Study On Mobility Prediction Over Wireless LAN And Its Applications

With the development of wireless technology and kinds of wireless applications, especially those need QoS support,mobility prediction techniques become more and more important. Mobility prediction impacts greatly on different aspects of wireless networks including reducing handoff delay of mobile hosts, improving admission control and flow control in wireless networks, reducing energy consumption of wireless network, improving the multicast routing protocols, and so on. Now the research on mobility prediction includes two different directions: one uses real time coordinates to make prediction with the assistance of GPS, while the other only uses user mobile history, which contains the sequence of APs the mobile host has associated with. The latter is the focus of this research paper. The main contents and conclusions in this dissertation can be summarized as follows:â… . Techniques for mobility prediction in WLAN(1) Among mobility prediction models in WLAN, Markov models are important for mobility prediction because of their simplicity and quite good prediction accuracy. But high-orders have the problem of high state-space complexity. To solve this problem, this paper puts forward a new Markov model named hybrid Markov model, which mixes several multi-step Markov models. This paper adopts EM algorithm to compute the mixing coefficients based on the Maximum Likelihood Principle. Error analysis and simulation experiments prove that the new hybrid Markov model can get almost the same mobility prediction accuracy as high-order Markov model while it overcomes the shortcoming of state-space expansion of high-order Markov models. Then, in order to get better mobility prediction accuracy, the paper puts forward a new dynamic hybrid Markov model based on the hybrid Markov model. An integration scheme is also provided based on the dynamic hybrid Markov model. These two models have both proved to have better mobility prediction accuracy than the hybrid Markov model.(2) By taking the mobility prediction problem as a time sequence prediction problem, BP neural network and Elman neural network are used to deal with it. Then an improved Elman neural network is given to get better prediction accuracy, which has the output-input feedback mechanism. The stability of the model is proved by using discrete-type Lyapunov stability analysis. The simulation experiments proved its better mobility prediction accuracy.(3) In fact, the mobility patterns are key factors in the mobility prediction problem. Therefore, the paper brings forward a new mobility prediction scheme based on data mining, especially on time sequence mining. At first, regular mobility patterns are picked out by using data mining techniques. Then by pattern matching, mobility prediction is made. In order to trace the newer mobility patterns, an incremental data mining technique is used to enhance the whole scheme, especially to improve its online prediction ability. Simulation experiments proved that the incremental data mining technique really promoted the prediction accuracy of the scheme.(4) The forementioned methods are all based on statistical principles, and focus on the efforts to get the mobility patterns. They all ignored the point that the intention of the mobile host is crucial for the real mobile track it will experience. So this paper tries to use an hidden Markov model to improve the prediction accuracy. Firstly, it takes the intention of the mobile host as the hidden state of the HMM. Secondly, it takes the real mobile paths which are possible under the current intention of the mobile host as the observation values. Thus, an HMM is set up for the mobility prediction problem. Then, a mobility prediction algorithm is put forward based on the HMM. The algorithm is composed of two steps, forward prediction and backward probability modification. In the scheme, how to decide the intention of the mobile host is very important This paper provides two basic methods to get the intention of the mobile host. But it's not enough and there is still a long way to go to get some satisfying results.â…¡. Study on related applications of mobility predictionThere is no doubt that mobility prediction acts as a key role in current wireless networks. It has been pointed out by very many papers and other documents. Mobility prediction can be used to improve performance in a lot of different fields of wireless environment. Aiming at the MAC handoff delay, especially the channel scanning delay which accounts for 90% of the whole handoff delay, this paper puts forward a new WLAN MAC handoff scheme based on mobility prediction information. By integrating the mobility prediction and WLAN MAC handoff problem reasonably and effectively, the scheme greatly reduced the channel scanning delay in WLAN MAC handoffs, thus reducing the whole handoff delay. Theoretical analysis and simulation experiments all proved the remarkable effect on the cut-down of WLAN MAC handoff delay. This improvement is important for the QoS in wireless networks.In order to solve the state-space expansion problem of high-order Markov models, this paper provides a new method which substitutes high-order Markov models by a hybrid Markov model. It is significant for similar problems in many other fields. Then the paper adopts three different techniques to study mobility prediction, which are helpful for the progress of mobility prediction. Also the paper puts forward a new method to reduce the channel scanning delay by using mobility prediction information, which provides a new view for reducing handoff delay in WLAN.With the development of Pervasive Computing and applications supporting wireless QoS, mobility prediction will become more and more important and contribute more to the future wireless networks. The results of the dissertation will contribute to the study on mobility prediction algorithms and its applications.