Machine Learning-based Prediction of PMI Report for DL-Precoding in 5G-NR System

To apply a suitable precoder in a downlink (DL) MU-MIMO system, it is imperative to use a reference signal for estimating the DL channel quality at the UE which is then reported back to the gNB. The feedback report sent is used for precoder selection from the codebook at the gNB and applied for multiplexing different users for downlink data transmission. Since a typical wireless channel is fading in nature, it is important to estimate the channel at least once every coherence interval. The feedback step does result in significant improvement of the system performance but incurs large overhead and hence reduces the spectral efficiency of the system, especially in a fast-fading environment. Consequently, a high mobility user is not scheduled in the MU-MIMO set up so as to avoid large overheads and also to reduce inter-user interference. Therefore, to facilitate such scenarios, we consider the application of machine learning (ML) for predicting the channel quality reports and thus enabling MU-MIMO transmission for a high mobility user. In this paper, we analyze the performance of different ML algorithms on the prediction of CSI Type-II parameters and subsequently study their impact on the overall system performance w.r.t different periodicity of the feedback report. The obtained results are promising, such that improved system performance is achieved compared to the conventional approach in addition to enhanced spectral efficiency. Furthermore, simulation results corroborate the efficacy of the proposed machine learning techniques.