Deep Learning Based Resource Allocation in NOMA Wireless Power Transfer Networks

Abstract

This paper investigates resource allocation of simultaneous wireless information and power transfer (SWIPT) in non-orthogonal multiple access (NOMA) system with power splitting (PS) technology. Our goal is to optimize both the harvested energy and the transmission rate subject to the minimum requirements on each user's harvested energy and transmission rate. Considering the battery is capable of storing power which is converted into throughput in reverse link, we establish the equivalent-sum-rate (ESR) maximazation function by combining the transformed throughput and the transmission rate with a weight coefficient. Since the formulated nonconvex problem is difficult to solve, we develop a deep learning-based approach to obtain an efficient resource allocation strategy. In particular, deep belief network (DBN), which comprises preparing data samples, training and running, has been adopted. The simulation results verify the effectiveness of the proposed DBN-based scheme, and prove that the ESR performance of the considered NOMA-SWIPT is better than that of the OFDMA-SWIPT.