Design and analysis of dynamic reconfiguration mechanisms for next-generation mobile networks

Resumo

The 5G standards bring a number of improvements to the physical layer, with the primary aim of significantly improving performance, particularly in terms of data rate and latency. However, newly conceived use cases, such as autonomous vehicles, immersive augmented reality, remote healthcare and large-scale Internet of things (IoT) networks, impose very stringent quality of service (QoS) requirements, close to the theoretical peak values supported by the 5G technology. As a result, 5G networks cannot support all these use cases with a universal configuration, but must adapt their operation to the dynamic user loads and requirements. The aim of this thesis is to propose and evaluate different solutions for providing better QoS based on network reconfiguration, aiming at enabling mobile networks to better support emerging use cases. These solutions cover both the radio access network (RAN) and the core network and fall under the following fields: dynamic adaptation of radio resources to network conditions, performance evaluation mechanisms for validating service level agreement (SLA) compliance in emerging use cases, and characterization and analysis of new use cases and paradigms. Within the topic of dynamic adaptation of radio resources to network conditions, a novel solution for reallocating wireless services to radio devices in SDR infrastructures is proposed. The solution allows operators or infrastructure providers to dynamically adjust the radio services and the density of base stations in operation at any time, with very little downtime, adapting the radio infrastructure in real time to the loads and requirements of the end users. Moreover, it enables operators to make better use of radio resources in shared infrastructures, resulting in improved performance. To complement this solution, two mechanisms are proposed to allow coordinated operation of Wi-Fi and 5G or LTE licence assisted access (LAA) carriers within the same frequencies and space. These mechanisms use protocols defined in the 802.11 standards to postpone Wi-Fi transmissions to predefined intervals or to adjust the bandwidth of the Wi-Fi basic service set (BSS) while keeping terminals informed. With these solutions, mobile and Wi-Fi networks can operate orthogonally. This coordinated operation not only allows the operator to define the amount of resources allocated to each technology, but also achieves higher average data rates than direct coexistence. With regard to performance evaluation mechanisms for validating SLA compliance in emerging use cases, this thesis describes a framework for measuring one-way latency in mobile networks. It aims to characterise the performance of the network in support of critical services, such as actuator networks, where the main issue is the time required to perform a transmission, irrespective of the time taken to receive the response. In the field of characterisation and analysis of new use cases and paradigms, work has been carried out on characterizing and proposing solutions for massive IoT networks and multi-access edge computing (MEC) use cases. A new framework is proposed for protecting next-generation mobile networks from distributed denial of service (DDoS) attacks, in anticipation of the large number of connected devices that these networks will need to support in IoT use cases. The solution relies on software defined networking (SDN) to detect devices that are behaving suspiciously in a lightweight manner and move their traffic to an isolated network segment for in-depth traffic analysis. On the other hand, this thesis analyzes the impact of MEC on the performance perceived by end users consuming services that require computation offloading. This study measures the performance of a machine learning (ML) application offloaded from a drone to MEC and cloud computing platforms. Evaluations were conducted using real-world private and commercial 5G networks, as well as a Wi-Fi setup. In addition, in order to assess the need for computational offloading, the performance of the application was also measured using on-board computing with different specialized low-power boards, comparing the impact of each solution on the flight time of the drone. Finally, this thesis introduces a novel framework to address the problem of allocating MEC resources to mobile users to meet their SLAs, taking into account that end users are not static and the computational load of MEC servers can change at any time. This solution relies on named data networking (NDN) to build a hierarchy of MEC servers based on their location, allowing the user to interact with any server close to the one currently processing their request if the SLA is not met. Unlike other proposals in the literature, this mechanism places the responsibility on the terminal to measure the performance of the application during offloading and to determine the new MEC server to which subsequent requests should be redirected.