Neuro-Fuzzy Modeling and Prediction of VBR MPEG Video Sources

T he work presented in this paper intends to apply neuro-fuzzy methods for the modeling and prediction on trac intensity of digital video sources which are coded with hybrid Motion Compensation/Di€erential Pulse Code Modulation/Discrete Cosine Transform (MC/DPCM/DCT) algorithm. Although current coding standards recommend constant bit rate (CBR) output by means of a smoothing bu€er, the hybrid algorithm inherently produces variable bit rate (VBR) output. This paper describes the novel application of a fuzzy predictor for the purposes of modeling and prediction on video sources. The computation requirement of the fuzzy predictor and its neural network implementation are also discussed. The proposed fuzzy prediction method and its neural network version can be applied to the development of connection admission control, usage parameter control and congestion control algorithms in ATM networks.