Laurent SCHMALEN  (Karlsruhe Institute of Technology)

Autoencoders in Optical Communications – From Modulation Format Optimization to Blind Equalization

Abstract
In the recent years, machine learning techniques have proven to be indispensable tools for designing communication systems. One particularly popular technique is the concept of auto-encoders. In this talk, we will introduce the concept of auto-encoders and show how they can be used in two distinct ways to optimize the physical layer of optical communication systems. In particular, we show how to design higher order constellations that are tailored to channels exhibiting laser phase noise by using a novel differentiable blind phase search algorithm. As a second application, we use the concept of variational auto-encoders to design novel blind equalizers for coherent optical communications compensating linear impairments. The proposed algorithm jointly carries out equalization and performs a channel estimation, which is useful for providing information to higher network layers.