Video source traffic flow prediction using neural networks

Prediction of traffic generated by multimedia sources can facilitate effective dynamic bandwidth allocation and implementation of quality-of-service (QoS) control strategies at the network edges. The time-series representing frame or VOP sizes of an MPEG-coded stream is extremely noisy and it has very long-range time dependencies. This paper proposes an approach to develop predictors for single-step-ahead (SS) and multi-step-ahead (MS) prediction of MPEG-coded real-time video traffic. The designed SS predictor consists of a recurrent neural network for I-VOPs and two feedforward neural networks for P- and B-VOPs, respectively. The SS prediction scheme is used for frame-by-frame prediction. A moving average of the frame or VOP sizes time-series is generated from the individual frame or VOP sizes and used for both SS and MS prediction. The designed MS predictor consists of a recurrent neural network and it is used to perform two-step-ahead and four-step-ahead prediction. Two-step and four-step-ahead prediction correspond to MS prediction horizons of 1 and 2 seconds, respectively