Predição de potência de saída em amplificadores a fibra dopada com érbio baseada em redes neurais artificiais

• Autor(a) principal: Chiang, Luis Henrique
• Data de Publicação: 2021
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/214611
• topic: Amplificadores de potência; Redes neurais (Computação); Telecomunicações

Erbium-doped fiber amplifiers represent a critical device in many communications systems. These devices feature a high output power, a relatively low noise figure, and a low sensitivity to polarization. However, in the presence of multiple wavelength-multiplexed channels, competition for the excited ions of erbium results in cross-gain modulation. The model typically adopted to consider this effect is the Giles-Desurvire model, which consists of a system of coupled differential equations. In this work we propose an erbium-doped fiber amplifier model based on artificial neural networks capable of considering the cross-gain effect between different channels without needing to integrate a system of coupled differential equations. In particular, we use a multilayer feedforward perceptron operating in regression mode with a single hidden layer to model each of the amplifier outputs. It is worth mentioning that the number of neurons in the hidden layer is optimized using stochastic cross validation. The numerical results reveal that the proposed model accurately predicts the output power in a fourchannel system. We also show that the complexity of neural networks is different for each channel, being greater in the central channels. For the studied object, the computational cost of the ANN model trained in this work is 370 floating point operations, which is predictably less than the complexity of numerically solving the Giles-Desurvire model.