Demodulação M-QAM empregando técnicas de Aprendizado de Máquina

• Ano de defesa: 2020
• Autor(a) principal: Toledo, Roberto Neves
• Orientador(a): Akamine, Cristiano
• Banca de defesa: Menezes, Mario Olimpio de , Lima, Eduardo Rodrigues de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Programa de Pós-Graduação: Engenharia Elétrica
• Departamento: Escola de Engenharia Mackenzie (EE)
• País: Brasil
• Palavras-chave em Português: aprendizado de máquina; demodulação; LLR, M-QAM
• Área do conhecimento CNPq: CNPQ::ENGENHARIAS
• Link de acesso: https://dspace.mackenzie.br/handle/10899/28600

This work presents the challenges faced in the demodulation of uniform and non-uniform high-order M-QAM (Quadrature Amplitude Modulation) signals, using the LLR (LogLikelihood Ratio) technique, which is nowadays one of the most widely used in the modern communication systems. The main theoretical concepts are reviewed, such as, modulation, demodulation, machine learning and cognitive radio. Comparative results are presented for several machine learning algorithms, acting as classification and regression, until the definition by the final model that is compatible with the current standards. It is proposed a new method to demodulate the M-QAM signal, evaluating its response to different modulation orders and SNR (Signal-to-Noise Ratio) values, when concatenated to a channel encoder LDPC (Low-density Parity-Check). The experimental results demonstrate a performance gain of up to 1485% for 4096-QAM in comparison with the classical LLR Max-Log-MAP demodulator, keeping the same BER (Bit Error Rate) level. Finally, this new demodulator scheme was implemented in the environment of GRC (GNU Radio Companion) to validate computational simulations.