Optimizing Training Lengths and Training Intervals in Time-Varying Fading Channels

In time-varying faded channels the transmissions are organized into frames where the channel estimation is mainly training-based. The optimal design of the training structure is formulated here by finding the training length (the optimal number of contiguous pilots) and the training interval (the interval among two successive training phases) to maximize system throughput. The optimal balance of training and payload depends on the combination of Doppler frequency and frame length. The level of the signal to noise ratio and the fading dynamics constrain the quality of the estimate from training. It is shown that the length of the training can be conveniently traded for lower training intervals to reduce the estimate out-dating. For fast-varying fading and for high enough signal to noise ratio, there is a definite advantage in fragmenting the frame with dispersed segments of training symbols of smaller length rather than having a highly reliable channel estimate by concentrating all the training symbols at the beginning of the frame. Extensive simulations corroborate the design criteria. System throughput is maximized either for noisy binary transmission and for Gaussian input symbol distribution (i.e., by using information theoretic analysis).