Placement and sum-rate maximization of IRS enhanced next generation networks using machine learning

Abstract

From the introduction of wireless communication to the development and deployment of forth generation of wireless telecommunication systems, the speed of reliability of communication between human users was prioritized. As other technologies grew and became widespread, more and more devices needed to be connected to the internet to be used efficiently. Automation and data driven systems started to contribute to a no ticeable portion of data services utilization. These devices now constitute a number of application scenarios, for which certain network metrics have to be met for the proper functioning of the systems. 6G systems aim to fully enable internet of things supporting machine to machine communication, along with high data rate and a number of other features necessary for several service classes. Intelligent Reflecting Surfaces (IRSs) have the potential to be one of the key enabling technologies in 6G with the ability to manip ulate the propagation environment of the signals. In this study we explore and develop potential methods for optimal IRS placement in a single base station SISO downlink OFDMA communication system. The problem of distributed IRS placement is ad dressed using smart clustering methods. The performance of these clustering techniques is enhanced using geometrical techniques and gradient descent. The results are then compared with classical grid-search optimized grid-search and the genetic algorithm. The results show that the intersection enhanced smart clustering algorithms, with the least computational complexity, perform close to the grid search algorithms. This makes them useful for energy constrained scenarios. The gradient descent optimized clustering technique provides the best performance in terms of outage but is more computationally complex than the intersection-based algorithm. Finally, the interference performance of the system is heavily dependant on the number and type of clustering. The increase in the total number of IRS elements in the system decreases the performance difference between systems with different number of clusters.