A hybrid model for congestion prediction in HF spectrum based on complete ensemble empirical mode decomposition

An improved hybrid model has been proposed in this paper. The hybrid model combines AR model with Volterra series expansion and adopts Complete Ensemble Empirical Mode Decomposition as preprocessing step for predicting congestion in high-frequency spectrum. In this model, we decompose original intricate spectral congestion series into several simpler components; AR model and Volterra series expansion are used to model the relatively stationary Intrinsic Mode Functions (IMFs) and the residue with tendency, respectively; LMS algorithm is employed to modify AR's and Volterra's coefficients. The effect of the order of the model on prediction performance has been investigated and we compare performance of the model with stand-alone use of AR model, Volterra adaptive filters and SVM for one-step prediction. The results have demonstrated that the hybrid model enhances the accuracy of prediction to behaviors of spectrum driven from nonlinear and non-stationary processes.