Support vector machine for channel prediction in high-speed railway communication systems

In this paper, the problem of fast time-varying channel prediction is investigated in high-speed railway communication systems. A channel prediction algorithm is proposed based on a support vector machine (SVM) model. In order to further improve the prediction accuracy, the penalty coefficient and Gaussian kernel width of the SVM model are optimized by a genetic algorithm (GA). Simulation results show that the proposed prediction model based on both the SVM and the GA (SVM-GA) has lower prediction error than traditional auto-regressive (AR) and single SVM prediction models. In addition, when the effect of the noise on prediction performance is considered, the SVM-GA prediction model is superior to the AR and the SVM models in terms of normalized mean squared error.