在無線網路中利用假設檢定達到偵測移動群組的機制

Abstract

本論文目的是在無線網路中，利用一群一起移動的行動裝置之間的距離、方向及速度來偵測出他們是一個群組，並利用統計中的假設檢定來增加偵測的正確率。隨著無線網路的快速發展，不論是利用基礎建設的WiMAX / 3G 系統，或是無需基礎建設的MANET（Mobile Ad Hoc Network） / VANET（Vehicles Ad Hoc Network）都是目前相當重要的研究議題。在WiMAX / 3G 系統上，各個行動裝置都需要與BS（Base Station）連線，再藉由BS連上網路。當行動裝置從一個地方移動到另一個地方時，為了維持穩定的連線，必須發起換手的程序將連線換到另一個訊號較好的BS。如果有一群行動裝置往相同的方向移動，例如在公車或是捷運上的一群乘客，他們會在同時間個別發起換手的程序，則其換手過程中所需耗費的控制訊息個數之總和是很大量的，這會造成換手延遲，甚至斷線。因此，如果我們可以偵測出一群一起移動的行動裝置，而且將他們視為一個群組，則我們就可以發起群組換手程序來取代個別換手程序，使handoff的負擔較低。在MANET / VANET上，行動裝置無法獲得整體的資訊。行動裝置只能藉由他與鄰居的相對位置及速度來偵測出哪一個鄰居是群組的成員。在MANET / VANET上，群組的資訊可以使散佈資料更有效率。如果可以識別出群組中的某一點，則我們可以將資料傳到群組中的一個點來取代將資料散佈到每一個點上。另外，行動裝置要能長時間的運作，其電量管理是必要的，而群組的資訊可以用來幫助電量管理。因為每一個行動裝置的電量都是由電池所提供，因此若是行動裝置可以知道他自己的群組成員，則他可以選擇使用耗電量較小的Short range Interface來溝通，而不需要一直使用耗電大的Long range Interface來溝通，這樣就能夠有效的節省電量。因此，無論是在基礎建設的WiMAX / 3G系統或是無需基礎建設的MANET / VANET下，群組偵測是很重要的研究。在本論文中，我們將探討如何利用群組內行動裝置之間的關係去識別出一群一起移動的行動裝置，並利用這樣的資訊將他們群組起來，這樣就可以改善網路的效能。在本論文中我們主要研究可以分為:(1)在基礎建設式網路下，已知Leader資訊的情況，以Centralized的方式偵測出群組，(2)在基礎建設式網路下，未知Leader資訊的情況，以Centralized的方式偵測出群組，(3)在隨意式網路下，以Distributed的方式偵測出群組。以達到在各種網路中，能夠有效率的分出群組的目的。最後利用模擬驗證所提出的方法。

The purpose of this thesis is to identify a set of mobile devices which are moving as a group in wireless networks. We use the statistical hypothesis test to improve the correct detection rate. With the rapid development of wireless networks, the advance research in both infrastructure based WiMAX / 3G systems, and Mobile Ad Hoc Networks (MANET) / Vehicles Ad Hoc Networks (VANET) are very important. In the WiMAX / 3G systems, every mobile devices access to the networks by connecting with Base Stations (BS). In order to maintain the constantly connectivity to the networks, a moving mobile device must handoff from a BS to other BS with better coverage. However, if a group of mobile devices in the same direction, such as a group of passengers in the bus or a rapid train, they may initiates a handoff process individually at the same time. As a result, the number of handoff messages, which the BS needs to handle simultaneously are enormous. It may cause a handoff delay or even multiple calls drops. Therefore, if we can detect a group of mobile nodes in advance and treat them as a group instead of individual nodes, then we can initiate a group handoff to trim down total handoff overhead for a mobile group, In MANET / VANET, a mobile device is unable to obtain global information such as positions, direction and velocities of all nodes. A mobile device detects its group member only by local information of its neighbored nodes. In MANET / VANET, group knowledge can help to spread information more efficiently. We can send a copy of information to one node of each group of mobile devices instead of spread information to every node if a group of nodes can be identified. Furthermore, group knowledge can help on energy management, which is an essential fact for the long duration operation of the MANETs. Since a mobile device of the MAVET is powered by the battery, if a node can know its group member, it can communicate with its group member using a short range interface instead of a more power hungry long range interface. In addition, we can use our method to implement cluster based algorithm for efficient routing and load balancing in MANET / VANET. Therefore, the group detection is an important research for both infrastructures based WiMAX / 3G systems, and MANET / VANET. In this thesis, we will explore the relationship between mobile devices to identify a group of nodes which travel as a group more accurately. We can make use of the group mobility to improve network performance. In this thesis, we focus on the following three topics: (1) Deign a centralize algorithm to detect a set of nodes which are traveling together in infrastructure based networks with a group lead. (2) Deign a centralize algorithm to detect a set of nodes which are traveling together in infrastructure based networks without a group lead. (3) Deign a distributed algorithm to detect a set of nodes which are traveling together in non-infrastructure based networks. Finally, simulation results show that our proposed approaches have a good group detection rate.