Abstract
Towards realizing the fifth generation (5G) of wireless networks, the Internet of Things (IoT), and the Tactile Internet, intelligent communications and computing is key part of the technological stack. The next generation of wireless networks are expected to be characterized by limited availability of resources, thus, in this dissertation, we tackle the problem of efficient allocation of several types of communications and computing resources, while achieving high quality of service and experience for the devices or the users. Considering the interdependence of the devices while trying to access and share common resources as well as their increasing intelligence which enables them to make choices on supporting self-beneficial properties, it seems natural to adopt more user-centric approaches leading to decentralized solutions. In this PhD dissertation, we considered designing decision making frameworks where devices take advantage of the network's capabilities in order to reduce their ...
Towards realizing the fifth generation (5G) of wireless networks, the Internet of Things (IoT), and the Tactile Internet, intelligent communications and computing is key part of the technological stack. The next generation of wireless networks are expected to be characterized by limited availability of resources, thus, in this dissertation, we tackle the problem of efficient allocation of several types of communications and computing resources, while achieving high quality of service and experience for the devices or the users. Considering the interdependence of the devices while trying to access and share common resources as well as their increasing intelligence which enables them to make choices on supporting self-beneficial properties, it seems natural to adopt more user-centric approaches leading to decentralized solutions. In this PhD dissertation, we considered designing decision making frameworks where devices take advantage of the network's capabilities in order to reduce their resource consumption and more effectively perform their assigned tasks. First, the prolongation of battery life of mobile machine-to-machine (M2M) devices is considered, in order to guarantee and sustain the operation of the IoT system for a longer period of time, while taking into account the management of information of the same nature in a more efficient way, by focusing on the use of social properties and characteristics of the devices. For that reason, a joint interest, physical and energy-aware cluster formation mechanism is proposed so that devices are effectively grouped and a high energy clusterhead can be assigned for each cluster. The clusterhead is then responsible to provide to the rest of the devices enough power to send their data via wireless energy transfer (WET), collect all the information from the devices on its cluster and forward the information to the eNB for further processing. Then, a setting of Multi-access Edge Computing (MEC) is discussed, where servers offer computing resources at the edge of the network to mobile end-users. A multi user - multi MEC server environment is considered where users are willing to offload some of their computational tasks and the MEC servers are setting a price in order to process them. The user is able to chose the server to offload the data to, as well as determine the portion of the task that will be offloaded, while the server will set the price it will charge for each task. In order to achieve the best server selection, a reinforcement learning framework based on stochastic learning automata is adopted, while the amount of data offloading is determined via a non-cooperative game among users, and the optimal announced prices are determined via an optimization problem. The information exchange between the users and the MEC servers until the final offloading decision, is handled and realized by a Software Defined Network (SDN) controller. In the rest of the dissertation, we introduced the concept of users' behavioural characteristics in order to capture and reflect the fact that users do not act as neutral maximizers but instead exhibit risk-aware behaviour. A MEC setting is considered as well, where multiple user devices are willing to offload their tasks to a MEC server responsible for handling them. Under this setting, the MEC server is considered as a Fragile Common Pool Resource (CPR), meaning that the more the server is used, the higher the probability of failure to execute its assigned tasks, resulting in losses for the users. The problem is modeled as a non-cooperative game between the users, where each user should choose the portion of the tasks to be offloaded to the server by selecting the amount of data to send. Towards capturing the users' behavioral characteristics in the data offloading decision-making process, we adopt the principles of Prospect Theory in order to model the users' decisions under risk and uncertainty of outcome. Additionally, a usage based pricing policy is considered to balance the usage of the MEC server by the users, since the additional cost prohibits users to over-exploit the servers' resources and thus reduces the Probability of Failure (PoF) of the server. Finally, we extended the aforementioned concept on a multi-user multi-server environment where the additional problems of users' server selection and MEC servers' price selection arise. In order to more holistically address the users' decision-making process, we considered the server selection and the amount of offloading data selection as a joint optimization problem, allowing users to choose the combination that maximizes their perceived utility. In order to tackle the MEC servers' price selection problem we proposed two different approaches, a game-theoretic approach and a reinforcement learning one, considering different information availability scenarios on the system. The overall framework is modeled as a Stackelberg game where the servers are considered leaders, making their pricing decisions based on one of the proposed approaches, and the users are considered followers, making their data offloading decisions based on the prospect theoretic principles.
show more
