A survey about deep learning for constellation design in communications

Abstract

The performance of communication systems dependson the choice of constellations, designed in an end-to-endmanner. In case of a mathematical intractability, either becauseof complexity or even lack of channel model only sub-optimal solutionscan be provided with an analytical approach. We presentend-to-end learning, a recent technique in communications tolearn optimal transmitter and receiver architectures based ondeep neural networks (DNNs) architectures. We discuss cases inwhich this technique has been used to design constellations inwhich channel model intractability repressed from a mathematicalanalysis.