Wavelet Analysis. Willy Hereman. Department of Mathematical and Computer Sciences Colorado School of Mines Golden, CO Sandia Laboratories

Similar documents
A Tutorial on Wavelets and their Applications. Martin J. Mohlenkamp

Wavelets in Scattering Calculations

Lectures notes. Rheology and Fluid Dynamics

Course and Wavelets and Filter Banks

From Fourier to Wavelets in 60 Slides

Wavelets. Lecture 28

Chapter 7 Wavelets and Multiresolution Processing

Lecture 7 Multiresolution Analysis

On the fast algorithm for multiplication of functions in the wavelet bases

MGA Tutorial, September 08, 2004 Construction of Wavelets. Jan-Olov Strömberg

1 Introduction to Wavelet Analysis

1 The Continuous Wavelet Transform The continuous wavelet transform (CWT) Discretisation of the CWT... 2

1 Continuity Classes C m (Ω)

Quantum Computing Lecture 2. Review of Linear Algebra

Space-Frequency Atoms

LECTURE 5: THE METHOD OF STATIONARY PHASE

( nonlinear constraints)

Chapter III Beyond L 2 : Fourier transform of distributions

We have to prove now that (3.38) defines an orthonormal wavelet. It belongs to W 0 by Lemma and (3.55) with j = 1. We can write any f W 1 as

Lecture Notes 5: Multiresolution Analysis

Boundary functions for wavelets and their properties

Chapter 7 Wavelets and Multiresolution Processing. Subband coding Quadrature mirror filtering Pyramid image processing

Wavelets and Filter Banks

Two Channel Subband Coding

Space-Frequency Atoms

Polynomial Approximation: The Fourier System

Application of wavelets to singular integral scattering equations

Lecture 16: Multiresolution Image Analysis

Wavelets, Multiresolution Analysis and Fast Numerical Algorithms. G. Beylkin 1

12. Cholesky factorization

Fourier analysis, measures, and distributions. Alan Haynes

TD 1: Hilbert Spaces and Applications

An Adaptive Pseudo-Wavelet Approach for Solving Nonlinear Partial Differential Equations

Toward a Realization of Marr s Theory of Primal Sketches via Autocorrelation Wavelets: Image Representation using Multiscale Edge Information

Construction of Multivariate Compactly Supported Orthonormal Wavelets

Wavelets and Image Compression Augusta State University April, 27, Joe Lakey. Department of Mathematical Sciences. New Mexico State University

DAVID FERRONE. s k s k 2j = δ 0j. s k = 1

Kernel Method: Data Analysis with Positive Definite Kernels

Hilbert Space Problems

Scientific Computing: An Introductory Survey

Homework I, Solutions

Matrices: 2.1 Operations with Matrices

October 7, :8 WSPC/WS-IJWMIP paper. Polynomial functions are renable

Nonseparable multivariate wavelets. Ghan Shyam Bhatt. A dissertation submitted to the graduate faculty

Digital Image Processing

Math 671: Tensor Train decomposition methods

Denoising and Compression Using Wavelets

Linear Algebra (Review) Volker Tresp 2018

Lecture 15 Review of Matrix Theory III. Dr. Radhakant Padhi Asst. Professor Dept. of Aerospace Engineering Indian Institute of Science - Bangalore

Schemes. Philipp Keding AT15, San Antonio, TX. Quarklet Frames in Adaptive Numerical. Schemes. Philipp Keding. Philipps-University Marburg

Harmonic Analysis: from Fourier to Haar. María Cristina Pereyra Lesley A. Ward

1. Fourier Transform (Continuous time) A finite energy signal is a signal f(t) for which. f(t) 2 dt < Scalar product: f(t)g(t)dt

Development and Applications of Wavelets in Signal Processing

Adaptive Nonparametric Density Estimators

Matrices A brief introduction

Numerical Approximation Methods for Non-Uniform Fourier Data

Chapter 1: Systems of Linear Equations and Matrices

3 (Maths) Linear Algebra

8.1 Concentration inequality for Gaussian random matrix (cont d)

University of Illinois at Chicago Department of Physics. Quantum Mechanics Qualifying Examination. January 7, 2013 (Tuesday) 9:00 am - 12:00 noon

MATH 590: Meshfree Methods

Wavelets For Computer Graphics

Short-time Fourier transform for quaternionic signals

Cuntz algebras, generalized Walsh bases and applications

AN INTRODUCTION TO COMPRESSIVE SENSING

Digital Image Processing

Lecture 6. Numerical methods. Approximation of functions

Math 304 (Spring 2010) - Lecture 2

A matrix over a field F is a rectangular array of elements from F. The symbol

ARCS IN FINITE PROJECTIVE SPACES. Basic objects and definitions

Georgia Tech PHYS 6124 Mathematical Methods of Physics I

A Novel Fast Computing Method for Framelet Coefficients

Wavelets and applications

Transformations from R m to R n.

