Predictive adaptive loading for HIPERLAN II

Adaptive loading is able to significantly reduce the average symbol error rate (SER) of OFDM-based wireless local networks by exploiting the channel frequency diversity. In time-varying channels however these gains disappear quickly when the channel state feedback rate is limited. We show that by combining ML channel estimation with robust MMSE prediction filters, we can predict the optimal constellation sizes and regain most of the achievable gain of adaptive loading. With only 4 channel measurements spaced one MAC frame apart, the loss compared to the performance of adaptive loading with perfect channel knowledge is limited to 0.7 dB and a gain of 10 dB is achieved over non-adaptive OFDM for an SER=10^-6