Channel Gain Prediction in Wireless Networks Based on Spatial-Temporal Correlation

The location information of a mobile terminal has become more available due to the popularity of GPS-aided smartphones. In this paper we propose a general predictive model for multistep-ahead average channel gain that exploits both spatial and temporal correlation, aided by the location information. The proposed Bayesian framework is composed of an autoregressive process and a multivariate Gaussian process, based on our statistical analysis on the average Rayleigh fading channel. Numerical results shows that the proposed algorithm achieves more accurate prediction than the autoregressive time series model, especially for long-range forecasting; and more robust prediction against inaccurate information of location than the support vector machine-based model.