Localization applying an efficient neural network mapping

Node location information is essential for many applications in Autonomic Computing. This paper presents and evaluates a new cooperative node localization scheme. We apply an efficient nonlinear data mapping technique, the Curvilinear Component Analysis (CCA), to produce accurate node position estimates employing only a small number of anchor nodes. Being a light-weight neural network, CCA has the learning ability to self-organize maps of nodes, and to project node coordinates with improved accuracy and efficiency. We present the distributed CCA-MAP scheme that derives node locations in either range-based or range-free scenarios. Unlike other schemes, no further refinement is needed to improve the position estimates generated by the devised CCA projection method. Through extensive simulation studies, we evaluate the performance of our scheme for both regular and irregular networks of different configurations. Comparisons with other related localization schemes are also presented, demonstrating the improved location estimate accuracy and performance efficiency.