Channel Estimation in Cellular Massive MIMO: A Data-Driven Approach

Abstract

Massive MIMO has provided immense improvement in the performance of wireless communication systems when it comes to spectral efficiency, which led to it becoming the main driving technology behind 5G. It is also expected to support Internet of Things (IoT) Connectivity [1] such as massive machine type communication (mMTC) and ultra-reliable low-latency communication (URLLC). For a massive MIMO system to perform well, an accurate estimate of the wireless channel response has to be acquired. The traditional approach for channel estimation makes use of empirical assumptions about the wireless channel statistics which is sufficient for deriving theoretical results. However, they can be inadequate for practical purposes. In this work, we propose a data-driven approach for channel estimation using the multilayer perceptron (MLP) neural network. Such an approach should be valid irrespective of the propagation environment. We demonstrate that this approach significantly outperforms the conventional Minimum-Mean-Square-Estimator (MMSE) except for the high signal-to-noise ratio (SNR) regime at which the performance of MLP estimator starts to saturate. To deal with this problem, we propose a heuristic algorithm which switches from the MLP estimator to the MMSE estimator at the high SNR regime.