Alocação de recursos em redes sem fio multiportadoras com ondas milimétricas utilizando aprendizado por reforço baseado em modelo Markoviano

• Ano de defesa: 2022
• Autor(a) principal: Carneiro, Daniel Porto Queiroz
• Orientador(a): Vieira, Flávio Henrique Teles
• Banca de defesa: Vieira, Flávio Henrique Teles, Soares, Anderson da Silva, Cardoso, Alisson Assis, Lemos, Rodrigo Pinto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Engenharia Elétrica e da Computação (EMC)
• Departamento: Escola de Engenharia Elétrica, Mecânica e de Computação - EMC (RG)
• País: Brasil
• Palavras-chave em Português: Alocação de recursos em rede 5G; Aprendizado por reforço; Processo de decisão de Markov; DQN adaptativa
• Palavras-chave em Inglês: 5G resource allocation; Reinforcement learning; Markov decision process; Adaptive DQN
• Área do conhecimento CNPq: ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/12253

In this dissertation, we present reinforcement learning-based resource allocation algorithms for a multicarrier communication system considering multiple users and the effects of multipath and average path loss in a transmission assuming millimeter waves. To this end, it is proposed that the communication system can be described by a Markovian model represented by queue states in buffers and channel states. For the resource allocation algorithms of this work, we introduce reward functions to be used in the reinforcement learning algorithm Q-learning. The results obtained in the simulations show that the application of the proposed algorithms for resource scheduling provides, in general, an improvement in the performance parameters of the considered communication system, such as, for example, increased throughput and decreased packet loss. Comparisons with other algorithms presented in the literature are carried out, also showing that the use of the proposed reward function and considered Markovian model makes the scheduling of users and the sharing of resources more efficient. Furthermore, a solution for resource and power allocation using a Deep Q-Network is presented. The modeling of states proposed for the DQN network covers some limitations encountered with the matrix representation of states and extends the limits for the size of the buffer.