Supervised learning applied to the decoding of SCMA codewords

This work puts together two technologies that are in the interest of the scientific community, on the one hand, access methods for fifth generation systems of mobile communications, in this case Sparse Code Multiple Access (SCMA), and on the other hand supervised learning based on neural networks. SCMA is one of the proposed access techniques for fifth generation mobile communication systems. Until now, the detection algorithm in the receiver is based on Message Passing Algorithm (MPA) or minimum Euclidean distance. In this work, a new approach is proposed, which is based on supervised learning using neural networks to decode SCMA codewords. The simulation results show that the receiver based on neural networks learns quickly and obtains 100% accuracy in predictions on channels with high noise. In addition to being simpler in its implementation than its predecessors (MPA and minimum Euclidean distance).