Downlink Channel Prediction for Time-Varying FDD Massive MIMO Systems

Abstract

Accurate channel estimation is important for massive multiple-input multiple-output (mMIMO) to ensure good performance. Considering mMIMO application in 5G, frequency division duplexing (FDD) can provide higher data rate and wider coverage than the time division duplexing (TDD) mode. However, since the uplink/downlink channel is not straightforwardly reciprocal, FDD downlink channel estimation requires heavier training and computation than TDD mode due to the massive number of antennas. In addition, fast channel variation renders the real time estimation even more difficult. In this paper, we propose a downlink channel prediction scheme for FDD mMIMO, which requires only the TDD overhead. Specifically, the downlink channel matrix is represented by three components: steering matrix (frequency dependent), fading coefficients (time varying), and time delays (semi-static). By the proposed scheme, these three components can be obtained through uplink training. In addition, fast tracking and prediction is leveraged to obtain the real time channel state information. Simulation results show that accurate channel prediction is obtained via low cost and complexity by the proposed scheme.