Dynamic Radio Access Selection and Slice Allocation for Differentiated Traffic Management Beyond 5G Networks

Beyond 5G (B5G) claims a three-dimensional (3D) ecosystem with cooperation between terrestrial networks (TNs) and non-terrestrial networks (NTNs) to achieve seamless coverage, improve capacity, and enable cutting-edge applications with strict quality of service (QoS) and quality of experience (QoE) requirements. This complex environment requires a disaggregated radio access network (RAN) deployment with open interfaces, such as the architecture promoted by the Open-RAN (O-RAN) Alliance. This framework, supporting the slicing paradigm, is a prominent solution to guarantee dynamism and differentiated traffic management. Furthermore, intelligence is critical for future wireless networks to enable machine learning (ML)-based optimization for autonomous RANs, handling ultra-dense heterogeneous environments, and adapting to numerous scenarios. Multiple types of end devices (EDs) with different mobility behaviors and tariff plans compete for finite resources, making critical an effective strategy to satisfy all requests with at least the minimum QoS requirements according to the service and user profiles. In such a context, each ED must be associated with the most suitable base station (BS) and network slice (NS) in his/her service area, ensuring the always best-connected (ABC) paradigm and seamless connectivity through the effective TNs-NTNs integration. Bearing the previous explanation, this Ph.D. study aims to dynamically manage RAN selection and slice allocation over the envisioned B5G heterogeneous environment. First, the thesis presents an overview of the research umbrella and the motivation, specifying the objective and contributions in Chapter 1. Then, Chapter 2 overviews the related works and main theoretical concepts, and Chapter 3 establishes the system model and problem formulation. The core content regarding the objectives, contributions, and associated publications are detailed in Chapters 4 to 8. Chapters 4 and 5 present heuristic algorithms based on multi-attribute decision-making (MADM), whereas Chapters 6 and 7 propose federated deep reinforcement learning (F-DRL) solutions inserted into the O-RAN framework. These proposals dynamically handle the RAN selection and slice allocation over the envisioned B5G environment to satisfy multiple service requests anywhere and anytime. The presented solutions aim to maximize the long-term QoS of all users in the network and optimize the slicing resource utilization based on the defined service level agreement (SLA). In case of overloading, a cooperative game theory (CGT) strategy is applied to adjust resources on demand and avoid abrupt QoS degradation. Additionally, Chapters 5 and 7 analyze the importance of exploiting multicast/broadcast services (MBS) capabilities over a softwarized framework to increase network capacity and avoid congestion. Finally, Chapter 8 proposes a QoE-based energy-aware radio resource allocation process over 5G ultra-dense heterogeneous networks (HetNets). We analyze the trade-off between user perception and electricity consumption, focusing on sustainability. The proposed solutions handle diverse network conditions, service requests, different types of EDs, mobility patterns, and priorities. Along the Ph.D. study, we conduct several simulations to prove the dynamic adaptation of our proposals in B5G heterogeneous systems while guaranteeing efficient differentiated traffic management. We demonstrate that the ML algorithm behaves similarly to the heuristic proposal, ensuring an efficient learning process based on multiple interactions with the environment. Moreover, we prove the importance of F-DRL solutions to reduce communication overhead, computational complexity (CC), and enhance data privacy, taking advantage of disaggregated and flexible architectures such as O-RAN. On the other hand, we evidence the necessity to save network resources and reduce energy consumption while maintaining satisfactory levels of QoE.