Dynamic spectrum allocation in hybrid networks with imperfect load prediction

The paper presents an investigation into the effects of imperfect load prediction on the performance of a dynamic spectrum allocation (DSA) scheme. Previous work has assessed the performance of DSA, but this has assumed that the DSA could make an accurate prediction of the future loads on networks sharing the spectrum. The paper investigates the performance of nine different load prediction schemes in adapting to seven different traffic scenarios. The results show how the performance of DSA can be improved with time-series prediction schemes combined with load histories, under unexpected traffic scenarios. In addition, the paper investigates when the prediction should be based on a time-series prediction algorithm or on a load history, and also the effect of the time interval between spectrum reallocations on the load prediction.