Deep Residual Denoising Network for IRS-Cascaded Channel Estimation with a Receiver Impairment

In intelligent reflecting surface (IRS) assisted communication systems, channel estimation is of great importance to realize the performance gains promised by an IRS that adapts to the channel environments through reflection pattern optimization. Recently, deep learning networks have been applied to estimate the IRS cascaded channel to find an estimator close to the optimal one which is intractable due to the non-Gaussian nature of the cascaded channel. This paper applies a deep residual denoising network to estimate the cascaded channel more robustly in the presence of a residual carrier frequency offset (CFO). Simulation results show that the learning-based channel estimation provides a significant improvement in mean square error compared with the linear estimators.