MPEG-4 Traffic Prediction Using Density Estimation for Dynamic Bandwidth Allocation in IEEE 802.16 Networks

Efficient transmission of variable-bit rate (VBR) video traffic in Broadband Wireless Access (BWA) networks is currently an active research topic. Due to the dynamic changes in bandwidth requirements of VBR video and the limited network bandwidth, dynamic bandwidth allocation (DBA) is required. Traffic prediction is a promising approach to improve the effectiveness of DBA in BWA networks. In this paper, we propose DEEP (Density Estimation basEd Predictor), a novel prediction scheme for MPEG traffic. In DEEP, the density of probability of MPEG frames is estimated through kernel density estimation method. This density is then used in forecasting the future bit rate of I, P, and B video frames. Simulation results show that DEEP is able to predict the bit rate of MPEG traffic more accurately than the conventional Least Mean Squares (LMS) algorithm, and is less sensitive to traffic variation. We provide application guidelines of this predictor in the IEEE 802.16 standard.