Summary: | 碩士 === 國立臺灣大學 === 資訊管理研究所 === 90 === GPRS is a better solution to mobile data transfer before the coming of 3G network. By using packet-switched technique, one to eight time-slot channels can be dynamically assigned to GPRS users on demand. Because GSM and GPRS users share the same channels and resources, the admission control of different type of traffics is needed to optimize the data channel allocation. The methodology of dynamic slot allocation plays an important role on both maximizing the system revenue and satisfying users’ QoS requirements. We try to find out the best methodology to get the maximum system revenue considering the call blocking probability constraints.
We propose two mathematical models to deal with the slot allocation problem in this thesis. The goal of our model is to find a slot allocation policy to maximize the system revenue considering the capacity and the call blocking probability constraints. The main difference between two models is the time type. The first model is discrete-time case, and the other is continuous-time case. Markovian decision process can be applied to solve the problem of maximizing system revenue without the call blocking probability constraints. In order to take the call blocking probability constraints into account, we propose three approaches: linear programming in Markovian decision process, Lagrangian relaxation with Markovian decision process, and the expansion of the Markovian decision process. The most significant and contributive part of this thesis is that we combine the Lagrangian relaxation and the Markovian decision process and use it to successfully solve the Markovian decision process with additional constraints.
The computational results are good in our experiments. We can find a slot allocation policy to maximize the system revenue under the call blocking probability constraints. Compared to the policy that the vendors often used, the policy we found has great improvement in system revenue. Thus, our model could provide much better decisions for system vendors and network planners.
|