Level based sampling techniques for energy conservation in large scale wireless sensor networks.

摘要

As the size and node density of wireless sensor networks (WSN ) increase, the energy conservation problem becomes more critical and the conventional methods become inadequate. This dissertation addresses two different problems in large scale WSNs where all sensors are involved in monitoring, but the traditional practice of periodic transmissions of observations from all sensors would drain excessive amount of energy.;In the first problem, monitoring of the spatial distribution of a two dimensional correlated signal is considered using a large scale WSN . It is assumed that sensor observations are heavily affected by noise. We present an approach that is based on detecting contour lines of the signal distribution to estimate the spatial distribution of the signal without involving all sensors in the network. Energy efficient algorithms are proposed for detecting and tracking the temporal variation of the contours. Optimal contour levels that minimize the estimation error and a practical approach for selection of contour levels are explored. Performance of the proposed algorithm is explored with different types of contour levels and detection parameters.;In the second problem, a WSN is considered that performs health monitoring of equipment from a power substation. The monitoring applications require transmissions of sensor observations from all sensor nodes on a regular basis to the base station, which is very costly in terms of communication cost. To address this problem, an efficient sampling technique using level-crossings (LCS) is proposed. This technique saves communication cost by suppressing transmissions of data samples that do not convey much information. The performance and cost of LCS for several different level-selection schemes are investigated. The number of required levels and the maximum sampling period for practical implementation of LCS are studied. Finally, in an experimental implementation of LCS with MICAz mote, the performance and cost of LCS for temperature sensing with uniform, logarithmic and a combined version of uniform and logarithmically spaced levels are compared with that using periodic sampling.