Offset Learning Based Channel Estimation for Intelligent Reflecting Surface-Assisted Indoor Communication

The emerging intelligent reflecting surface (IRS) can significantly improve the system capacity, and it has been regarded as a promising technology for beyond fifth-generation (B5G) communications. For IRS-assisted multiple input output (MIMO) systems, accurate channel estimation is critical challenge. This severely restricts practical applications, particularly resource-limited indoor scenario contains numerous scatterers parameters to be estimated, while number of pilots limited. Prior art tackles these issues associated optimization using mathematical-based statistical approaches, but are difficult solve increase. To estimate channels with an affordable piloting overhead, we propose offset learning (OL)-based neural network estimation. proposed OL-based estimator dynamically trace state information (CSI) without any prior knowledge structure well statistics. In addition, inspired by powerful capability convolutional (CNN), CNN-based inversion blocks developed in module build operator. Numerical results show that achieve more CSI lower complexity compared benchmark schemes.