研究與發展專為無線網路系統客製化之最佳化演算架構

Abstract

於現代無線網路傳輸技術高度發展的今日，無論是電信業者或電腦網路中整體應用 系統所涵蓋之部分，包括多媒體編碼，網路傳輸協定，同儕式平台架構，服務品質控制 等方面，皆有許多最佳化問題亟待解決與處理。而最佳化技術的本身，更是現實世界中 非常重要的課題，各種研究、工程領域中相當多的困難與問題，最終都需要仰賴最佳化 技術來進行處理。因此本計畫的目標，是奠基在演化計算 (Evolutionary Computation) 之 富有彈性的最佳化方法論上，發展一套穩定、有效，且可客製化 (Customizable) 並用於 各種數值型態問題 (例如：整數型問題、實數型問題、混合型問題… 等) 的最佳化演算 法。並以之針對無線傳輸控制、網路服務品質機制中各項最佳化問題加以量身訂製，同 時建立數學模型、處理與解決，以期發展可專門用於相關領域最佳化問題之客製最佳化 系統 (Customized Optimization Framework)，以利無線網路技術之發展與進步。在第一 年中，吾人將以原本設計用來處理二進位變數的延伸式精簡基因演算法 (ECGA) 為基 礎，配合本實驗室過去所發展出的多項創新技術，包括整數型邊際機率模型 (iMPM)、 隨選分割 (SoD) 技術…等，設計並實作出可進行客製化之彈性最佳化演算架構。在完 成此最佳化演算架構之設計與實作後，本子計畫在第二年即以無線網路傳輸系統中所存 在之數種最佳化問題為目標，對此最佳化系統進行調整、改變、與客製化。以此為研究 方法與步驟，除了可以處理做為標的之無線網路最佳化問題外，更可達成本計畫中，發 展專為無線網路技術訂製之最佳化演算架構的目標。

As the wireless techniques are highly developed nowadays, lots of optimization problems exist in the domains, including multimedia encoding, network protocols, P2P architectures, QoS control, and the like, covered by the application networking systems established by either the mobile phone companies or the Internet service providers. The optimization technology itself is important in all kinds of real-world applications. Many problems in research and engineering fields are solved or handled by using optimization technologies. Hence, the goal of this research project is firstly to design and develop a stable, effective, as well as customizable optimization framework which can be used to handle optimization problems consisting of different types of decision variables, such as integer variables, real-number variables, mixed typed variables, etc., based on the flexible optimization methodology in evolutionary computation. Moreover, the optimization framework will be customized specifically for the optimization problems existing in wireless transmission and networking quality control. In the first year, we will employ the extended compact genetic algorithm (ECGA), which was designed only for handling binary decision variables, as the backend optimization engine and then integrate several novel techniques, such as integer marginal probabilistic models (iMPM) and split-on-demand (SoD), developed by our lab into ECGA to devise and implement the optimization framework. After accomplishing the customizable optimization framework, in the second year, we will focus on adjusting, modifying, and customizing the optimization framework for the problems needed to be handled in wireless networking systems. By taking these steps, we cannot only solve these immediate optimization problems, but we can also achieve the goal of this research project---developing a customized optimization framework for wireless networking systems.