AI-enhanced constellation design for NOMA system: A model driven method

Grant-free Non-orthogonal Multiple Access (GF-NOMA) is a promising technology for massive access users and sporadic small-packet transmission for Beyond the 5th Generation mobile communication system(B5G) / the 6th Generation mobile communication system (6G). One of the key aspects in GF-NOMA system is the signature/constellation design. However, due to the channel variation and random activation of users, conventional optimization approaches seem unsuitable for such complex models. In this paper, as an initial attempt, we propose a human intelligence (HI)-guided artificial intelligence (AI)-enhanced signature/constellation design method. By separate design of modulation and power allocation inspired by prior knowledge, the proposed deep neuron network (DNN) for NOMA signature/constellation design not only has smaller size of DNN and less training data, but also has stronger interpret-ability. In the last section, via simulations we demonstrate that in terms of bit error rate, the proposed scheme can achieve significant performance gain over the conventional NOMA schemes.