Deep Learning-Assisted Power Minimization in Underlay MISO-SWIPT Systems Based On Rate-Splitting Multiple Access

• Published in: IEEE access Vol. 10; pp. 62137 - 62156
• Main Authors: Camana, Mario R., Garcia, Carla E., Koo, Insoo
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Array signal processing; Artificial neural networks; Cognitive radio; cognitive radio network; Deep learning; Energy harvesting; Interference; Machine learning; MISO (control systems); MISO communication; Multiaccess communication; Multiple access; NOMA; Optimization; Penalty function; Power transfer; Rate-splitting (RS); semidefinite relaxation (SDR); simultaneous wireless information and power transfer (SWIPT); Splitting
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2022.3182552

In this article, we consider a multi-user multiple-input single-output underlay cognitive radio system with simultaneous wireless information and power transfer (SWIPT) based on the rate-splitting multiple access (RSMA) framework. The system model is composed of a set of secondary users that only decode information, and another set of secondary users that simultaneously decode information and harvest energy based on a power-splitting (PS) ratio. Precoders are designed to minimize the transmission power of the secondary transmitter subject to a minimum rate requirement, an energy harvesting requirement, and maximum allowable interference with the primary network. The optimization problem is non-convex and challenging. Thus, we divide it into two subproblems where the outer problem is solved by a deep neural network (DNN)-based scheme with an autoencoder, and the inner problem is solved based on the semidefinite relaxation (SDR) technique. The inner problem takes the solution of the DNN-based scheme to provide the precoder vectors and PS ratios based on SDR, where a penalty function is proposed to guarantee feasible solutions to the problems. Our simulation results prove that the proposed framework based on RSMA outperforms the conventional methods and can achieve performance close to that of the optimal solutions, with a significant reduction in computational complexity.