(a) Explain why the wavelet expansion (9.136) defines a...

(a) Explain why the wavelet expansion (9.136) defines a linear transformation on $\mathbb{R}^n$ that takes a wavelet coefficient vector $\mathbf{c}=\left(c_0, c_1, \ldots, c_{n-1}\right)^T$ to the corresponding sample vector $\mathbf{f}=\left(f_0, f_1, \ldots, f_{n-1}\right)^T$. (b) According to Theorem 7.5, the wavelet map must be given by matrix multiplication $\mathbf{f}=W_n$ c by a $2 \times 2^n$ matrix $W=W_n$. Construct $W_2, W_3$ and $W_4$. (c) Prove that the columns of $W_n$ are obtained as the values of the wavelet basis functions on the $2^n$ sample intervals. (d) Prove that the columns of $W_n$ are orthogonal. (e) Is $W_n$ an orthogonal matrix? Find a formula for $W_n^{-1}$. (f) Explain why the wavelet transform is given by the linear map, $\mathrm{c}=W_n^{-1} \mathbf{f}$.

(a) Explain why the wavelet expansion (9.136) defines a linear transformation on $\mathbb{R}^n$ that takes a wavelet coefficient vector $\mathbf{c}=\left(c_0, c_1, \ldots, c_{n-1}\right)^T$ to the corresponding sample vector $\mathbf{f}=\left(f_0, f_1, \ldots, f_{n-1}\right)^T$.
(b) According to Theorem 7.5, the wavelet map must be given by matrix multiplication $\mathbf{f}=W_n$ c by a $2 \times 2^n$ matrix $W=W_n$. Construct $W_2, W_3$ and $W_4$. (c) Prove that the columns of $W_n$ are obtained as the values of the wavelet basis functions on the $2^n$ sample intervals. (d) Prove that the columns of $W_n$ are orthogonal. (e) Is $W_n$ an orthogonal matrix? Find a formula for $W_n^{-1}$. (f) Explain why the wavelet transform is given by the linear map, $\mathrm{c}=W_n^{-1} \mathbf{f}$.

Key Concepts

Matrix Inversion

Matrix inversion is the process of finding a matrix that reverses the effect of a given linear transformation. For orthogonal or orthonormal matrices like those often found in wavelet transformations, the inverse is directly related to the transpose of the matrix (up to a scaling factor). This allows the original signal to be recovered unambiguously from its wavelet coefficients through the inverse transformation.

Orthogonality

Orthogonality in the context of wavelet transforms refers to the property that the inner product between any two distinct basis functions is zero. This property implies that the columns of the transformation matrix are orthogonal, which simplifies computations, ensures energy preservation, and facilitates the inversion of the transform.

Discrete Wavelet Transform (DWT)

The discrete wavelet transform involves the transformation of a digital signal into wavelet coefficients. This process discretizes the continuous wavelet transform and provides a finite-dimensional representation, typically organized in a structured matrix, that facilitates both the analysis and reconstruction of signals at different resolutions.

Basis Functions Evaluation

The columns of the transformation matrix are obtained by evaluating the wavelet basis functions on discrete sampling intervals of the signal. This evaluation links the abstract wavelet functions to concrete numerical values, ensuring that each column encodes the influence of a single basis function over the samples.

Linear Transformation

A linear transformation is a mapping between vector spaces that respects the operations of vector addition and scalar multiplication. When a wavelet expansion is applied to a signal, it acts as a linear transformation, translating a vector of wavelet coefficients into a vector of signal samples by summing weighted basis functions.

Wavelet Expansion

Wavelet expansion is the process of expressing a function or signal as a linear combination of wavelet basis functions, which are localized in both time (or space) and frequency. This expansion allows for the analysis of data at various scales and is fundamental to techniques for signal compression, noise reduction, and feature extraction.

Matrix Representation of Linear Transformations

Any linear transformation between finite-dimensional vector spaces can be represented by a matrix. In the context of wavelet transforms, the mapping from wavelet coefficients to signal samples is represented by a matrix whose columns are the sampled values of the corresponding wavelet basis functions. This representation enables efficient computation and analysis using standard linear algebra techniques.

Transcript

00:02 Okay, this question asks us about the matrix a from last time, but this time it wants us to normalize it.

00:10 So here i am in python again, and we have our matrix a, and in order to normalize it, i just took each column and divided it by its magnitude.

00:25 And just to see, we can print this matrix u, and as we can see, it's our original matrix, except this time.

00:37 Everything is scaled so that we have normal columns.

00:44 So now what it wants us to do is to compute u transpose u and u -u -transpose.

00:53 So first we're going to print the matrix multiplication of u -transpose and u -transpose.

01:07 And remember, this should give us the identity.

01:11 So let's check.

01:13 And we have some rounding going on with the computer.

01:16 Code, but we can see we have 1, 1, 1, 1 on the diagonal, and then 0 everywhere else, because these are just rounding errors by the computer.

01:30 So we know that u transpose u gives us the identity matrix, which makes sense because we have orthonormal columns now.

01:39 But then it wants us to see what u times u transpose gives us.

01:47 And this actually gives us something completely different.

01:56 And the most obvious difference is in the matrix size.

02:00 So first, we had a 4 by 8 multiplied by an 8 by 4, which gave us a 4 by 4.

02:11 But now we have an 8 by 4 times a 4 by 8, giving us an 8 by 8.

02:20 Next, it tells us to generate a random vector in r8.

02:33 And i didn't think of this ahead of time.

02:39 So we're just going to generate a vector.

02:41 Let's say 1, 4, negative 8, negative 5, 10, 2, 3, 3, 3, 1.

02:56 Let's count 1, 2, 3, 4, 6, 7, 8.

03:00 Up, and we have an extra...

Please give Ace some feedback