MATH 205C: STATIONARY PHASE LEMMA

A trigonometric orthogonality with respect to a nonnegative Borel measure

1. Let m 1 and n 1 be two natural numbers such that m > n. Which of the following is/are true?

Subband Coding and Wavelets. National Chiao Tung University Chun-Jen Tsai 12/04/2014

Graph Signal Processing

The Closed Form Reproducing Polynomial Particle Shape Functions for Meshfree Particle Methods

Sparse linear models

Wavelets and multiresolution representations. Time meets frequency

MLISP: Machine Learning in Signal Processing Spring Lecture 10 May 11

EXPOSITORY NOTES ON DISTRIBUTION THEORY, FALL 2018

Compactly Supported Wavelets Based on Almost Interpolating and Nearly Linear Phase Filters (Coiflets)

be the set of complex valued 2π-periodic functions f on R such that

POINT VALUES AND NORMALIZATION OF TWO-DIRECTION MULTIWAVELETS AND THEIR DERIVATIVES

Construction of Multivariate Compactly Supported Tight Wavelet Frames

An Introduction to Wavelets and some Applications

Conformal maps. Lent 2019 COMPLEX METHODS G. Taylor. A star means optional and not necessarily harder.

Assignment #10: Diagonalization of Symmetric Matrices, Quadratic Forms, Optimization, Singular Value Decomposition. Name:

Basic Postulates of Quantum Mechanics

Orthogonality of hat functions in Sobolev spaces

Linear Algebra Practice Problems

Digital Image Processing

Generalized pointwise Hölder spaces

ABSTRACT. Title: Wavelet-space solution of the Poisson equation: An algorithm for use in particle-incell simulations

Module 4 MULTI- RESOLUTION ANALYSIS. Version 2 ECE IIT, Kharagpur

20. Gaussian Measures

Free Probability Theory and Non-crossing Partitions. Roland Speicher Queen s University Kingston, Canada

Transcription:

Wavelet Analysis Willy Hereman Department of Mathematical and Computer Sciences Colorado School of Mines Golden, CO 8040-887 Sandia Laboratories December 0, 998

Coordinate-Coordinate Formulations CC and CC2 Integral formulation of the Dirichlet problem (ψ T (x s ) = 0) leads to F i (x) = L 2 L 2 Z D p (x, x )N T (x )dx N T is the normal derivative of the field on the surface (surface current) F i a linear combination of the incident field ψ i (x s ) and the normal derivative N i evaluated on the surface F i (x) = βψ i (x s ) + αn i (x s ). Note that F i comes from a linear combination of two equations: ψ i (x s ) = (SN T )(x s ) and 2 N T (x s ) = N i (x s ) PV n (SN T )(x s ). The impedance kernel (D=Dirichlet) is Z D p (x, x ) = 2 αδ(x x ) + α PV G p (x, x ) + βg p (x, x ) involves a Dirac delta function term, the periodic Green s function G p (x, x ) = i λ 4π L and its normal derivative j= β j e ik [α j (x x )+β j s(x) s(x ) ] G p (x, x ) = [n j j G p (x, x )] z=s(x) z =s(x ) n j is the (non-unit) surface normal PV represents Cauchy Principal Value

Discretization of F i (x) = L 2 L 2 Z D p (x, x )N T (x )dx yields a matrix whose row and column entries both result from sampling in coordinate space (coordinate-coordinate method, CC for short). The real parameters α and β in specify the type of equation: For α = 0 first kind integral equation referred to as CC For β = 0 second kind integral referred to as CC2 For β = and α arbitrary ZN T = F i. F i (x) = βψ i (x s ) + αn i (x s ) combined field integral equation (CFIE) Solution of the unknown N T yields the amplitudes A n. The amplitudes lead to the energy (normalized): n A n 2 Re(β n ) β 0. We will always test how close this is to.

Daubechies Orthogonal Wavelets with Compact Support Example: DAUB4 (2 vanishing moments) Scaling and wavelet function families: φ j,k (x) = 2 j 2φ(2 j x k) ψ j,k (x) = 2 j 2ψ(2 j x k). The two-scale difference equations for φ(x) and ψ(x) : φ(x) = 2 3 ψ(x) = 2 3 h k φ(2x k) g k φ(2x k) The functions φ(x) and ψ(x) are supported on interval [0, 3]. The scaling function φ(x) is orthogonal to all its integer translates. The scaling function φ(x) is properly normalized φ(x)dx =. The wavelet function ψ(x) has two vanishing moments ψ(x)dx = xψ(x)dx = 0. In the Fourier space, the low pass filter with M = 2 vanishing moments H(z) = 3 h k z k = 2( + z 2 )2 Q(z), with z = exp ( 2πiω) with Q(z) is polynomial of degree.

