Reconstrução de sinais em redes de sensores sem fios com técnicas de geoestatí­stica

Abstract

Wireless sensor networks are formed by mobile devices that collect and process data from an enviroment, and transmit them to a data center wich is responsible for taking decisions. This work aims to analyze the signal reconstruction in these networks using geostatistic techniques. Three processes of kriging are used: simple, ordinary and bayesian. Three approaches to simple krigingwere found in the literature, according to the way themean of the data is estimated,were assessed themall. A newBayesian approach is proposed: use general least square to estimate the mean, and set it as a constant into the Bayesian inference. The effect of clustering techniques is assessed, namely without clusters and with clusters formed by LEACH and SKATER algorithms. Bayesian kriging presents the best qualitative results in almost all scenarios, but it is not available to systems that require fast aswers; in this case we recommend ordinary kriging. The proposed variant of Bayesian kriging reduces the time required, without hampering the quality of the reconstructed signal, but the time reduction is not enough for real-time systems