Nonparametric Regression for MU-MIMO Channel Prediction: From KNN to Local Linear Regression

Channel aging poses a huge challenge for the MU-MIMO (Multi-user Multiple-Input-Multiple Output) communications. To alleviate the problem, channel prediction is widely regarded as a promising mean. Existing channel prediction methods usually rely heavily on the assumption of parametric models. Severe performance degradation might occur when modeling error exists, which is unfortunately common in practice due to the complicated nature of wireless channels. To address the problem, we propose to use nonparametric regression methods for prediction. Nonparametric methods enjoy the advantage of not relying on the model assumption at all, enabling it very suit for making inference from complex channel data. In this paper, we present two nonparametric regression methods for MU-MIMO’s channel prediction, i.e., k-nearest neighbors (kNN) regression and its improved version, local polynomial regression (LPR). In addition, we propose to employ the Bayesian Information Criteria (BIC) to select the parameters in LPR, whereby we get the conclusion that local linear regression is recommended in practical applications. Furthermore, for the case of predicting the right singular matrix of the channel in specific, we provide a useful preprocessing procedure. Simulation results illustrate that our proposed nonparametric regression methods can outperform significantly the conventional parametric methods for channel prediction.