請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/65756
標題: | 離散正交諧波轉換 Discrete Orthogonal Harmonic Transforms |
作者: | Chun-Lin Liu 劉俊麟 |
指導教授: | 貝蘇章(Soo-Chang Pei) |
關鍵字: | 離散轉換,諧波分析,傅立葉轉換,微分方程式,正交,特徵向量, Discrete Transforms,Harmonic Analysis,Fourier Transforms,Differential Equations,Orthogonality,Eigenvectors, |
出版年 : | 2012 |
學位: | 碩士 |
摘要: | 在數位訊號處理的領域中,離散正交轉換對於信號分析,信號表示,資料壓縮,數位濾波或者是訊號偵測上都是相當重要的工具。然而現有的離散正交轉換,必須要先對轉換的核函數作完整的定義。仍然還有一些連續的正交轉換,並沒有相對應的離散正交轉換。即使我們找到了,離散正交轉換的表達式都相當的複雜,同時也欠缺一般性。
在本篇論文當中,我們提出了一套有系統的方法來計算離散正交轉換。我們首先考慮一維的信號與轉換。先從連續轉換正交轉換的核函數的微分方程式,將其轉換成離散特徵值和特徵向量的問題,最後再用數值方法解此問題。因此,我們探討了一些積分核函數,例如從古典正交多項式多出得正交函數,扁橢圓球波函數以及量子力學中的薛丁格方程式。解出的離散訊號也都用模擬驗證了正確性。同時我們也實作了一些離散正交轉換:例如 chromatic derivatives,Airy 轉換,尺度轉換,一般化分數傅立葉轉換以及線性正則變換。這些離散正交轉換在行為上和常用的類傅立葉轉換有相當大的不同。舉例來說,chromatic derivatives 擁有區域近似的能力,Airy 轉換可以將訊號轉到介於時間和頻率之間的空間上,而尺度轉換對於信號的尺度變化有不變性。我們也探討了一般化的分數傅立葉轉換還有線性正則變換的特徵函數,以及他們的離散實現。這些結果可以直接在離散的角度直接用於訊號的分析。 在高維度的情形下,我們考慮的二維傅立葉轉換的特徵函數,發現不可分離的特徵函數事實上可以寫成 Hermite Gaussian 函數的線性組合。變換組合係數可以得到不同的特徵函數,同時也可以確保正交性。用這樣的概念,我們在離散的角度實作了正交的 gyrator 轉換(和旋轉的 Hermite Gaussian 函數有關)以及正交的Laguerre Gaussian 轉換。gyrator 轉換和 Laguerre Gaussian 轉換都適合用在分析離散的影像信號上。此外,也可以從 Laguerre Gaussian 轉換推導出一個不因旋轉而變的特徵。在三度空間,我們在離散的空間解了球座標簡諧波函數以及球座標簡諧轉換。他們可以用於分析三度空間的訊號,也可以推出不因旋轉而變的特徵。最後我們實做了二維不可分離的線性正則變換。我們提出了一個快速而且準確的演算法,基於分解成許多的步驟,並且套用快速傅立葉演算法。 In digital signal processing, the discrete orthogonal transforms are an important tool for signal analysis, signal representation, data compression, digital filtering, and signal detection. However, the existing discrete orthogonal transforms require the explicit definition of the discrete transform kernels. There are still lots of continuous orthogonal transforms, whose corresponding discrete orthogonal transforms are not found. Even they are found, the expressions are very complicated and not general. In this thesis, a systematic method to computing the discrete orthogonal harmonic transforms is proposed. We first consider the one-dimensional signals and transforms. We can start from the differential equations of the continuous transform kernels, model them as the discrete eigen-problems, and then solve them numerically. As a result, we explore some transform kernels, such as the classical orthogonal functions, derived from the orthogonal polynomials, the prolate spheroidal wave functions, and the Schrodinger equation in quantum mechanics. The discrete equivalents of these functions are all verified by simulations. We also implement some discrete orthogonal harmonic transforms, such as the chromatic derivatives, the Airy transforms, the scale transforms, the generalized fractional Fourier transforms, and the linear canonical transforms. These discrete orthogonal harmonic transforms own different behaviors when they are the conventional Fourier-based transforms. For instance, the chromatic derivatives have the local approximation ability, the Airy transforms convert the signal to the domain which is inbetween the time domain and the frequency domain, the scale transforms are invariant to scale changes of the signals. We also explore the generalized fractional Fourier transforms and the eigenfunctions of the linear canonical transforms, in conjunction with their discrete implementations. These results can be used to analyze the information of the digital signals, directly in the discrete domain. For the higher dimensional case, we consider the eigenfunctions of the two dimensional Fourier transforms and find a simple linear combination of the Hermite Gaussian functions to express the non-separable eigenfunctions. Changing the combination coefficients results in different eigenfunctions and the orthogonality is guaranteed. With this concept, in the two-dimensional case, the gyrator transforms, which are related to the rotated Hermite Gaussian function, and the Laguerre Gaussian transforms are implemented in the discrete domain and with perfect orthogonality. The gyrator transforms and the Laguerre Gaussian transforms are suitable for analyzing the discrete images. In addition, the rotational invariant features can be derived from the Laguerre Gaussian transforms. In the three-dimensional case, we solve the spherical harmonic oscillator wavefunctions and the spherical harmonic oscillator transforms in the discrete domain. They are suitable for analyzing three-dimensional volume data and for deriving the rotational invariant features. Finally, we also implement the two-dimensional non-separable linear canonical transforms. A fast and accurate algorithm is proposed based on decomposing the linear canonical transforms into many stages and utilizing the fast Fourier transform algorithm. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/65756 |
全文授權: | 有償授權 |
顯示於系所單位: | 電信工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-101-1.pdf 目前未授權公開取用 | 15.14 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。