A Survey on RIS-Assisted Millimeter-Wave Channel Estimation

Abstract

Nowadays, with the rapid development of Reconfigurable Intelligent Surface (RIS), it has gradually become one of the most important research areas in 6G communication. By dynamically adjusting the phase shift of the reflection units, RIS can smartly arrange the wireless communication environment, so as to effectively solve the problems in millimeter-wave communication such as blockage by obstacles and the loss of the communication path. However, because of the passive nature of RIS and the large-scale antenna dimensions, channel estimation has become an important task in order to reduce the pilot expenses. Considering this, this article provides an overview of RIS-assisted channel estimation methods using in millimeter-wave systems. This article divides the estimation methods into three main methods: compressed sensing, deep learning, and beam training. Compressive sensing uses the sparse characteristic of millimeter-wave channels in the angular domain, thus turning the channel estimation problem into recovering the sparse channel. Deep learning uses the neural networks' fitting ability to learn and simulate the mapping relationship of the channel matrix from the received signals. Beam training avoids estimating the complete channel matrix. Instead, it directly tests preset phase codebooks to find the best beam direction. Then, this article provides a detailed comparison of these three methods in terms of pilot expenses, computational complexity, and hardware limitations. Finally, this article discusses the open challenges like the beam splitting in broadband millimeter-wave systems and the joint channel estimation problem in STAR-RIS systems, giving some direction for future study.