Accurate Prediction of Real-Time MPEG-4 Variable Bit Rate Video Traffic

Abstract

In this letter, we propose a novel algorithm to predict MPEG‐coded real‐time variable bit rate (VBR) video traffic. From the frame size measurement, the algorithm extracts the statistical property of video traffic and utilizes it for the prediction of the next frame for I‐, P‐, and B‐ frames. The simulation results conducted with real‐world MPEG‐4 VBR video traces show that the proposed algorithm is capable of providing more accurate prediction than those in the research literature.

Variable bit rate video traffic prediction based on kernel least mean square method

“…The methods used for VBR traffic prediction can be classified into autoregressive (AR), moving average (MA), ARMA, AR integrated MA (ARIMA), fuzzy and neuro-computational. The AR model was used to predict the size of each frame types (I, P and B) separately in [5][6][7][8][9][10], the size of GOP in [6,11,12], the size of B frames from the previous sizes of P and B frames [13] and the size of each frame of an aggregate of videos in [9,14]. In the latter, the VBR traffic time series values are obtained by aggregation of several video frame size time series to obtain smoother traffic.…”

“…The MA model was used in [8,10] to forecast the next frame size based on the probability density function obtained by the cubic-spline interpolation. This method outperforms the AR model with the least mean square (LMS) method, neural network and adaptive network fuzzy inference system (ANFIS) predictors.…”

“…The LMS and NLMS algorithms are commonly used online and adaptive algorithms for finding AR model coefficients in linear adaptive systems [28,29]. In [7,9,12,14,15] the LMS, and in [5,6,8,10,11,30,31] the NLMS were utilised to calculate the coefficients. The use of a fixed step size in the LMS results in slow convergence [5-11, 14, 30, 31].…”