Variational Mode Decomposition Based Channel Prediction in Massive MIMO Systems

Accurate channel state information (CSI) is mo-mentous to achieve the immense benefits of massive multiple-input multiple-output (MIMO) systems. However, in mobility scenarios, the acquired CSI at the base station (BS) provides the outdated information, which results in dramatic performance degradation. Timely channel prediction is thus implemented to compensate for the performance loss caused by mobility. In this paper, a decomposition based channel prediction framework is presented by exploiting the Doppler structure information of the multipath channel. Furthermore, a specific variational mode decomposition (VMD) based multi-autoregressive (AR) prediction method, namely VMD-MAR, is proposed to process wireless channel with mixed time-varying patterns into several modes with distinct dynamics and then carry out the mode-level prediction. Due to the discrepancy in the dynamics between modes, the AR model built on each mode adopts its own AR order according to the Bayesian information criterion (BIC). Moreover, we theoretically analyze the physical meaning of modes from channel modeling perspective. Compared to the AR pre-diction method and empirical mode decomposition (EMD) based prediction method, experimental results show that the proposed VMD-MAR achieves superior channel prediction performance.