Self organisation for 4G/5G networks
Nowadays, the rapid growth of mobile communications is changing the world towards a fully connected society. Current 4G networks account for almost half of total mobile traffic, and in the forthcoming years, the overall mobile data traffic is expected to dramatically increase. To manage this increas...
| Autor: | |
|---|---|
| Formato: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2016 |
| País: | España |
| Recursos: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/403990 |
| Acesso em linha: | http://hdl.handle.net/10803/403990 https://dx.doi.org/10.5821/dissertation-2117-105813 |
| Access Level: | acceso abierto |
| Palavra-chave: | Àrees temàtiques de la UPC::Enginyeria de la telecomunicació 621.3 |
| Resumo: | Nowadays, the rapid growth of mobile communications is changing the world towards a fully connected society. Current 4G networks account for almost half of total mobile traffic, and in the forthcoming years, the overall mobile data traffic is expected to dramatically increase. To manage this increase in data traffic, operators adopt network topologies such as Heterogeneous Networks. Thus, operators can de ploy hundreds of small cells for each macro cell, allowing them to reduce coverage hales and/or lack of capacity. The advent of this technology is expected to tremendously increase the number of nodes in this new ecosystem, so that traditional network management activities based on, e.g., classic manual and field trial design approaches are just not be viable anymore. As a consequence, the academic J literature has dedicated a significant amount of effort to Self-Organising Network (SON) algorithms. These solutions aim to bring intelligence and autonomous adaptability into cellular networks, thereby reducing capital and operation expenditures (CAPEX/OPEX). Another aspect to take into account is that, these type of networks generate a large amount of data during their normal operation in the form of control, management and data measurements. This data is expected to increase in SG due to different aspects, such as densification, heterogeneity in layers and technologies, additional control and management complexity in Network Functions Virtualisation (NFV) and Software Defined Network (SDN), and the advent of the Internet of Things (loT), among others. In this context, operators face the challenge of de signing efficient technologies, while introducing new services, reaching challenges in terms networks, which are self-aware, self-adaptive, and intelligent. This dissertation provides a contribution to the design, analysis, and evaluation of SON solutions to improve network opera tor performance, expenses, and users' experience, by making the network more self-adaptive and intelligent. It also provides a contribution to the design of a self-aware network planning tool, which allows to predict the Quality of Service (QoS) offered to end-users, based on data al ready available in the network . The main thesis contributions are divided into two parts. The first part presents a novel functional architecture based on an automatic and self-organised Reinforcement Learning (RL) based approach to model SON functionalities, in which the main task is the self-coordination of different actions taken by different SON functions to be automatically executed in a self-organised realistic Long Term Evolution (LTE) network. The proposed approach introduces a new paradigm to deal with the conflicts genera ted by the concurrent execution of multiple SON functions, revealing that the proposed approach is general enough to modelali the SON functions and their derived conflicts. The second part of the thesis is dedicated to the problem of QoS prediction. In particular, we aim at finding patterns of knowledge from physical layer data acquired from heterogeneous LTE networks. We propose an approach that not only is able to verify the QoS level experienced by the users, through physical layer measurements of the UEs, but it is a lso able to predict it based on measurements collected at different time, and from different regions of the heterogeneous network. We propose then to make predictions independently of the physical location, in order to exploit the experience gained in other sectors of the network, to properly dimension and deploy heterogeneous nodes. In this context, we use Machine Learning (ML) as a tool to allow the network to learn from experience, improving performances, and big data analytics to drive the network from reactive to predictive. |
|---|