Abstract
A plethora of diverse services within every vertical segment is rapidly emerging within the context of the Internet of Things (IoT) wireless networks. This thesis models and analyzes large-scale wireless IoT networks from the communication and computation views. Rigorous frameworks are developed and validated by extensive simulations for different traffic models from a spatiotemporal viewpoint. For the computation aspect, multi-access edge computing is utilized to enable efficient offloaded task execution at the serving base stations. Latency reduction gains are demonstrated through different use-cases within the automotive and the industrial vertical segments.
Over the past decades, the features and functionalities of large-scale IoT networks have become more complex, which calls for a new way of thinking via designing the wireless networks from a combined communication and computation perspective. This thesis discusses the need to rigorously study the spatiotemporal dynamics of large scale IoT networks for diverse requirements, deployments and use cases.
For the communication pillar, analytical models are presented to model wireless networks while considering random base stations deployment and dynamic temporal traffic models. Different analytical frameworks are proposed that entail the macroscopic and microscopic scales of large-scale wireless-networks as highlighted in Figure 1.
The microscopic scale is viewed per device level and addresses the temporal dynamics at each device, whereas the macroscopic scale representsa holistic view of the network that captures the mutual interaction (i.e., mutual interference and contention among the resources) among the devices (i.e., queues). Hence, both scales can be regarded as, respectively, a zoom in that shows the behavior of each device and a zoom out that shows the behavior of the entire network. In this regard, prioritized multi-stream traffic is investigated and a systematic and tractable scheme is presented. Specifically, dedicated and shared channel priority-aware access strategies are studied, and bench-marked against a priority-agnostic scheme. To characterize the information freshness in large-scale uplink networks, a spatiotemporal framework for the time and event-triggered traffic is proposed.
For the computation pillar, because of the massive number of running services, demanding computation processes within the network are inevitable. One solution is to let such computations be executed at a remote data center. However, such approach is not only inefficient due to bandwidth constraints, but also hinders the performance of time-sensitive and location-dependent applications due to the imposed network delay. Consequently, we investigate throughout this thesis the advents of Multi-access Edge Computing (MEC) deployments in large-scale IoT networks, focusing on the latency and task execution efficiency in task-offloading use cases as shown in Figure 2.
In such networks, it is expected that devices utilize MEC computation processing capabilities to properly address the demanding tasks generated locally. In this regard, a computation-oriented association criterion is proposed to exploit jointly the available communication and the computation resources within the network. When it comes to safety-critical use cases within the automotive vertical, we showcase that in contrast to conventional remote cloud-based networks, MEC deployment can assist in pruning the experienced end-to-end latency and the peak age of information.
In summary, this thesis provides a unified insight on modeling, designing, and assessing future wireless large-scale IoT networks, utilizing different communication and computation key enablers.
