Eigenvector Prediction: A Low-Complexity Approach to Channel Prediction in 5G Sub-6 GHz Multiuser Massive MIMO

In this paper, we investigate a low-complexity method to mitigate the effects of channel aging in 5G new radio (NR) downlink (DL) multiuser multiple-input multiple-output (MU-MIMO) system by prediction of eigenvectors of the channel matrix. In a typical uplink channel estimation procedure of 5G NR using sounding reference signals (SRS), the channel is estimated at discrete time intervals. Due to time-varying nature of the channel, the estimates become outdated with time. Thus, use of the stale channel estimates renders inaccuracy to the DL beamforming (BF) weights, resulting in degraded spatial multiplexing and throughput deterioration. Conventional channel prediction techniques predict the raw channel matrix in the time-slots where no SRS transmission is scheduled. However, this necessitates eigenvalue decomposition (EVD) of the predicted channel matrix in every DL time-slot for calculation of DL BF weights. Our eigenvector prediction method directly predicts the eigenvectors from the estimated channel matrices, thus eliminating the need for EVD in every DL slot which dramatically reduces the implementation complexity as compared to the conventional system relying on prediction of raw channel estimates. We evaluate the method using a 5G NR system level simulation employing the 3GPP’s 3D urban micro channel model, demonstrating improvement in system throughput at negligible additional computational cost.