Application of Deep Learning to Energy-Efficient Power Allocation Scheme for 5G SC-NOMA System with Imperfect SIC

Abstract

Non-orthogonal multiple access (NOMA) is a multiuser access technique for 5G communication system, which has drawn more attention from both academia and industry because of its higher spectrum efficiency than that of orthogonal multiple access (OMA). NOMA allows more than one user to access the same time-frequency resource and separates multiuser signal by successive interference cancellation (SIC). However, error propagation in SIC procedure called imperfect SIC can cause a severe performance degradation. In this paper, we propose the application of deep learning to power allocation to reduce the effect of imperfect SIC under energy-efficient perspective for downlink NOMA system. Firstly, we formulate a non-convex optimization problem aiming to maximize energy efficiency subject to a minimum data rate of each user and a total transmit power of base station (BS). Secondly, on the formulated problem, we obtain the optimal power allocation via exhaustive search method. Lastly, we train a deep neural network to predict the obtained power allocation. Simulation results verify that the proposed scheme provides the near optimal performance with very low computational complexity.