Radio Channel Prediction Based on Parametric Modeling

Doctoral thesis, 2007

Long range channel prediction is a crucial technology for future wireless communications. The prediction of Rayleigh fading channels is studied in the frame of parametric modeling in this thesis. Suggested by the Jakes model for Rayleigh fading channels, deterministic sinusoidal models were adopted for long range channel prediction in early works. In this thesis, a number of new channel predictors based on stochastic sinusoidal modeling are proposed. They are termed conditional and unconditional LMMSE predictors respectively. Given frequency estimates, the amplitudes of the sinusoids are modeled as Gaussian random variables in the conditional LMMSE predictors, and both the amplitudes and frequency estimates are modeled as Gaussian random variables in the unconditional LMMSE predictors. It was observed that a part of the channels cannot be described by the periodic sinusoidal bases, both in simulations and measured channels. To pick up this un-modeled residual signal, an adjusted conditional LMMSE predictor and a Joint LS predictor are proposed. Motivated by the analysis of measured channels and recently published physics based scattering SISO and MIMO channel models, a new approach for channel prediction based on non-stationary Multi-Component Polynomial Phase Signal (MC-PPS) is further proposed. The so-called LS MC-PPS predictor models the amplitudes of the PPS components as constants. In the case of MC-PPS with time-varying amplitudes, an adaptive channel predictor using the Kalman filter is suggested, where the time-varying amplitudes are modeled as auto-regressive processes. An iterative detection and estimation method of the number of PPS components and the orders of polynomial phases is also proposed. The parameter estimation is based on the Nonlinear LS (NLLS) and the Nonlinear Instantaneous LS (NILS) criteria, corresponding to the cases of constant and time-varying amplitudes, respectively. The performance of the proposed channel predictors is evaluated using both synthetic signals and measured channels. High order polynomial phase parameters are observed in both urban and suburban environments. It is observed that the channel predictors based on the non-stationary MC-PPS models outperform the other predictors in Monte Carlo simulations and examples of measured urban and suburban channels.