Leveraging Active and Nearly Passive Reconfigurable Intelligent Surfaces Using Deep Learning Algorithm for 6G Wireless Networks

Abstract

Active Reconfigurable Intelligent Surface (ARIS) shows promise in boosting the desired signal at the receiver user. However, the “fully-connected” architecture of ARIS needs high power due to additional active components. This paper adopts sub-connected ARIS to enhance achieved data rates with good energy efficiency at the Cell Edge Users (CEUs) and addresses the “multiplicative fading” effect caused when the signal propagates through a longer path (i.e., the serving Base Station (BS)-ARIS-CEU) than the straight route across the serving BS and the CEUs. Additionally, a Nearly Passive RIS (NP-RIS) is proposed to mitigate interfering signals from other BSs by creating destructive interference at the CEUs. The reflection matrix of the NP-RIS is extracted using Deep Learning (DL) techniques, with a select few NP-RIS reflecting elements being active. This model improves achieved data rates by around 58% for M = 16 RIS elements compared with the baseline model with the same number of elements in NP-RIS without ARIS implementation. Moreover, the proposed model enhances data rates by approximately 31.8% compared with a baseline using negative resistance Reflecting Elements (RE). However, the Spectral Energy Efficiency (SEE) using the second baseline will be improved over the “fully-connected” ARIS leading to the sub-connected ARIS solution to improve the SEE by nearly 25%.