Channel Prediction Using an Adaptive Kalman Filter

We present an adaptive Kalman filter based channel estimation and prediction algorithm for multicarrier systems with time-varying mobile radio channels. We prove that for a widesense stationary uncorrelated scattering channel in which P plane waves superpose at the receiver antenna, each current/future sample of the channel transfer function can be represented as a linear combination of at least P past channel transfer function samples. Based on this result, a vector auto-regressive model is used to model the dynamics of the time-varying mobile radio channel. The adaptive Kalman filter initially uses a state transition matrix derived from the Jakes' model for the temporal correlation and the one-sided exponential power delay profile for the spectral correlation between the channel transfer function samples. Afterwards, the state transition matrix of the adaptive Kalman filter is periodically updated by computing the correlation between the estimated channel transfer function samples. It is shown that the periodical update of the state transition matrix results in a significant prediction performance improvement over the one based on the correlation estimates using the Jakes' model and the one-sided exponential power delay profile. Furthermore, the proposed adaptive Kalman filter yields a satisfactory performance when used as an interpolation filter in the frequency domain even for cases where the channel transfer function is sampled at sub-Nyquist rate.