Channel Estimation for Intelligent Reflecting Surface Aided Wireless Communications Using Conditional GAN

Channel estimation is very challenging, especially in an intelligent reflecting surface (IRS)-aided wireless system. This letter proposes a deep learning (DL) based approach for IRS-assisted systems. Specifically, a conditional generative adversarial network (cGAN) is designed to estimate the cascaded channels with the received signals as conditional information. Two DL networks are trained adversarially to learn an adaptive loss function to generate the more realistic cascaded channels. Numerical results show that the proposed cGAN-based method outperforms the state-of-the-art DL-based approach and achieves high robustness in the IRS-assisted system.