Machine Learning Concepts in Optical Communication Systems
The goal of this symposium is to bridge the gap between researchers working in the areas of optical communication and machine learning. More specifically, the symposium will address powerful statistical signal processing methods, used by the machine learning community, and link them to current problems in optical communications.
Optical communication systems are becoming increasingly complex, especially with the introduction of space division multiplexing, and therefore advanced tools are necessary in order to perform system as well as component characterization, and enable optical communication systems to operate close to channel capacity. Especially, one of the major challenges is nonlinearity associated with the optical channel (fiber) and components.
Techniques from machine learning can be equally well applied to linear and nonlinear dynamical systems as well as to systems with non-additive Gaussian noise, which makes them quite powerful especially for the nonlinear optical fiber communication channel. In general terms, methods from machine learning can be used to learn the impairments from the observed data and built a probabilistic model of the impairment. This knowledge can later be used to either perform impairment compensation or to quantify the amount of distortion coming from components or specific subsystems. Finally, the probabilistic model can be used for synthetic impairment generation which may be very useful for performance analysis and prediction in optical networks.
Department of Photonics Engineering
Technical University of Denmark, Denmark
Faculty of Electrical Engineering
Helmut Schmidt University, Germany