Deep Gaussian Process for Channel Estimation in LIS-assisted mm-Wave Massive MIMO Systems

In this letter, we introduce deep Gaussian process (DGP) models for channel estimation in large intelligent surface (LIS)-assisted millimeter-wave (mm-Wave) massive MIMO (m-MIMO) systems. In our proposed approach, a properly structured and trained DGP is employed that provides a flexible and efficient mapping of the input data into the desired channel estimates. Specifically, using the received pilot signals, each user accesses the trained DGP to estimate both its direct and cascaded channels in the LIS-assisted m-MIMO system. In addition to channel estimation, our approach provides an uncertainty as a measure of the model's reliability. Simulation results confirm the better channel estimation performance and competitive computational complexity of the proposed DGP-based approach compared to other learning-based methods.