Next generation >200 Gb/s multicore fiber short-reach networks employing machine learning

• Autor(a) principal: Piedade, Derick Augusto Évora
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/26698
• topic: Short-reach network; Multicore fiber; Intercore crosstalk; Redes neuronais -- Neural networks; Machine learning; Kramers-Kronig receivers; Redes de curto alcance; Fibra multi-núcleo; Diafonia entre núcleos; Receptores Krames-Kronig

This work proposes and evaluates the use of machine learning (ML) techniques on >200 Gb/s short-reach systems employing weakly coupled multicore fiber (MCF) and Kramers-Kronig (KK) receivers. The short-reach systems commonly found in intra data centers (DC) connections demand low cost-efficient direct detection receivers (DD). The KK receivers allow the combination of higher modulation order, such as 16-QAM used in coherent systems, with the low complexity and low cost of DD. Thus, the use of KK receivers allows to increase the bit rate and spectral efficiency while maintaining the cost of DD systems as this is an important requirement in DC. The use of MCF allows to increase the system capacity as well as the system cable density, although the use of MCF induces additional distortion, known as inter-core crosstalk (ICXT), to the system. Thus, low complexity ML techniques such as k-means clustering, k nearest neighbor (KNN) and artificial neural network (ANN) (estimation feedforward neural network (FNN) and classification feedforward neural network) are proposed to mitigate the effects of random ICXT. The performance improvement provided by the k-means clustering, KNN and the two types of FNN techniques is assessed and compared with the system performance obtained without the use of ML. The use of estimation and classification FNN prove to significantly improve the system performance by mitigating the impact of the ICXT on the received signal. This is achieved by employing only 10 neurons in the hidden layer and four input features. It has been shown that k-means or KNN techniques do not provide performance improvement compared to the system without using ML. These conclusions are valid for direct detection MCF-based short-reach systems with the product between the skew (relative time delay between cores) and the symbol rate much lower than one (skew x symbol rate « 1). By employing the proposed ANNs, the system shows an improvement of approximately 12 dB on the ICXT level, for the same outage probability when comparing with the system without the use of ML. For the BER threshold of 10−1.8 and compared with the standard system operating without employing ML techniques, the system operating with the proposed ANNs show a received optical power improvement of almost 3 dB and a ICXT level improvement of approximately 9 dB when the mean BER is analized.