O-NIIS: Open and Intelligent IRS-SWIPT For Wireless Energy-Efficiency Enhancement

초록

This study proposes an Open and Intelligent IRS SWIPT (O-NIIS) framework to enhance energy efficiency in wireless networks, leveraging the openness and interoperability of O-RAN to seamlessly integrate emerging technologies, including intelligent reflecting surfaces (IRS), simultaneous wireless information and power transfer (SWIPT), and artificial intelligence (AI), within a cellular system. The IRS redirects scattered RF signals toward desired directions under the control of the RAN intelligent platform called AI-based xApps, thereby boosting the received signal strength at user equipment (UE) and maximizing both energy harvesting and information decoding through SWIPT technology. The O-NIIS framework then formulates the network problem as a complex-valued NP-hard optimization problem, aiming to maximize energy efficiency under throughput constraints, and develops a tailored unsupervised deep neural network (uDNN) to solve this problem. uDNN learns by inter acting with a contextual database that stores this telemetric data received from the base station and IRS, enabling it to generate optimal beamforming weights for both of them. Furthermore, a solution error mitigation mechanism is incorporated into uDNN learning to quantify and correct discrepancies between solutions derived from estimation and those obtained from the physical network, enhancing alignment with real-world performance. The results demonstrate that uDNN outperforms benchmark schemes such as fixed beamforming (FB) and alternating optimization (AO). In particular, it achieves a near-global optimal solution with lower time complexity than the exhaustive search (ES) method.

키워드