A Deep Learning-Based Channel Estimation Approach for MISO Communications with Large Intelligent Surfaces

We consider multi-antenna wireless systems employing large intelligent surfaces (LIS); a new physical layer technology for improving coverage and energy efficiency by intelligently controlling the propagation environment. In practice, achieving the promised gains of LIS requires accurate channel estimation. Recent solutions have been presented based on the simple, but sub-optimal, least-squares (LS) approach. Here, we propose an improved channel estimator based on the minimum mean-squared-error (MMSE) criterion. While a closed-form MMSE solution is intractable, we obtain an approximate MMSE solution by employing a deep learning-based denoising convolutional neural network (DnCNN) that takes as input the noisy LS channel estimate, and produces a cleaned channel matrix at its output. Simulation results show that the proposed DnCNN-based estimator achieves a 3 dB improvement in mean squared error compared with the existing LS approach.