Extending the prediction horizon in dynamic bandwidth allocation for VBR video transport
Chapter in Scopus
- Additional Document Info
- View All
This paper presents a dynamic bandwidth allocation system for real-time variable bit rate (VBR) video transport in asynchronous Transfer mode (ATM) networks. This system takes advantage of scene changes in the video trace on a scale larger than a second, and it adapts the bandwidth as needed. An improvement on efficiency is achieved by assigning bandwidth for the transport of VBR video and having more than one predictor in parallel with different prediction horizons, hence this scheme reduces the processing time for the bandwidth adaptation with no significant degradations on the queue statistics. The link capacity required for a specific session is a function of the input trafic, which is characterized by its spectral characteristic. In particular, we use the low frequency band of the power spectrum, which is extracted from the stochastic input by a low pass filter. The predictor is a neural network (NN) called "Pi-Sigma Network", and the output of this predictor is interpreted according to the prediction horizon in use. © Springer-Verlag Berlin Heidelberg 2000.