Use of optimal wavelet packet decomposition for the long-term prediction of variable-bitrate video traffic

This study systematically investigates the long-term prediction of online, real-time variable-bit-rate (VBR) video traffic, which is a key and complicated component of advanced predictive dynamic bandwidth control for future networks such as multimedia satellite networks. As per the time variation, non-linearity and long-range dependence in the VBR video traffic trace, a novel method of feature extraction based on multi-scale decomposition is proposed. After an analysis of the time-frequency distribution characteristics of the video trace, the wavelet packets that have the trait of arbitrary distinction and decomposition are selected. After space partition of wavelet packets, the best wavelet packet basis for feature extraction is picked. Based on the best basis, fast arbitrary multi-scale wavelet packet transform (WPT) can be done and each higher dimension wavelet coefficient matrix can be obtained. Then, wavelet coefficients prediction is done based on normalised least mean square error (NLMS) or least mean-support vector machine (LS-SVM) algorithm. The long-term prediction of the VBR video traffic is obtained through reverse wavelet transforms on the predicted wavelet coefficients. Numerical and simulation results are provided to validate the claims.