藍芽個人無線區域網路上數據通訊問題之設計與分析

Abstract

藍芽（Bluetooth）技術係一新興崛起之個人區域網路標準，並預期在未來無線通訊領域發展中將扮演著重要的角色；藍芽當初原本是設計用來取代纜線設備（例如：滑鼠，鍵盤與個人電腦之連接線），但在後來的演進中發現，藍芽的應用面並不僅止於此；隨著市面上低價藍芽模組的陸續出現，大規模的藍芽設備佈署將很有可能在近期內實現；在本論文中，我們針對藍芽相關數據通訊議題，提出一系列的解決方案或分析結果，其細節分項敘述如下： 對於可攜式裝置（例如：Notebooks，PDAs，cellular phones）而言，一個重要的議題便是省電（power saving），由於體積小（compact size）係藍芽裝置之主要設計目標之一，在無法配備大體積高容量電池的情況下，省電問題在藍芽技術中更顯重要；藍芽係採主從裝置之網路結構（master-slave configuration），其基本的網路單元稱為微網（piconet），由於主裝置（master）係微網之中心控制設備，一般我們並不會讓主裝置進入睡眠狀態去省電，而另一方面，由於從屬裝置（slave）所扮演的角色責任較輕，因此我們可以讓此種裝置進入低功率省電模式，以節省不必要之電池能源消耗；在第一個議題中，我們深入探究藍芽低耗電監聽模式（sniff mode）之管理問題，在監聽模式下，從屬裝置（slave）可以進入睡眠狀態並週期性的醒來接收/傳送其與主裝置（master）間之數據資料；由於在藍芽規格裡主裝置與從屬裝置可以透過一些控制封包（control packet）來協調從屬裝置進入監聽模式之運作參數（sniff parameters），然而規格裡並無明確交代該如何設計這些參數，使得在數據流量需求（traffic requirement）與省電需求（power-saving requirement）之間能夠取得一個平衡，以達到整體網路效能提昇；因此，針對這個議題，我們提出了一個調整監聽模式相關參數之策略，讓主裝置與從屬裝置可以根據目前的數據流量，在不犧牲工作效能的條件下，動態協調出一組合適的監聽參數，使得從屬裝置能進入低功率之監聽模式以節省不必要的電池電源消耗，達到網路生命週期╱效能提昇之目的。 接下來在第二個議題中，我們繼續在微網（piconet）的環境下，探討主裝置詢問（poll）從屬裝置之機制策略；由於每個從屬裝置有著不同的數據流量，主裝置送往從屬裝置（master-to-slave；downlink）與從屬裝置送往主裝置（slave-to-master；uplink）邏輯連結之頻寬需求通常亦不對稱（asymmetric traffic），因此主裝置能否根據各從屬裝置之不同需求，來有智慧地執行詢問機制（polling scheme），以有效率不浪費的方式來分配有限頻寬給旗下所有從屬裝置，便是一個基本且重要的問題；在這項研究裡，我們提出了一個樣式配對（Pattern Matching）之詢問策略（polling policy），當主裝置與從屬裝置的數據流量穩定且平均流率可被預估時，我們的方法將定義一組最有效率的詢問樣式（polling pattern），並精心排程相對應的詢問時間點（polling timings），其中詢問樣式代表的是一個有序數列，定義了主裝置詢問（polling）與從屬裝置回應（replying）所必須使用的數據資料封包型態（data packet type），由於藍芽規格裡主要提供了三種數據封包大小，分別是佔1個，3個，及5個時槽（1-/3-/5-slot），每一個封包大小所能載送之資料量並不同，藉由把不同數據封包的選擇設計進主裝置之詢問機制（polling scheme）中，則上╱下行（up-╱down-link）之不對稱數據交通量的問題，將可獲得有效的解決；我們的目標是希望減少負載欄位（payload field）沒填滿，或甚至空的數據封包，期望更有效率地使用每一個藍芽時槽（time slot），藉此提昇藍芽微網最大可容納之生產量（throughput），使得更多的通訊活動能夠同時被支援。 將數個微網連結起來可以形成一個更大的網路結構，稱為散網（scatternet），然而，在藍芽現有的標準規格裡，散網的結構與運作方式並未被清楚定義，端視各家廠商如何設計﹔因此，一個簡單且具通訊效率的藍芽散網拓樸結構能否被提出，便成為藍芽是否能成為個人無線區域網路（Wireless Personal-Area Network，WPAN）上主要標準技術元件的關鍵所在;在第三個議題中，我們提出一個藍芽散網之環形結構，稱為藍環（BlueRing），係將數個微網以環狀的方式串接起來，連接微網與微網之藍芽主機稱為橋接器（bridge），橋接器負責跨微網（inter-piconet）之封包轉送；除此以外，在這項研究裡，亦清楚交待藍環（BlueRing）之形成（formation）、繞徑（routing）、及維持管理（maintenance）機制，經過初步的模擬驗證，藍環之散網結構不論是在通訊效能（communication efficiency）及容錯能力（fault tolerance）上，均有優於先前提出之其它拓樸結構的良好表現。 最後，我們研究在多個微網共存的環境（multi-piconet environment）中，從數學機率分析的角度，來評估因為共同頻道干擾（co-channel interference）而導致封包碰撞（packet collision）所造成對整體網路效能的影響程度，並提供理論與實驗數據結果，以供將來藍芽網路佈署時有用之參考依據。

