A MIMO Channel Prediction Scheme Based on Multi-Task Learning

This paper proposes a multi-input multi-output (MIMO) channel prediction scheme using multi-task learning algorithm. Based on the spatially correlated MIMO channel Channel State Information (CSI) observations, a multi-task least square support vector machine (MTLS-SVM) is trained, where the CSI prediction for each antenna pair can be modeled as one task and jointly learning between these tasks are implemented. Then the future CSI is predicted by this MTLS-SVM. By using the relatedness of the multiple tasks, the spatial correlations between different antenna pairs can fully be exploited and hence better channel prediction performance can be achieved compared with the single task prediction scheme.