On the Complexity of Machine Learning based Algorithms for Intelligent Reflecting Surface aided Communication System: A Survey

The demand for high data rates for applications such as telemedicine and online gaming has fueled the application of mmWave in wireless communication to harness the high bandwidth that will support high data rate. If the scatterers are badly distributed in the propagation environment, the attenuation loss is large thus affecting the mmWave communication. Researchers have revealed that, the metasurface devices known as Intelligent Reflecting Surfaces (IRS), can be used to manipulate the propagation environment to achieve constructive interference to the intended user and destructive interference to none intended user. Consequently, increasing the received energy to the intended user, thus improving the system data rate. It can also be used for security by exploiting the destructive interference feature, where none intended users will receive almost no signal at all. In this work, we provide a summary of communication systems that are aided with IRS, while using machine learning methods to find the proper phase shift that maximizes the system performance. The summary is expected to inform the system developers on appropriate machine learning algorithm that can be chosen depending on their requirements. Specifically, the study presents an analysis of their computational complexity versus the number of IRS elements. The results shows that, more than $50 \%$ of the algorithm's computational complexity increases exponentially with the number of IRS elements. This shows that, there is a trade off on achieving high beamforming gain by IRS surface with large number of array and the affordable computational complexity of the algorithm.