Machine learning applications in communication systems decoding

Abstract

The usage of machine learning (ML) techniques in different academic and professional fields confirms its theoretical and practical utility. The communications field is no exception. In fact, models that learn from data were already in use prior to the recent advancement in the ML field. This research investigates different kinds of usage that can be done with ML models in three different problems, seeking to show their high flexibility and to present alternative ways of obtaining classical results which employ well established algorithms, or even outperform them in some scenarios. The first problem discusses the so-called Markov-Gaussian channels and compares an ML model with the already common hidden Markov models approach. The second problem deals with non-orthogonal multiple access transmissions and compares an ML model with the usually employed decoding algorithm. The third presents a chaos-based communication system and compares the maximum likelihood decoding to a neural network-based one.