Map-based Adaptive Positioning in Wireless Sensor Networks

Abstract

Frequent localization in sensor networks may be needed due to the dynamically changing topology and the possible mobility of sensor nodes. We present a distributed adaptive localization method that we refer to as: map-based adaptive positioning (MAP). The main idea is to construct a relative local map at every node in the network, consisting of the node itself and its immediate neighbors, and merge the local maps together to form a global map. We consider two algorithms that can be used to estimate the relative local maps: multidimensional scaling (MDS) and semidefinite programming (SDP). The performance of these algorithms depend on two parameters: size of a local map, i.e., number of nodes, and the average connectivity of the node at the center of the local map and its 1-hop neighbors. We use machine learning to adaptively select the appropriate algorithm to estimate the relative local maps. Simulation results show that MAP outperforms both MDS and SDP, with better improvement for networks with less uniform node deployment.