Dealing with uncertainty and conflicting information in heterogeneous wireless networks

Inspired by challenges of multi-constraint path selection and the need for providing a desired QoS, this dissertation focuses on devising an efficient network selection algorithm that satisfies multiple user constraints with uncertainty in a heterogeneous wireless network (HWN), while under imprecise and dynamic network conditions. We start by determining the impact of the partial network knowledge on the optimal solution. We introduce a Dynamic Programming (DP) solution approach to the routing problem using a well established routing metric. We then compare the impact of using a more realistic scenario with stochastic metrics and formulate an approximate optimal strategy for routing between mobile devices (MD). A fuzzy logic model is then proposed which aims at translating the uncertainty of the network conditions to accurate values. We perform a thorough analysis of the metric values offered by various wireless technologies, and derive crisp values for imprecise network parameters. A sensitivity analysis is performed that reflects the performance and relative importance of the metrics on each network. These results are shown to impact user's decision in handing data over to an appropriate interface. While earlier works focused on multi-constrained routing or handover decision in a HWN, we consider dynamically changing network conditions. This is expected in a realistic deployment where a user is uncertain about what exactly is required under a given circumstance, indicates their preference in vague terms, and expects multiple deployments, with scenarios that are prone to failures, reliability strategies are considered in order to try to determine when to stop retransmitting a message in order to ensure proper delivery while still being energy efficient. A simple effective link-attribute estimator is presented that is capable of identifying the quality of communication between neighboring mobile devices while maintaining scalability. By relying on this link-quality estimator, a maximum number of attempts is computed which (probabilistically) ensures delivery while maintaining an energy-efficient network. Simulations show that our estimator maintains acceptable message delivery ratio while increasing the overall energy efficiency. Finally, a study is made regarding dealing with conflicting information, and how devices can cope with data that may overlap or even conflict with each other through a localized protocol. Specifically, this dissertation looks at the impact of transient and permanent failures on the accuracy of decision making. The behavior of a wireless network is analyzed with respect to the detection of an event by increasing the number of failures. We compare four different schemes: simple majority between neighboring MDs, a more adaptive reputation-based protocol, fuzzy logic to quantify the MDs' uncertainty and a combination of fuzzy logic and the Transferable Belief Model (TBM) framework. Through simulations, we show that our proposed TBM-based solution has the lowest number of incorrect decisions, even when used in deciding and detecting anomalies under an extremely large percentage of faulty MDs.