Performance Analysis Of RIS Machine Learning For Next-Generation Wireless Communication Systems

Abstract

By dynamically altering the propagation environment, reconfigurable intelligent surfaces (RIS), which are made up of meta-materials with electromagnetic wave control capabilities, are becoming a key technology in 6G wireless communication. In order to achieve the maximum path gain(dB) at the two receivers provided by only a single transmitter through a couple RIS, this investigation addresses gradient-based optimization of RIS configurations. The phase profiles of the RIS are improved by machine learning approaches, utilizing a training model that employs the Adam optimizer and gradient descent.

This study identifies the RIS area that delivers the greatest path gain improvement through a comparison of the initial path gain to the optimized path gain, with a focusing lens for different RIS areas. The effect of various antenna configurations, such as isotropic, hw_dipole(sionna), dipole and TR 38.901 antennas, on the path gain performance in RIS-assisted communication is also examined in this work. Simulation results illustrate iterative improvements in path gain, indicating the potential of machine learning in setup RIS according to waves with ultimate path gain, fostering 6G and next-generation wireless Communication applications. All simulations are performed using Sionna 0.19.1 and Python 3.9.