We analyze the application of the vector channel synthesis and prediction filter introduced by Arredondo et al.. Ray-tracing simulations show that predictions of the downlink channel can result in improvements in signal-to-interference ratio in excess of 10 dB in nonline-of-sight scenarios, compared with beamforming without predictions.

Many researchers have done significant work to reduce fast fading in single channel wireless systems using prediction. We introduce a novel synthesis and prediction filter at the smart antenna base station to predict the vector channel in time-division duplex systems. We show the advantage of modeling each component of the vector channel with the same coefficients over modeling each channel separately. Experimental results from measurements taken in downtown Austin show that prediction of the vector channel is feasible, even as far as ten steps ahead. Ray tracing simulations of downtown Austin, assuming noiseless line-of-sight and nonline-of-sight channels, show that these predictions enhance downlink beamforming resulting in improvements in downlink received power in excess of 10 dB in nonline-of-sight scenarios compared to beamforming without predictions. Furthermore, this improvement allows an increase in the duplex interval by more than three times, assuming constant mobile velocity.

Researchers have studied the behavior of the vector channel in order to study the quality of beamforming methods and to increase downlink signal power and signal-to-interference ratio. Jeng (see IEEE Trans. Antennas Propagat., vol.46, p.749-57, 1998) showed that knowledge beyond simply directions of arrival of the vector channel at one frequency may not be useful at another. We propose that if the ratio between two frequencies, f/sub 1/ and f/sub 2/, is close to one, then the autoregressive (AR) coefficients which model the behavior of the vector channel at f/sub 1/ are essentially the same AR coefficients as those which model the vector channel at f/sub 2/. This implies that accurate predictions of the vector channel at f/sub 2/ can be performed knowing only the channel at f/sub 1/, as well as enough knowledge of the channel at f/sub 2/ in order to provide initial conditions. Ray-tracing simulations of downtown Austin, TX, show that this hypothesis holds for f/sub 1/ = 1.85 GHz and f/sub 2/ = 1.93 GHz, which correspond to uplink and downlink frequencies, respectively, in US personal communications systems.