Variational Bayesian Inference-based Robust Channel Estimation for IRS-assisted mmWave MIMO systems

Intelligent Reflecting Surfaces (IRS) play a significant role in providing enhanced coverage, and improved energy efficiency for millimeter-wave (mmWave) multiple input multiple output (MIMO) systems. Performance gains of IRSassisted systems rely heavily on accurate channel state information. Existing works in literature leverage the underlying sparsity of $\mathbf{m m W a v e}$ channels to estimate the cascaded channel between base station (BS), IRS, and user equipment (UE) with low pilot overhead. However, the conventional sparse modelingbased channel estimation approaches yield less accurate channel estimates in practical scenarios where the statistics of the channel and noise are erratic. This work develops novel Variational Bayesian Inference (VBI)-based Sparse Bayesian Learning (SBL) algorithms by employing a hierarchical model-based approach capable of capturing the uncertainties in the channel and noise. Simulation results illustrate the superior performance of VBIbased schemes over the conventional SBL channel estimation approaches. Further, the robustness of the proposed schemes for non-ideal and non-Gaussian noise conditions is also demonstrated.