Machine Learning-Based Channel Prediction Exploiting Frequency Correlation in Massive MIMO Wideband Systems

Abstract

In this paper, we focus on the channel prediction for massive multiple-input multiple-output (MIMO) wideband systems. In massive MIMO systems, the channel prediction is necessary to deal with out-dated channel state information (CSI). A wideband channel is converted into parallel narrowband channels via the orthogonal frequency division multiplexing (OFDM) technique where there exists a frequency correlation among narrowband channels. Thus, we propose a machine learning (ML)-based channel prediction technique, which exploits the frequency correlation. Numerical result shows that, under certain scenarios, the channel predictor trained with a single narrowband channel can support other narrowband channels.