A Comparison of Wireless Channel Predictors: Artificial Intelligence Versus Kalman Filter

Accurate channel state information (CSI) is a prerequisite to reap the benefits of fading-adaptive wireless communications. In practice, however, the available CSI is generally outdated due to processing and feedback delays, which deteriorate system's performance severely. Channel prediction that is able to forecast future CSI provides a promising solution. In addition to statistical methods, namely modelling a time-varying channel as an autoregressive process and using a Kalman filter to predict, artificial intelligence techniques with the capability of time-series prediction are also being discussed recently. This paper compares performance and complexity of these two kinds of predictors. The numerical results on prediction accuracy measured by mean squared error in both noiseless and noisy Rayleigh fading channels, together with their achieved performance in a transmit antenna selection system, are comparatively illustrated.