Building a green connected future: smart (Internet of) Things for smart networks

The vision of Internet of Things (IoT) promises to reshape society by creating a future where we will be surrounded by a smart environment that is constantly aware of the users and has the ability to adapt to any changes. In the IoT, a huge variety of smart devices is interconnected to form a network of distributed agents that continuously share and process information. This communication paradigm has been recognized as one of the key enablers of the rapidly emerging applications that make up the fabric of the IoT. These networks, often called wireless sensor networks (WSNs), are characterized by the low cost of their components, their pervasive connectivity, and their self-organization features, which allow them to cooperate with other IoT elements to create large-scale heterogeneous information systems. However, a number of considerable challenges is arising when considering the design of large-scale WSNs. In particular, these networks are made up by embedded devices that suffer from severe power constraints and limited resources. The advent of low-power sensor nodes coupled with intelligent software and hardware technologies has led to the era of green wireless networks. From the hardware perspective, green sensor nodes are endowed with energy scavenging capabilities to overcome energy-related limitations. They are also endowed with low-power triggering techniques, i.e., wake-up radios, to eliminate idle listening-induced communication costs. Green wireless networks are considered a fundamental vehicle for enabling all those critical IoT applications where devices, for different reasons, do not carry batteries, and that therefore only harvest energy and store it for future use. These networks are considered to have the potential of infinite lifetime since they do not depend on batteries, or on any other limited power sources. Wake-up radios, coupled with energy provisioning techniques, further assist on overcoming the physical constraints of traditional WSNs. In addition, they are particularly important in green WSNs scenarios in which it is difficult to achieve energy neutrality due to limited harvesting rates. In this PhD thesis we set to investigate how different data forwarding mechanisms can make the most of these green wireless networks-enabling technologies, namely, energy harvesting and wake-up radios. Specifically, we present a number of cross-layer routing approaches with different forwarding design choices and study their consequences on network performance. Among the most promising protocol design techniques, the past decade has shown the increasingly intensive adoption of techniques based on various forms of machine learning to increase and optimize the performance of WSNs. However, learning techniques can suffer from high computational costs as nodes drain a considerable percentage of their energy budget to run sophisticated software procedures, predict accurate information and determine optimal decision. This thesis addresses also the problem of local computational requirements of learning-based data forwarding strategies by investigating their impact on the performance of the network. Results indicate that local computation can be a major source of energy consumption; it’s impact on network performance should not be neglected.