Maximizing througput with adaptive M-QAM based on imperfect channel predictions

Uncoded adaptive M-QAM transmission over flat Rayleigh fading channels is here optimized in a novel way in the presence of channel prediction errors. The modulation rate is determined based on the predicted channel state. The modulation rate limits are adjusted by maximizing the throughput in error-free link-level frames, averaged over the pdf of the true channel state. No bit error rate constraint is imposed. This approach is appropriate when fast link-level retransmissions can be used to attain required error levels. The resulting scheme is evaluated analytically in a multiuser environment where predictive link adaptation is used in combination with a scheduling strategy that provides multiuser diversity gain. Prediction errors typical of prediction 1/3 wavelength ahead in space will then result in only 8%-12% reduction in the spectral efficiency, as compared to a case with perfectly known channels. The resulting performance is very robust with respect to the prediction error variance assumed when optimizing the rate adaptation scheme.