Intelligent Bandwidth Management Using Fast Learning Neural Networks
Fahad Ullah, Gul Muhammad Khan, Sahibzada Ali Mahmud · 2012
A fast learning neural network based scheduling system is presented to predict the frames on a single and multi-user MPEG-4 traffic and to distribute the bandwidth accordingly. MPEG-4 video stream traffic from various sources is used to evaluate the capability of this algorithm. A Fast learning Neural network algorithm also termed as Cartesian Genetic Programming Evolved Artificial Neural Network (CGPANN) is used as a forecaster to predict the size of the next frame based on the historical data consisting of previous 10 frames in the buffer for each individual user. A range of scenarios are exploited and analyzed for the frame size prediction error, bandwidth utilization efficiency and the frame drop rate for the whole system as well as every user involved obtaining outstanding results. For the best case, the system - with 50 users using the streaming service - has 35% of bandwidth efficiency with very low frame drop frequency.