Deep Learning Aided Channel Estimation in Intelligent Reflecting Surfaces

Intelligent Reflecting Surfaces (IRS) consist of multiple independently controllable passive reflecting elements that can change the phase and the amplitude of the reflected signals to achieve passive beamforming. To facilitate passive beamforming, channel state information (CSI) needs to be available at the Base Station so that the optimum reflection pattern can be calculated. Therefore, the objective of this research is to present a Deep Learning (DL) based approach to improve the accuracy of the channel estimation in an IRS aided Multiple Input Single Output - Orthogonal Frequency Division Multiplexing (MISO-OFDM) wireless network. A Convolutional Neural Network (CNN) that treats OFDM frames as images is adapted for IRS and applied to the direct and cascaded channels. The CNN presented in the study is trained with channel coefficients obtained by Least Squares (LS) method and Discrete Fourier Transform (DFT) as the reflection pattern at the IRS. The results show that the CNN improves channel estimation efficiency by reducing the effects of noise and improving the Normalized Mean Square Error (NMSE) parameters.

Automated summary

Intelligent Reflecting Surfaces (IRS) is a technology that uses passive reflecting elements to achieve passive beamforming. This research proposes a Deep Learning (DL) based approach to improve channel estimation in an IRS-aided wireless network. A Convolutional Neural Network (CNN) is used to treat OFDM frames as images and is trained with channel coefficients obtained by LS and DFT methods. The results show that the CNN improves channel estimation efficiency by reducing noise effects and improving NMSE parameters.