Deep Neural Networks for Transmission Impairment Mitigation in Long-Reach 5G Access Networks

The mitigation of transmission impairments in optical communication systems requires implementation of digital signal processing that can easily adapt to changing conditions and work on continuous streams of data. In this paper, the application of continual learning in deep neural networks for transmission impairments in long haul transmission systems is considered. This work investigates the novel use of continual learning to overcome the problem of catastrophic forgetting and remembrance in deep neural networks that can be used for dispersion and nonlinearity mitigation in long-haul transmission systems. We perform comparisons between state-of-the-art methods such as iCaRL, DGR, EWC, XdG and traditional DNNs implementations typically used in optical networks. We demonstrate improvements in training accuracy towards signal recovery, and increased flexibility for DNN equalization applications. The results show that the method can be successfully used for fiber lengths ranging from less than 1km to 250km and over with no need to retrain the network from scratch for each fibre length. The introduction of continual learning enables training DNNs on a continuous stream of data with no loss of prior knowledge.