| Summary: | 碩士 === 元智大學 === 資訊工程學系 === 94 === Due to a large number of picocells and different user mobility in multimedia wireless networks, the probability of handover becomes higher and the traffic pattern can be arbitrary with time and following a non-Poisson and non-stationary process. As a result, without enough available resource for request call arrival, it may cause much higher blocking rate. Hence, how to reserve and control the resource allocation is one of most important issues in next generation wireless networks.
This thesis proposed a predictive resource reservation method with time-space series model for estimating resource requirement of handover call and new call from different service classes. Using local information and recent history from a spatial neighborhood of observed cells, a time-space series model is utilized to perform a traffic prediction. First, Seasonal-ARIMA(SARIMA) model is adopted to fit the actual traffic data measured in the reference cells. Then, spatial dependencies between the target cells and its neighbor cells are incorporated into this model for actual network environment. Due to spatial and temporal dependencies are modeled simultaneously for approximating and forecasting traffic, the prediction model is solved using artificial neural network. Based on time-space series prediction model, we estimate the resource requirement with online traffic predictor, it could be more efficient for supporting resource forecast, control, and allocation in the variable wireless networks, and the system performance could be increased by using the precise reservation levels.
In order to demonstrate the difference performance of our mechanism between those of several reservation methods, a service-dependent system model is adopted to integrate real-time and non-real-time services for call admission control (CAC). The performance indicators of our evaluations in terms of handover request call blocking probability (HDP) and the new call dropping probability (NBP) are represented. The results show that it performs well at heavy traffic load. The total reduction in call blocking rate is about 7~26%. It can also keep the total call dropping rate below 5%. At various call arrival rates, the bandwidth utilization can be increase 11%. However, using an unsuitable update interval may cause worst performance for fixed PDC scheme, even if in PDC scheme, it may causes heavy system overhead for repeated prediction.
|
|---|
