Cognitive Solutions for Resource Management in Wireless Sensor Networks

Abstract

Wireless Sensor Networks (WSN) is an important technology that can be used to provide new data sets for many applications ranging from healthcare monitoring to military surveillance. Due to the increasing popularity of WSNs, user demands have evolved as well. To achieve the end-to-end goals and requirements of the applications, managing the resources of the network becomes a critical task. Cognitive networking techniques for resource management have been proposed in recent years to provide performance gains over traditional design methodologies. However, even though several tools have been considered in cognitive network design, they show limitations in their adaptability, complexity, and their ability to consider multiple conflicting goals. Thus, this thesis proposes novel cognitive solutions for WSNs that include a reasoning machine and a learning protocol. Weighted Cognitive Maps (WCM) and Q-Learning are identified as suitable tools for addressing the aforementioned challenges and designing the cognitive solutions due to their ability to consider conflicting objectives with low complexity. WCM is a mathematical tool that has powerful inference capabilities. Thus, they are used to design a reasoning machine for WSNs. Two case studies are proposed in this thesis that illustrate the capabilities of WCMs and their flexibility in supporting different application requirements and network types. In addition, an elaborate theoretical model based on Markov Chains (MC) is proposed to analyze the operation of the WCM system. Extensive computer simulations and analytical results show the ability of the WCM system to achieve the end-to-end goals of the network and find compromises between conflicting constraints. On the other hand, Q-Learning is a well known reinforcement learning algorithm that is used to evaluate the actions taken by an agent over time. Thus, it is used to design a learning protocol that improves the performance of the WCM system. Furthermore, to ensure that the learning protocol operates efficiently, methods for improving the learning speed and achieving distributed learning across multiple nodes are proposed as well. Extensive computer simulations show that the learning protocol improves the performance of the WCM system in several metrics.