Information Technology Reference
In-Depth Information
A Prediction Algorithm for Real-Time Video Traffic
Based on Wavelet Packet
Yingyou Wen
1,2
, Zhi Li
1,2
, Jian Chen
1
, and Hong Zhao
1
1
Northeastern University, Shenyang, 110003, China
2
Laboratory of Medical Image Computing, Shenyang, Liaoning, 110179, China
Abstract.
Long-term prediction is a key problem in real-time video traffic ap-
plications. Most of real-time video traffic belong to VBR traffic and has specif-
ic properties such as time variation, non-linearity and long range dependence. In
this paper, feature extraction method of real-time video traffic based on multi-
scale wavelet packet decomposition is proposed. On this basis, LMS algorithm
is adopted to predict wavelet coefficients. Through reverse wavelet transforms
of the predicted wavelet coefficients, the long-term prediction of real-time vid-
eo traffic is realized. Numerical and simulation results show that this long-term
prediction algorithm can accurately track the variation trend of video signal and
obtain an excellent prediction result.
Keywords:
real-time video traffic, wavelet packet, multi-scale decomposition,
long-term prediction, LMS.
1
Introduction
In recent years, the proportion of video traffic transmission in network is gradually
increased. Prediction of video traffic in coming period will do much help to improve
the quality of video transmission [1-3]. Therefore, study of real-time video traffic
prediction algorithm has important significance considering the requirements of effi-
cient bandwidth allocation. Traditional bandwidth analysis method for pre-encoded
video is not suitable for the analysis of real-time video traffic because signal encoding
method can not be obtained in advance.
Self-similarity and long-range dependence are essential to accurately traffic
prediction
[4]. Related studies have shown that the broadband network traffic
and video traffic both have nature of long-range dependence and self-similarity
in addition to short-range dependence [5, 6]. Some traffic prediction algorithms
have been proposed [7, 8], most of which belongs to short-term prediction
(1 to 10 frames), such as ARX and ARMA etc. Long-term prediction is one of
the most difficult problems in the area of video traffic prediction. In this paper,
we proposed an optimal multi-scale decomposition method of real-time video
traffic and realize a long-term prediction of decomposed video traffic based on LMS
algorithm.
