Massive MIMO channel prediction using machine learning: Power of domain transformation

Abstract

however, the time overhead of training phase directly affects the latency of the system. In this paper, we propose a novel ML-based channel prediction technique, which can reduce both the time overhead of collecting training data and training a neural network by transforming the domain of channels from subcarrier to antenna in wideband massive MIMO systems. Numerical results show that the proposed technique can not only reduce the time overhead but also improve channel prediction performance compared to the ML-based channel prediction techniques without domain transformation.

To compensate the loss from outdated channel state information in wideband massive multiple-input multiple-output (MIMO) systems, channel prediction can be performed by leveraging the temporal correlation of wireless channels. Machine learning (ML)-based channel predictors for massive MIMO systems were designed recently