A novel long-span traffic predictor for real-time VBR videos via ρ-domain rate model

To predict the traffic of frames that may be far from the current frame, this letter extends the use of the ρ-domain rate model from macroblock-based source rate control to frame-based long-span traffic prediction. Moreover, this work enhances the linearity and convergence speed of ρ-domain frame-based rate function by adding a parameter that is the number of nonzero motion vectors. Simulation results reveal that the proposed predictor can significantly lower the prediction error compared with two conventional LMS methods. More importantly, the process of the proposed predictor is unique but simple for different video contents and prediction spans.