Data-Centric Energy Efficient Adaptive Sampling Techniques for Wireless Pollution Sensor Networks

Abstract

Air pollution is one of the gravest problems being faced by modern world, and urban traffic emissions are the single major source of air pollution. This work is founded on collaboration with environmental scientists who need fine grained data to enable better understanding of pollutant distribution in urban street canyons. “Wireless sensor networks” can be used to deploy a significant number of sensors within a space as small as a single street canyon and capture simultaneous readings both in the time and space domain. Sensor energy management becomes the most critical constraints of such a solution, because of the energy hungry gas sensors. Hence, the main research objective addressed in this thesis is to propose novel temporal and spatial adaptive sampling techniques for wireless pollution sensor nodes that take into account the pollution data characteristics, and enable the sensor nodes to sample, only when, an important event happens to collect accurate statistics in as efficient a manner as possible. The major contributions of this thesis can be summarised as: 1) Better understanding of underlying pollution data characteristics (based on real datasets collected during pollution trials in Cyprus and India) using techniques from time series analysis and more advanced methods from multi-fractal analysis and nonlinear dynamical systems. 2)Proposal of novel adaptive temporal sampling algorithm called Exponential Double Smoothing based Adaptive Sampling (EDSAS) that exploits the presence of slowly decaying autocorrelations and local linear trends. The algorithm uses a time series prediction method based upon exponential double smoothing for irregularly sampled data. This algorithm has been compared against a random walk based stochastic scheduler called e-Sense and found to give better sampling performance. EDSAS has been extended to the spatial domain by incorporating distributed hierarchical agglomerative clustering mechanism. 3)Proposal of a novel spatial sampling algorithm called Nearest Neighbour based Adaptive Spatial Sampling (NNASS) that exploits the non-linear dynamics existing in pollution data to compute predictability measures to adapt the sampling intervals for the sensor nodes. NNASS has been compared against another spatial sampling algorithm called ASAP and found to give comparable or better sampling performance.