Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/48674
標題: Study in Channel Coding with Side Information and Its Applications to the Digital Watermarking Problems.
具有額外資訊之通道碼研究與其在數位浮水印之應用
作者: 陳後守
關鍵字: CCSI
應用研究
電信工程, 資訊科學--軟體
SCSI
dirty paper coding
distributed source coding
turbo codes
LDPC codes
額外資訊之通道碼
額外資訊之訊源碼
分散式訊源碼
髒紙碼
渦輪碼
低密度檢查碼
摘要: The proposal of this project investigates the problems of channel coding with side information (CCSI), along with its applications to informed coding problems in digital watermarking, also known as dirty paper codes. The mathematical models and solutions of CCSI is very closed to its dual problem, the source coding with side information (SCSI), which also has wide applications, such as distributed source coding in wireless sensor networks. In the past two decades, we have witnessed dramatic advances in pure channel coding problems, including the invention of turbo codes, and the rediscovery of LDPC codes. Both codes approach the Shannon capacity using the iterative sum product algorithm. The information-theoretic bounds for CCSI and SCSI are already solved in certain sources, such as binary and Gaussian sources, and certain channels, such as BSC and AWGN. However, the current gap between these theoretical bounds and practical approaches remains substantial. In other words, the study of structured schemes for CCSI or SCSI, such as the use of turbo or LDPC codes, becomes an important issue. The research proposal is a two-year project: we will study the general problems in CCSI in the first year and then apply these results to digital watermarking problems in the second year.In the first year, we will provide a comprehensive study on the general problems of CCSI, including the binary and Gaussian sources transmitted over the noiseless channels, or noisy BSC or AWGN channels. Theoretically, we can use a nave random binning to achieve the CCSI capacity. However, because of its infeasible complexity, the study of structured schemes, such as the nested binary codes or nested lattices, will be our main concern in the first year. In principle, a good CCSI code is a good channel code which is partitioned into some certain cosets of good source codes. This is a challenging joint source-channel problem, where we start a code with a large packing radius, which contains a subcode with a small covering radius. In the second year, we will apply these results to the problems of digital watermarking. We propose two informed coding algorithms, based on bit flipping and Hamming trellis respectively, for the reduction of encoding and decoding complexity. Some of the preliminary results of these two algorithms will be addressed in the project.
本計劃將探討額外資訊之通道碼所遭遇之問題與其應用於數位浮水印之技巧,此技術又稱為髒紙碼。由於額外資訊之通道碼的數學模型與解決方法跟額外資訊之訊源碼有異曲同工之處,所以額外資訊訊源碼之原理與技術也可以應用於額外資訊通道碼。而額外資訊訊源碼主要應用於無線感測網路之分散式訊源碼。過去20年傳統通道碼最大之成就為引進渦輪碼與低密度檢查碼,此兩種通道碼主要是使用遞迴式和積演算法逼近薛農之極限。如同傳統之通道編碼與訊源編碼,在二元或高斯訊號源通過二元對稱通道或高斯雜訊通道下,額外資訊通道編碼之理論上限與額外資訊訊源編碼的理論下限均已經有數學公式。但是如何使用實際之編碼技術達到理論之極限仍是一件非常挑戰之工作。如同渦輪碼與低密度檢查碼之於通道碼,如何設計有結構且達到理論極限之額外資訊通道碼將是我們此計劃之重大議題。我們提出兩年之研究計畫,第一年主要是研究額外資訊通道碼之一般問題,而第二年將利用第一年所得到結果應用於數位浮水印。第一年我們將廣泛研究額外資訊通道碼的理論基礎,主要包含二元或高斯訊號源通過二元對稱通道或高斯雜訊通道下之數學模式與其編解碼架構。理論上我們可以使用自然隨意分群法(random binning)逼近理論值,但此方法複雜度過大,因此研究有架構之編碼方式將是我們第一年之主要重心。比如二元訊號源通過二元對稱通道,可以使用套狀碼(nest codes),或在高斯訊號源通過高斯雜訊通道,可以使用套狀格(nest lattices)。設計一個好的額外資訊通道碼的基本原則是將一個好的通道碼分割成數個好的訊源碼,亦即一個好的額外資訊通道碼本身須有很大之合併半徑(packing radius)且包含有很小覆蓋半徑(covering radius)之子碼。第二年我們將應用額外資訊通道碼之結果於數位浮水印,我們提出位元翻轉解碼與漢明籬笆解碼來降低額外資訊通道碼之編碼與解碼複雜度,初步之結果將於計畫中簡述。
URI: http://hdl.handle.net/11455/48674
其他識別: NSC99-2221-E005-081-MY2
文章連結: http://grbsearch.stpi.narl.org.tw/GRB/result.jsp?id=2122446&plan_no=NSC99-2221-E005-081-MY2&plan_year=99&projkey=PB9907-2448&target=plan&highStr=*&check=0&pnchDesc=%E5%85%B7%E6%9C%89%E9%A1%8D%E5%A4%96%E8%B3%87%E8%A8%8A%E4%B9%8B%E9%80%9A%E9%81%93%E7%A2%BC%E7%A0%94%E7%A9%B6%E8%88%87%E5%85%B6%E5%9C%A8%E6%95%B8%E4%BD%8D%E6%B5%AE%E6%B0%B4%E5%8D%B0%E4%B9%8B%E6%87%89%E7%94%A8
Appears in Collections:通訊工程研究所

文件中的檔案:

取得全文請前往華藝線上圖書館



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.