Geração de chave na camada física para sistemas FDD baseado em aprendizagem profunda

• Autor(a) principal: Silva, Gustavo Marques da
• Data de Publicação: 2023
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/19463
• topic: Aprendizagem profunda; Duplexação por divisão de frequência; Geração de chave aleatória na camada física; Inteligência artificial; Segurança na camada física; Deep learning; Frequency division duplexing; Physical layer random key generation; Artificial intelligence; Physical layer security; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO; ENGENHARIAS::ENGENHARIA ELETRICA::TELECOMUNICACOES

The advancement of telecommunications and the transition to the next generation of networks (6G) opens new challenges and opportunities. Security at the physical layer (PLS, Physical Layer Security) is crucial to ensure network reliability, especially considering the increasing computational power of potential attackers. Secret key generation at the physical layer is a PLS technique that offers the advantage of being less complex and resource-intensive, making it a significant enabler of security for systems with computational limitations, such as Internet of Things (IoT) devices, for example. However, in Frequency Division Duplexing (FDD) systems, which have uplink and downlink channels at different frequencies, the reciprocity required for reliable random key generation is not directly applicable due to the distinct characteristics of the channels. In this context, this work proposes the use of deep learning techniques to establish artificial reciprocity between communication channels in FDD systems. The work focuses on the construction and training of a neural network with four hidden layers, using an extensive dataset of uplink and downlink channel conditions under various scenarios. Based on this approach, a random key generation scheme is applied, and security aspects such as the Key Error Rate (KER) and Key Generation Ratio (KGR) are analyzed. The main contributions of the study are: i) the construction and training of the neural network for random key generation in FDD systems; ii) the analysis of network parameters to ensure robustness and generalization; iii) the evaluation of key security using metrics such as KER and KGR."