H(z) satisfies the quadrature mirror filter (QMF) condition H(z)H( z ) + H( z)h( z ) = 2. The polynomial Q(z) for M = 2 is computed from factorization of Q(z)Q( z ) = M ( ) k (M + k)! ( z) 2k (4z) k = 2 k!(m )! 2z z 2. The averaging (low pass) filter H = {h 0, h, h 2, h 3 }, where h 0 = 4 2 ( + 3) = 0.48296293445343 h = 4 2 (3 + 3) = 0.83656303737808 h 2 = 4 2 (3 3) = 0.22443868042034 h 3 = 4 2 ( 3) = 0.29409522552604 The differencing (high pass) filter G = {g 0, g, g 2, g 3 } = {h 3, h 2, h, h 0 }. Both H and G have L = 4 filter taps. The high pass filter annihilates constant and linear trends: g 0 + g + g 2 + g 3 = 0 and 0g 0 + g + 2g 2 + 3g 3 = 0.

Example: DAUB6 (3 vanishing moments) Scaling and wavelet function families: φ j,k (x) = 2 j 2φ(2 j x k) ψ j,k (x) = 2 j 2ψ(2 j x k). The two-scale difference equations for φ(x) and ψ(x) : φ(x) = 2 5 ψ(x) = 2 5 h k φ(2x k) g k φ(2x k) The functions φ(x) and ψ(x) are supported on interval [0, 5]. The scaling function φ(x) is orthogonal to all its integer translates. The scaling function φ(x) is properly normalized φ(x)dx =. The wavelet function ψ(x) has three vanishing moments ψ(x)dx = xψ(x)dx = x2 ψ(x)dx = 0. In the Fourier space, the low pass filter with M = 3 vanishing moments H(z) = 5 h k z k = 2( + z 2 )3 Q(z), with z = exp ( 2πiω). Q(z) for M = 3 is polynomial of degree 2 and computed via Q(z)Q( z ) = M ( ) k (M + k)! ( z) 2k (4z) k k!(m )! = 9 4 + 3z2 8 + 3 8z 2 9z 4 9 4z.

The averaging (low pass) filter H = {h 0, h, h 2, h 3, h 4, h 5 }, where h 0 = h = h 2 = h 3 = h 4 = h 5 = 6 2 ( + 0 + 5 + 2 0) = 0.3326705529500826 6 2 (5 + 0 + 3 5 + 2 0) = 0.806895093093 6 2 (0 2 0 + 2 5 + 2 0) = 0.4598775028495 6 2 (0 2 0 2 5 + 2 0) = 0.350020002546 6 2 (5 + 0 3 5 + 2 0) = 0.085442738820267 6 2 ( + 0 5 + 2 0) = 0.0352262988570955 The differencing (high pass) filter G = {g 0, g, g 2, g 3, g 4, g 5 } = {h 5, h 4, h 3, h 2, h, h 0 }. Both H and G have L = 6 filter taps. The high pass filter annihilates constant, linear and quadratic trends: and g 0 + g + g 2 + g 3 + g 4 + g 5 = 0 0g 0 + g + 2g 2 + 3g 3 + 4g 4 + 5g 5 = 0 0g 0 + g + 4g 2 + 9g 3 + 6g 4 + 5g 5 = 0

The Wavelet Transform For CC, SC or SS formulations: we have to solve a linear system Ax = b x is the vector of unknowns, A and b are given. Apply wavelets to: the integral equations, the matrix equations. Wavelet transform methods are applied to: sparsify the matrix, make the matrix inversion faster, speed up the solution process for x. Different types of wavelets can be used, e.g. Daubechies wavelets. The wavelet transform is described as a similarity transform (implemented as convolution). If W is the orthogonal matrix which performs the wavelet transform on the columns of a matrix, then the wavelet transform à of A is à = WAW T since W = W T (similarity transform). Applying this transformation to Ax = b we get or where x = Wx and b = Wb. WAW T Wx = Wb à x = b,

Pick a threshold τ and replace à by Ã(t) as follows à (t) ij = Solve the new system for x. 0, Ãij τ max Ãmn, m,n à ij otherwise. à (t) x = b Then compute x = W T x, which is the thresholded solution Ax = b. Observations: Cost: the times necessary to apply the wavelet transform and its inverse, and to do a matrix search to threshold can be many times the inversion time of the untransformed impedance matrix. This is particularly true for the SC and SS methods whose small matrix sizes make the inversion much faster. We can not reach a sparsity of 0 % (only 0 % of the matrix elements remain after thresholding). A 0 % level of sparsity our matrices become singular unless we dramatically oversampled to begin with. The accuracy decreased dramatically as sparsity increased. Sparsification may help iterative techniques. We have control over sampling. Proponents of wavelets go immediately to large matrices which are often the result of oversampling. We achieved better results with small dense matrices and no wavelet techniques.