Studies of networking algorithms in wireless sensor networks
Date of Issue2015
School of Electrical and Electronic Engineering
In traditional wireless sensor networks (WSNs), sensors powered by battery are largely deployed in a field to observe certain physical phenomena, such as temperature, noise and many more. The purpose of WSN is to be able to sense from environment and detect events for a long time. By improving the network lifetime, WSNs are able to detect events longer. This agrees with the nature of WSNs, which is sense-to-detect events through long operations (hopefully) of batteries. However, limited battery capacity and low event detection probability have always been imposing serious challenges and constraints to network applications, stimulating the developments of various new technologies. Existing works on maximizing lifetime include networking clustering technologies which let most sensors communicate with the nearby cluster head (CH) rather than the remote base station; and energy-harvesting (EH) sensors which allow battery recharging by collecting energy from surrounding environment, etc. Existing works on maximizing event detection probability study the case of deterministic and known sensors deployment, where sensors transmit their local observations or decisions over perfectly known channels to a Gateway, which collects the observations and fuses them in order to perform event detection. In this thesis, we propose a centralized approach to finding the optimal locations for cluster heads and energy harvesting nodes, where we consider nodes powered by ambient energy harvesting as dedicated relay nodes for cluster heads, and propose joint clustering and relay node placement algorithms for network lifetime maximization. In addition, we also propose a distributed EH-CH matching algorithm with given locations for CHs and EHs. To achieve good EH-CH matching in a distributed manner, we borrow ideas adopted in virtual-output-queued (VOQ) networks to develop a simple matching scheme. For the problem of event detection, we develop the first reported solution for wireless sensor network under random deployment. We consider the practical case where the spatial distribution of the sensors follows a Finite Binomial Point Process (FBPP) and Infinite Poisson Point Process (IPPP) for both homogeneous and inhomogeneous cases. In addition, we present novel distributed event detection algorithms based on a statistical approach that tolerates Byzantine attacks and nonlinear transmission schemes where malicious (compromised) or nonlinear sensors send false sensing data to the gateway leading to increased false alarm rate using Moment Matching and Series Expansion methods. We provide analytical and extensive simulation results to demonstrate the optimality and convergence of all these algorithms, and therefore help provide useful benchmarks for various centralized and distributed scheme designs.
DRNTU::Engineering::Electrical and electronic engineering::Wireless communication systems