Channel Prediction Using Ordinary Differential Equations for MIMO Systems

Abstract

Channel state information (CSI) estimation is part of the most fundamental problems in 5G wireless communication systems. In mobile scenarios, outdated CSI will have a serious negative impact on various adaptive transmission systems, resulting in system performance degradation. To obtain accurate CSI, it is crucial to predict CSI at future moments. In this paper, we propose an efficient channel prediction method in multiple-input multiple-output (MIMO) systems, which combines genetic programming (GP) with higher-order differential equation (HODE) modeling for prediction, named GPODE. In the first place, the variation of one-dimensional data is depicted by using higher-order differential, and the higher-order differential data is modeled by GP to obtain an explicit model. Then, a definite order condition is given for the modeling of HODE, and an effective prediction interval is given. In order to accommodate to the rapidly changing channel, the proposed method is improved by taking the rough prediction results of Autoregression (AR) model as a priori, i.e., Im-GPODE channel prediction method. Given the effective interval, an online framework is proposed for the prediction. To verify the validity of the proposed methods, We use the data generated by the Cluster Delay Line (CDL) channel model for validation. The results show that the proposed methods has higher accuracy than other traditional prediction methods.