Prediction Algorithms for Real-Time Variable-Bit-Rate Video

Accurate prediction of variable bit rate (VBR) video traffic can be used to improve the network utilization efficiency while supporting guaranteed QoS requirements of VBR video. On-line prediction algorithms have been proposed in the literature to forecast real-time VBR video traffic for dynamic bandwidth allocation. In this paper, we survey a number of algorithms both in time domain and wavelet domain for video traffic prediction. The features of the existing algorithms are summarized, and on the basis of it we propose a time-domain and a wavelet-domain normalized least mean square (NLMS) based adaptive prediction scheme respectively. Our proposed time-domain scheme combines the separation and differential techniques in the literature to reduce short-term bit rate variation of VBR video traffic and smooth the data for more accurate perdition. Our proposed wavelet-domain prediction scheme uses the a trous wavelet transform instead of conventional decimated wavelet transform to improve the prediction accuracy by exploiting the redundant information in the wavelet transform coefficients. Simulations using three half-an-hour long full-motion moving picture experts group (MPEG) video traces show that our proposed methods can achieve better performance than those in the literature