As an emerging personal-area networking solution, Bluetooth technology is highly expected to play an important role in the evolution of wireless communications. Bluetooth was originally viewed as a way of cable replacement, and later found that it was capable of more than this purpose. With low-cost Bluetooth modules available on market, a large deployment of Bluetooths is very likely to take place in the near future. In this thesis, we focus on Bluetooth-related data communication issues and propose a series of solutions or evaluation results to them. The details of related works are described below. One essential issue for almost all kinds of portable devices is power saving. This could be even more important for Bluetooth, which has a design goal of being very compact. Bluetooth has a master-slave configuration, called a piconet. Since the master device is the central controller in a piconet, we do not put such device into sleep mode to save power. On the other hand, slave devices, with less responsibility, can enter power-saving mode to reduce unnecessary battery energy expenditure. In the first work, we study the problem of managing the low-power sniff mode in Bluetooth, where a slave is allowed to be awake only periodically. One challenging problem is how to schedule each slave's sniffing period in a piconet so as to resolve the tradeoff between traffic requirement and power-saving requirement, to which we refer as the sniff-scheduling problem. We propose an adaptive protocol to dynamically adjust each slave's sniff parameters, with a goal of catching the varying, and even asymmetric, traffic patterns among the master and slaves. Unspecified in the Bluetooth standard, the link polling policy adopted by a master may significantly influence the bandwidth utilization of a piconet. In the second work, we propose a Pattern Matching Polling (PMP) policy for data link scheduling to efficiently use the limited bandwidth. A polling pattern is a sequence of Bluetooth packets of different type combinations to be exchanged by a master-slave pair that can properly reflect the traffic ratio (i.e., asymmetry) of the pair. By judiciously selecting a proper polling pattern together with polling times for the link, the precious wireless bandwidth can be better utilized. The ultimate goal is to reduce the unfilled, or even null, payloads in each busy slot. In addition, an overflow mechanism is included to handle unpredictable traffic dynamics. A larger-area Bluetooth network can be formed by interconnecting multiple piconets, called scatternet. However, the structure of scatternets is not defined in the Bluetooth specification and remains as an open issue at the designers' choice. It is desirable to have simple yet efficient scatternet topologies with well supports of routing protocols, considering that Bluetooths are to be used for personal-area networks with design goals of simplicity and compactness. In the literature, although many routing protocols have been proposed for mobile ad hoc networks}, directly applying them poses a problem due to Bluetooth's special baseband and MAC-layer features. In the third work, we propose an attractive scatternet topology called BlueRing which connects piconets as a ring interleaved by bridges between piconets, and address its formation, routing, and topology maintenance protocols. Finally, by sharing the same set of 79/23 frequency channels, a Bluetooth piconet will inevitably encounter the interference problem from other piconets. With a special channel model and packet formats, one research issue is how to predict the packet collision effect in a multi-piconet environment. The analysis model proposed in our fourth work considers all three data packet types (1-/3-/5-slot) supported by Bluetooth, and remove the assumption that each piconet must be fully loaded. Thus our result reflects a more general scenario.