NOMANet: A Graph Neural Network Enabled Power Allocation Scheme for NOMA

This paper proposes a graph neural network (GNN)-enabled power allocation scheme for non-orthogonal multiple access (NOMA) networks. We consider a downlink scenario where one base station serves multiple users over several subchannels. The number of available subchannels is less than the number of users, and thus, some users have to share a subchannel via NOMA. Our goal is to maximize the system energy efficiency subject to the users' rate requirements and the overall transmit power budget. We propose a deep learning-based approach termed NOMA-Net (NOMANet) to solve the considered problem. Particularly, NOMANet adopts a GNN-based architecture that maps channel state information (CSI) to the desired power allocation scheme for all subchannels. NOMANet adopts multi-head attention and residual/dense connections to enhance feature extraction. To guarantee the feasibility of NOMANet's output, we employ a customized activation function and a penalty method. Numerical results show that NOMANet trained in an unsupervised manner achieves performance close to that of the successive convex approximation (SCA) method but with an inference speed approximately 700 times faster. Besides, NOMANet exhibits scalability to both the number of users and subchannels.