within the locality of each node. The derived weights form the terrain for holding a coordinated leader election in which each node selects the node closer to the center of mass of its neighborhood to become its leader. The clustering algorithm also employs a dynamic density reachability criterion that groups nodes according to their neighborhood's density properties. Our simulation results show that the proposed algorithm can trace locally isotropic structures in non-isotropic network and cluster the network with respect to local density attributes. We also found out that TASC exhibits consistent behavior in the presence of moderate measurement noise levels.