Many machine learning techniques are currently used to optimize single components of a transmitter or receiver of an optical communication system. Typical such tasks are equalization, nonlinearity mitigation, carrier phase recovery etc. However optimizing the single components can be a suboptimal approach.
End-to-end neural network (NN) based autoencoders optimize transmitters and receivers as a single process.
Transmitter, channel and receiver are implemented as a single NN and jointly trained. This approach is particularly well suited when the system model is unknown or computationally prohibitive.
Autoencorders where applied to wireless communication systems for the first time in  and subsequent to IM/DD optics e.g.  and also to coherent optics e.g. .
The task of the student is to learn and understand the topic of autoencoders in communication systems and analyze different approach and/or possible problems.