MPEG-4 Internet Traffic Estimation Using Recurrent CGPANN

Abstract

Stretching across the horizon of data communication and networking, in almost every scenario, accurate bandwidth allocation has been a challenging problem. From simple online video streaming to the sophisticated communication network underlying a Smart Grid, efficient management of bandwidth is always desired. One way to achieve such efficiency lies in the science of predication: an intelligent system can be deployed that can estimate the sizes of upcoming data packets by analyzing patterns in the previously received data. This paper presents such a system that implements a fast and robust Neuro-Evolutionary algorithm known as Recurrent-Cartesian Genetic Programming evolved Artificial Neural Network (R-CGPANN). Based on the previously received 10 MPEG-4 video frames, the system estimates the size of the next frame. The simulation results show that the recurrence in CGPANN measurably outperform not only the feed forward version of the said algorithm but other contemporary methods in the field. © Springer-Verlag Berlin Heidelberg 2013.