Channel Estimation for STAR-RIS-Aided Communications Based on Deep Iterative Networks

With the development of 6G communication, the reconfigurable intelligent surfaces (RIS) is proposed to be deployed in 6G systems to assist communications. RIS is composed of multiple passive units with no signal storage or processing capabilities, which is a low-power and low-cost emerging technology. However, it can only serve users on one side, thus the simultaneously transmitting and reflecting RIS (STAR-RIS) is proposed. Compared with the traditional RIS, STAR-RIS can serve users in the whole space through transmitting and reflecting signals. In order to obtain a high-quality communication effect, accurate channel state information (CSI) is indispensable. However, due to the passive characteristics of RIS and the larger channel dimension caused by the deployment of RIS, the pilot cost and computational complexity of channel estimation rise sharply, so efficient and accurate estimation algorithms need to be proposed. In this paper, one deep learning algorithm based on gradient-descent-based deep-iterative-unrolling network (GD-Net) is proposed, and the superiority of this algorithm is verified by the simulation results.