COGNITION IN LARGE SCALE INFORMATION-CENTRIC SENSOR NETWORKS: NOVEL DEPLOYMENT AND DATA DELIVERY SOLUTIONS

Abstract

Smart Cities, enabled by Wireless Sensor Networks (WSNs), have emerged as one of the most promising applications of the Internet of Things (IoT). These Smart City environments require that the underlying sensor network infrastructure be enriched with smart devices, so that the network can understand and respond to requests from multiple users with diverse information requirements. Now, the use of artificial intelligence has enabled some amount of user-requirement awareness in sensor networks. However, there is no architectural framework around how cognition is incorporated in the network, or where the smart decision making is implemented. In addition, WSN implementations are mostly address-centric, where users must specify the location from where data must be gathered. But this is counter-intuitive to how users would like to access information in a smart city environment, especially in applications such as smart parking, where the network needs to provide the location of free parking spots to the user. Moreover, managing the large IP address space becomes problematic as the network size expands to vast counts of sensor nodes. In this thesis, we propose an architectural framework called Cognitive Information Centric Sensor Network (CICSN), to introduce cognition in WSNs. Cognitive nodes, capable of knowledge representation, learning, and reasoning, along with an information-centric approach to data delivery, are central to the idea of the CICSN. We propose a deployment strategy for cognitive nodes in the network such that connectivity of sensor nodes with the sink is maintained, and the number of cognitive nodes is minimized as well. Knowledge representation is done using attribute-value pairs. In addition, a Quality of Information (QoI) aware data delivery strategy, with Analytic Hierarchy Process (AHP) as the reasoning technique, is used to identify data delivery paths that dynamically adapt to changing network conditions and user requirements. Latency, reliability, and throughput are the attributes used to identify the QoI along the delivery path. Further, heuristic learning techniques are explored to improve the success rate of data delivery to the sink. Simulation results show that the proposed architecture significantly improves the QoI as well as the success rate of data delivered by the network.