Deep Convolutional Self-Attention Network forEnergy-Efficient Power Control in NOMA Networks

In this letter, we propose an end-to-end multi-modalbased convolutional self-attention network to perform powercontrol in non-orthogonal multiple access (NOMA) networks. Weformulate an energy efficiency (EE) maximization problem wedesign an iterative solution to handle the optimization problem.This solution can provides an offline benchmark but might notbe suitable for online power control therefore, we employ ourproposed deep learning model. The proposed deep learning modelconsists of two main pipelines, one for the deep feature mappingwhere we stack our self-attention block on top of a ResNet toextract high quality features and focus on specific regions in thedata to extract the patterns of the influential factors (interference,quality of service (QoS) and the corresponding power allocation).The second pipeline is to extract the shallow modality features.Those features are combined and passed to a dense layer toperform the final power prediction. The proposed deep learningframework achieves near optimal performance and outperformstraditional solutions and other strong deep learning models suchas PowerNet and the conventional convolutional neural network(CNN).