Contribution to the modelling and evaluation of radio network slicing solutions in 5G
Network slicing is a key feature of 5G architecture that allows the partitioning of the network into multiple logical networks, known as network slices, where each of them is customised according to the specific needs of a service or application. Thus, network slicing allows the materialisation of m...
| Autor: | |
|---|---|
| Formato: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2022 |
| País: | España |
| Recursos: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/674243 |
| Acesso em linha: | http://hdl.handle.net/10803/674243 https://dx.doi.org/10.5821/dissertation-2117-367549 |
| Access Level: | acceso abierto |
| Palavra-chave: | 621.3 |
| Resumo: | Network slicing is a key feature of 5G architecture that allows the partitioning of the network into multiple logical networks, known as network slices, where each of them is customised according to the specific needs of a service or application. Thus, network slicing allows the materialisation of multi-tenant networks, in which a common network infrastructure is shared among multiple communication providers, acting as tenants and each of them using a different network slice. The support of multi-tenancy through slicing in the Radio Access Network (RAN), known as RAN slicing, is particularly challenging because it involves the configuration and operation of multiple and diverse RAN behaviours over the common pool of radio resources available at each of the RAN nodes. Moreover, this configuration needs to be performed in such a way that the specific requirements of each tenant are satisfied and, at the same time, the available radio resources are efficiently used. Therefore, new functionalities that allow the deployment of RAN slices are needed to be introduced at different levels, ranging from Radio Resource Management (RRM) functionalities that incorporate RAN slicing parameters to functionalities that support the lifecycle management of RAN slices. This thesis has addressed this need by proposing, developing and assessing diverse solutions for the support RAN slicing, which has allowed evaluating the capacities, requirements and limitations of network slicing in the RAN from diverse perspectives. Specifically, this thesis is firstly focused on the analytical assessment of RRM functionalities that support multi-tenant and multi-services scenarios, where services are defined according to their 5G QoS requirements. This assessment is conducted through the Markov modelling of admission control policies and the statistical modelling of the resourc allocation, both supporting multiple tenants and multiple services. Secondly, the thesis addresses the problem of slice admission control by proposing a methodology for the estimation of the radio resources required by a RAN slice based on data analytics. This methodology supports the decision on the admission or rejection of new RAN slice creation requests. Thirdly, the thesis explores the potential of artificial intelligence, and specifically, of Deep Reinforcement Learning (DRL) to deal with the capacity sharing problem in RAN slicing scenarios. To this end, a DRL-based capacity sharing solution that distributes the available capacity of a multi-cell scenario among multiple tenants is proposed and assessed. The solution consists in a Multi-Agent Reinforcement Learning (MARL) approach based on Deep Q-Network. Finally, this thesis discuses diverse implementation aspects of the DRL-based capacity sharing solution, including considerations on its compatibility with the standards, the impact of the training on the achieved performance, as well as the tools and technologies required for the deployment of the solution in the real network environment. |
|---|