Deep Learning of Geometric Constellation Shaping including Fiber Nonlinearities

This addresses the challenge of improving data transmission efficiency in fiber optic networks, though it appears incremental as it builds on existing shaping methods with machine learning enhancements.

The paper tackled the problem of mitigating nonlinear effects in fiber optic communication by proposing a new geometric constellation shaping method using unsupervised machine learning, achieving gains of up to 0.13 bit/4D when trained with a simplified fiber channel model.