Self-similar Traffic Prediction Scheme For Satellite Internet Services

In contemporary, the prosperity of Internet promotes the rapid developmentof the Next Generation Network which forms the setup of air-space-groundintegrated network. As a kind of auxiliary network of terrestrial networks, thebroadband satellite networks which has advantages of global coverage, can releasethe pressure of terrestrial networks, however, it also faced a series of thornydifficulties to be solved. For example, the broadband satellite networks use TCP/IPprotocol as information carrier and transmission platform, the TCP protocol fromterriestrial to satellite requires smooth transition, on the other hand, the challengesof TCP/IP's QoS in ground networks still exist in broadband satellitecommunication network. Considering the complexity of ground-satellite hybridnetwork and the difficulties of integrated cyber source's distribution, thediscussion on broadband satellite Internet QoS becomes a necessary and importantresearch subject.Focus on Broadband Satellite Internet QoS requirements, the dissertationdeals with the following aspects:First of all, the dissertation detects that the aggregation and propagation oftraffic which is self-similar in satellite backbond networks, and then analyzes theself-similar traffic's properties which is transmits from wired ground Internet towireless satellite links through the ground gateway. Then, we build the ON/OFFmodel of self-similar traffic and the queuing model of Pareto/M/1/C to prove thatthe traffic in the satellite communication system is self-similar. Finally, we discussthat higher degree of similarity will lead to higher packet loss rates and biggerrequired bandwidth in the satellite communication system, which reduces theperformance of the communication system and its QoS.As the traffic on the satellite communication system are not available, weselect the appropriate self-similar traffic in ground network as a reference toestablish ARIMA traffic prediction model with the combination with wavelettransform. Under the guidance of the theory and Matlab simulation result, usingsym wavelet and3-layer wavelet decomposition to process the original networktraffic which is non-stationary, periodic and self-similar. After the pre-processingof the network traffic, we build the ARIMA prediction modle which uses100packets as prediction samples and predicts5values per step. This predictionmodel proposed in my dissertation can achieve the best forcasting results whichRRMSE is0.1773. The prediction results of this system are used on scheduling queue on satellitecommunicationg system and are provided to reliable reference data for cross-layerscheduling and bandwidth allocation of Satellite Internet. Accurate trafficprediction results can ease the network congestion and optimize the resouseallocation, and improve the performance of satellite network.