Energy-Efficient Power Allocation in Downlink Multi-Cell Multi-Carrier NOMA: Special Deep Neural Network Framework

Abstract

Energy-efficient resource allocation for multi-cell multi-carrier non-orthogonal multiple access (MCMC-NOMA) is a challenging task due to the interference and other related factors, which makes obtaining an applicable solution in the real-time is even more challenging. In this paper, we aim to design a model capable of efficient resource allocation in real-time. We formulate our problem as energy efficiency (EE) maximization. First, we propose an iterative solution to handle user scheduling and power allocation, which is not suitable for real-time application. Next, we design a dual-pipeline augmented deep convolutional neural network (ADCNN) to handle the power allocation in real-time. The first pipeline is to extract high quality features and spatially connect and refine them using attention-based network. The second pipeline is to extract low-quality spatial features. Because more discriminative features are obtained through fusion of the high and low quality features, a better prediction of power allocation can be obtained. Simulation results show the adequacy of the proposed model for the real-time application and larger problems compared with other models such as deep neural network (DNN).