Radio resource management techniques for 5G networks based on machine learning

• Ano de defesa: 2020
• Autor(a) principal: Costa Neto, Francisco Hugo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Palavras-chave em Português: Gerenciamento de recursos de rádio; Formatação híbrida de feixes; Controle de potência; Aprendizado de máquina
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/58915

The fifth generation (5G) of mobile communications has been envisioned to expand the capabilities of wireless networks and, consequently, to provide optimized support to several use cases and design requirements. In view of this, massive multiple-input multiple-output antenna arrays and the operation at the millimeter wave (mmWave) frequency range are important technical solutions able to support an expressive enhancement of the data traffic capacity, a recognizably relevant demand of 5G. In this context, the present thesis investigates radio resource management (RRM) techniques to explore these technologies and to overcome their main challenges, such as hostile propagation conditions, demanding channel state information (CSI) acquisition, and transceiver implementation complexity. Moreover, the proposed solutions rely on the main technical specifications from the third partnership project (3GPP) aiming to consider practical implementation aspects. In the first part of this thesis, devoted to the hybrid beamforming design based on the joint spatial division and multiplexing scheme, we propose a framework to exploit a limited CSI feedback and to reduce the inter-cell interference considering different mmWave propagation conditions. In the second part of this document, we investigate an uplink power control framework compliant with the beam-centric design of the air interface of 5G radio access technology. The proposed signaling scheme among base stations allows a flexible transmit power control able to increase the energy efficiency by the enhancement of the system data rate and to reduce the power consumption while limiting interference to neighbor cells. This thesis explores different machine learning (ML) paradigms to optimize 5G network deployment. We investigate how ML can help to uncover unknown properties of the wireless channel and establish successful RRM strategies from the knowledge determined by the interaction with the network. Numerical analyses are presented to validate the proposed methods and to demonstrate that, despite the limitations imposed by the 3GPP technical specifications, such as hardware restrictions and available signaling, the proposed solutions improve system performance and achieve relevant engineering requirements, such as data rate improvement and energy efficiency enhancement with reduced signaling overhead and computational complexity