MPEG-4 Fine Granularity Scalability (MPEG-4 FGS) encoded video stream provides fine granularity scalability, it can adapt to the transmission of various network bandwidth. Therefore, it is becomingthe main part of video streaming at present, and video traffic modeling for MPEG-4 FGS is getting more and more significant to network performance simulation and communication network design.This paper proposes a new video traffic model based on the MPEG-4 FGS encoded video. The model is presented on the basis of the statistical characteristics of MPEG-4 FGS encoded video. It can model both the video traffic with individual scene and that with scene changes. AR(1) Model is used to model the video with individual scene, Markov Modulated AR(1) Model is used to model the video with frequent scene changes. To model the video with individual scene, firstly, separates the whole trace file into traces, one for each frame type. Then, use AR(l) to model these traces with the same frame type and interleave to generate the modeled video traffic according the GOP mode. This paper proposes a method to allocate video rate dynamically. For video with frequent scene changes, the video sequence is first segmented into scenes by difference detection based on GOP. Individual scenes are then classified into classes by K-means clustering, and each class is modeled by AR(1) model. Finally, a Markov chain is used to modulate the transition from one scene class to another.Based on the comparison between real video traffic and video traffic generated by the proposed model, we found the first-order and second-order statistics of modeled video traffic to be in excellent agreement with their empirical counterparts. In the end, NS-2(Network Simulator-2) is exploited to simulate the modeled video traffic. The experimental results show that the proposed model can generate approximately the same size of each video frame and the correlation as MPEG-4 FGS encoder does.Moreover, this paper implemented a video traffic generator embedded in NS-2 according to the model proposed in this paper. And it is applied to network simulation. The proposed video traffic model can be extended to model various video traffics and various layered-encoded streams.
|