Intelligent and Dynamic Spectrum Management for Beyond 5G

The 5G network introduces virtualisation technology and network slicing, that allows an agile creation and (re)configuration of multiple network slices on the same infrastructure that can be independently managed. This provides the impetus for new business models and third party operators, that support private networks targeting verticals such as manufacturing industry, warehouse and logistics, mining and maritime ports, among others. This requires the optimised management of radio spectrum, where traditional rule-based techniques are facing serious challenges – they can hardly consider all network and slice combinations in 5G in order to fully exploit all the available degrees of freedom. To address this challenge, this thesis proposes an intelligent framework to automate spectrum management and adapt to the dynamically varying network conditions.<br/><br/>The proposed concept is engineered to address a spectrum sharing scheme with very high granularity, where we divide the network coverage area in pixels that also maps to network slices, and allocate to every pixel and network slice, the required spectrum to reflect demand in that area for a given time period. This avoids the need to overprovision spectrum to each slice, and in all territory where that slice is offered, according to the busy hour traffic demand. The proposed spectrum sharing scheme is implemented in software to exploit the programmability of 5G networks, and thus avoiding limiting the number of slices that can be offered in each pixel. Therefore, it can support all the spectrum sharing schemes proposed in the literature (TV white spaces, licensed shared access, citizens broadband radio service, etc.), and goes beyond the state-of-art by allowing to implement many other sharing schemes, with different priorities to access and leave the spectrum, and different interference protection levels.<br/><br/>To implement the proposed concept, a 3GPP-compliant radio management architecture is proposed, that leverages the 3GPP service-based 5G network management functions (MnF) to integrate machine learning (ML) algorithms to continuously optimise radio resource allocation (including spectrum sharing) in different parts of the network. More specifically, radio resources can be managed at the network slice level, access network and at the transport network level, for all cells, for a group of cells, in a single cell or in a single link. We provide some use case examples to demonstrate the applicability of the radio resource management architecture.<br/><br/>To evaluate the proposed concept, the use case that pertains to radio resource allocation at the access network and transport network level was chosen. We assumed a satellite-assisted 5G mobile network using integrated access and backhaul (IAB) to provide additional backhaul capacity to every slice of a congested base station. Moreover, we used reinforcement learning and a double deep Q-network (DDQN) agent, to select the most appropriated backhaul link capable to provide additional backhaul capacity to each network slice of a congested base station. For every slice, the backhaul link was selected from a pool of wired, wireless (IAB) and satellite links. To select this backhaul link, a reinforcement learning simulator was developed, that interacted with a network simulator that kept track of the usage level of each backhaul link.

Name dc:type.qualificationname

Doctoral Thesis

Level dc:type.qualificationlevel

Student thesis

Year dc:date.issued

2026

Author dc:creator

• Morgado, António Jorge Da Silva

Advisors dc:contributor.advisor

• Rodriguez, Jonathan
• Mumtaz, Shahid

• Integrated Access and Backhaul
• Spectrum Management
• Radio Systems
• Machine Learning
• Reinforcement Learning

Language dc:language

eng

Identifier

oai:pure.atira.dk:studenttheses/da0c681a-3805-4896-9b00-1a9b827dff26

OAI identifier oai:identifier

oai:pure.atira.dk:studenttheses/da0c681a-3805-4896-9b00-1a9b827dff26