loading page

Channel Estimation for RIS-Aided Communications Based on Attention-guided Deep Residual Networks
  • +1
  • Jing Zhang,
  • Qiang Zhang,
  • Zhibo Zhang,
  • Ying Su
Jing Zhang
Shanghai Normal University

Corresponding Author:[email protected]

Author Profile
Qiang Zhang
Shanghai Normal University
Author Profile
Zhibo Zhang
Shanghai Normal University
Author Profile
Ying Su
Shanghai Normal University
Author Profile

Abstract

Reconfigurable intelligent surface (RIS) is the meta-material to passively reflect electromagnetic wave and provide programmable and well-defined wireless environment. In RIS-aided transmission systems, accurate channel estimation is indispensable to realize performance gains. We propose an attention-guided denoising deep residual network (ADNet) to learn residual noise from least square estimation samples. The deep convolutional neural network (CNN) consists of a sparse block, a feature enhancement block, an attention block and a reconstruction block. The attention mechanism is exploited to strengthen the extraction of specific noise features. Simulation results show that the ADNet can achieve accuracy comparable to the ideal linear minimum mean square error estimation. It also obtains higher accuracy than the general CNN-based deep residual network and spatial-frequency CNN.