doi:10.1016/j.peva.2007.02.004
Copyright © 2007 Elsevier Ltd All rights reserved.
A new model for video traffic originating from multiplexed MPEG-4 videoconference streams
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Aggelos Lazarisa, 
, Polychronis Koutsakisb, 
, and Michael Paterakisa,
aDepartment of Electronic and Computer Engineering, Technical University of Crete, Greece
bDepartment of Electrical and Computer Engineering, McMaster University, Canada
Received 15 May 2005;
revised 5 February 2007.
Available online 17 February 2007.
Abstract
Due to the burstiness of video traffic, video modeling is very important in order to evaluate the performance of future wired and wireless networks. In this paper, we first study the behavior of single MPEG-4 videoconference traces and investigate the possibility of modeling this type of traffic with well-known distributions. Our results show that the Pearson type V distribution is the best fit among all the examined distributions, for all the traces under study. However, the behavior of single videoconference traces can never be perfectly “captured” by a distribution generating independently frame sizes according to a declared mean and standard deviation, due to the high autocorrelation of videoconference; therefore none of the fitting attempts can achieve high accuracy. Still, our results on attempting to model single MPEG-4 videoconference sources provide significant insight and help to build a Discrete Autoregressive (DAR(1)) model to “capture” the behavior of multiplexed MPEG-4 videoconference movies from VBR coders. Based on our results and on comparisons with other existing approaches, we discuss the contribution of our proposed method to the field.
Keywords: Videoconferencing; MPEG-4 video encoding; Video traffic modeling; Discrete autoregressive model
Fig. 1. Histogram for the frame size of the lecture camera trace.
Fig. 2. Q–Q plot for the ARD Talk I frames.
Fig. 3. Q–Q plot for the ARD Talk P frames.
Fig. 4. Q–Q plot for the ARD Talk B frames.
Fig. 5. KS-test (Comparison percentile plot) for the ARD Talk I frames.
Fig. 6. KS-test (Comparison percentile plot) for the ARD Talk P frames.
Fig. 7. Autocorrelation function of the N3 Talk trace.
Fig. 8. Autocorrelation function of the office camera trace.
Fig. 9. Comparison for a single trace between a 2000 frame sequence of the actual I frames of the ARD Talk trace and the respective DAR(1) model in number of cells/frame (Y axis).
Fig. 10. Comparison for a single trace between a 2000 frame sequence of the actual P frames sequence of the ARD Talk trace and the respective DAR(1) model in number of cells/frame (Y axis).
Fig. 11. Comparison for a single trace between a 2000 frame sequence of the actual B frames sequence of the ARD Talk trace and the respective DAR(1) model in number of cells/frame (Y axis).
Fig. 12. Comparison for 20 superposed sources between a 2000 frame sequence of the actual P frames sequence of the ARD Talk trace and the respective DAR(1) model in numbers of cells/frame (Y axis).
Fig. 13. Q–Q plot of the DAR(1) model versus the actual video for the B frames of the N3 Talk trace, for 20 superposed sources.
Fig. 14. Q–Q plot of the DAR(1) model versus the actual video for the P frames of the N3 Talk trace, for 20 superposed sources.
Fig. 15. Q–Q plot of the DAR(1) model versus the actual video for the I frames of the N3 Talk trace, for 20 superposed sources.
Fig. 16. Q–Q plot of the DAR(1) model versus the actual video for the I frames of the ARD Talk trace, for 20 superposed sources.
Fig. 17. Autocorrelation vs. number of lags for the I frames of the actual ARD Talk trace and the DAR(1) model, for 20 superposed sources.
Fig. 18. Autocorrelation vs. number of lags for the I frames of the actual N3 Talk trace and the DAR(1) model, for 20 superposed sources.
Fig. 19. Autocorrelation vs. number of lags for the P frames of the actual N3 Talk trace and the DAR(1) model, for 20 superposed sources.
Fig. 20. Autocorrelation vs. number of lags for the B frames of the actual N3 Talk trace and the DAR(1) model, for 20 superposed sources.
Fig. 21. Waiting time cdfs for the office camera traces and the model, for 10 superposed sources (offering a 20% average channel load). Unlimited buffer size.
Fig. 22. Waiting time cdfs for the office camera traces and the model, for 20 superposed sources (offering a 40% average channel load). Unlimited buffer size.
Fig. 23. Waiting time cdfs for the N3 Talk traces and the model, for 20 superposed sources (offering a 55% average channel load). Unlimited buffer size.
Fig. 24. Video packet dropping ratio versus offered load from superposition of N3 Talk traces. Unlimited buffer size.
Table 1.
Statistics for the I,P,B frames of the ARD Talk trace
Table 2.
Statistics for the I,P,B frames of the ARD Talk trace
Table 3.
Lag-1, lag-2 and lag-3 autocorrelation for various superposed video frames’ types
Corresponding author. Tel.: +1 9055259140x26542; fax: +1 9055212922.
Abstract
Purchase PDF (2318 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (2)
Purchase PDF (158 K)
