Abstract
Along with the advancement of edge computing and next-generation cellular networks, the demand for wireless communication systems is increasing significantly. The fifth-generation (5G) wireless systems are beginning to spread worldwide and gradually finding their application in many domains. However, 5G alone is not enough to solve the problems associated with more complex systems, where technologies such as machine learning (ML) and artificial intelligence are applied, the operational requirements of which can be covered by the use of sixth-generation (6G) networks. One of the most important aspects of implementing 6G is the reconfigurable intelligent surface (RIS), which uses reflective elements capable of manipulating electromagnetic waves. However, the use of a large number of passive reflecting elements in standard RIS complicates the task of channel estimation (CE). To achieve a significant increase in performance, a semi-passive RIS architecture is proposed using a few active sensors that can receive and reflect signals. In this paper, we consider semi-passive RIS and explore different respective CE approaches, classifying them into traditional and ML-based ones, with different metrics for optimization. Finally, in addition to the CE challenges, the issues with RIS deployment and user localization are also studied.