Chapter 2

Signal and System as Vectors

To solve forward and inverse problems, we need mathematical tools to formulate them. Since it is convenient to use vectors to represent multiple system parameters, variables, inputs (or excitations) and outputs (or measurements), we first study vector spaces. Then, we introduce vector calculus to express interrelations among them based on the underlying physical principles. To solve a nonlinear inverse problem, we often linearize an associated forward problem to utilize the numerous mathematical tools of the linear system. After introducing such an approximation method, we review mathematical techniques to deal with a linear system of equations and linear transformations.

2.1 Vector Spaces

We denote a signal, variable or system parameter as f(r, t), which is a function of position r = (x, y, z) and time t. To deal with a number n of signals, we adopt the vector notation (f1(r, t), …, fn(r, t)). We may also set vectors as (f(r1, t), …, f(rn, t)), (f(r, t1), …, f(r, tn)) and so on. We consider a set of all possible such vectors as a subset of a vector space. In the vector space framework, we can add and subtract vectors and multiply vectors by numbers. Establishing the concept of a subspace, we can project a vector into a subspace to extract core information or to eliminate unnecessary information. To analyze a vector, we may decompose it as a linear combination of basic elements, which we handle as a basis or coordinate of a subspace.

2.1.1 Vector Space and Subspace


Definition 2.1.1
A non-empty set V is a vector space over a field images or images if there are operations of vector addition and scalar multiplication with the following properties.

Vector addition

1. u + vV for every u, vV (closure).
2. u + v = v + u for every u, vV (commutative law).
3. (u + v) + w = u + (v + w) for every u, v, wV (associative law).
4. There exist 0V such that u + 0 = u for every uV (additive identity).
5. For all uV there existsuV such that u + ( − u) = 0 andu is unique (additive inverse).

Scalar multiplication

1. For images and uV, auV (closure).
2. For images and u, vV, a(bu) = (ab)u (associative law).
3. For images and u, vV, a(u + v) = au + av (first distributive law).
4. For images and uV, (a + b)u = au + bu (second distributive law).
5. 1u = u for every uV (multiplicative identity).

A subset W of a vector space V over F is a subspace of V if and only if au + vW for all images and for all u, vW. The subspace W itself is a vector space.


Example 2.1.2
The following are examples of vector spaces.
  • images, n-dimensional Euclidean space.
  • images.
  • C([a, b]), the set of all complex-valued functions that are continuous on the interval [a, b].
  • C1([a, b]): = {f ∈ C([a, b]):f′ ∈ C[a, b]}, the set of all functions in C([a, b]) with continuous derivative on the interval [a, b].
  • images, the set of all square-integrable functions on the open interval (a, b).
  • images.

 


Definition 2.1.3
Let G = {u1, …, un} be a subset of a vector space V over a field images or images. The set of all linear combinations of elements of G is denoted by span G:

images


The span G is the smallest subspace of V containing G. For example, if images, then span{(1, 2, 3), (1, − 2, 0)} is the plane images.


Definition 2.1.4
The elements u1, …, un of a vector space V are said to be linearly independent if

images

Otherwise, u1, …, un are linearly dependent.


If images is linearly independent, no vector uj can be expressed as a linear combination of other vectors in the set. If u1 can be expressed as u1 = a2u2 + ··· + anun, then a1u1 + a2u2 + anun = 0 with a1 = − 1, so they are not linearly independent. For example, {(4, 1, 5), (2, 1, 3), (1, 0, 1)} is linearly dependent since − (4, 1, 5) + (2, 1, 3) + 2(1, 0, 1) = (0, 0, 0). The elements (2, 1, 3) and (1, 0, 1) are linearly independent because a1(2, 1, 3) + a2(1, 0, 1) = (0, 0, 0) implies a1 = a2 = 0.


Example 2.1.5
In Figure 2.1, the two vectors {u, v} are linearly independent whereas the four vectors {u, v, p, q} are linearly dependent. Note that p and q are linearly independent.

Figure 2.1 Linearly independent and dependent vectors

2.1

2.1.2 Basis, Norm and Inner Product


Definition 2.1.6
Let W be a subspace of a vector space V. If span{u1, …, un} = W and {u1, …, un} is linearly independent, then {u1, …, un} is said to be a basis for W.

If images is a basis for W, then any vector vW can be expressed uniquely as images. If G′ is another basis of W, then G′ contains exactly the same number n of elements.


Definition 2.1.7
Let W be a subspace of a vector space V. Then W is n-dimensional if the number of elements of the basis of W is n; W is finite-dimensional if dim W < ∞; otherwise W is infinite-dimensional.

To quantify a measure of similarity or dissimilarity among vectors, we need to define the magnitude of a vector and the distance between vectors. We use the norm ||u|| of a vector u to define such a magnitude. In the area of topology, the metric is also used for defining a distance. To distinguish different vectors in a vector space, we define a measure of distance or metric between two vectors u and v as the norm ||uv||. The norm must satisfy the following three rules.


Definition 2.1.8
A normed vector space V is a vector space equipped with a norm || · || that satisfies the following:
1. images.
2. images.
3. images (triangle inequality).

Here, the notation images stands for “for all” and iff stands for “if and only if”.


Example 2.1.9
Consider the vector space images. For images and 1 ≤ p < ∞, the p-norm of u is

2.1 2.26

In particular, images is the standard distance between u and v. We should note that, when 0 < p < 1, ||u||p is not a norm because it does not satisfy the triangle inequality.

 


Example 2.1.10
Consider the vector space V = C([0, 1]). For f, g ∈ V, the distance between f and g can be defined by

images


In addition to the distance between vectors, it is desirable to establish the concept of an angle between them. This requires the definition of an inner product.


Definition 2.1.11
Let V be a vector space over images or images. We denote the complex conjugate of images by images. A vector space V with a function images is an inner product space if:
1. images;
2. images;
3. images and images.

In general, 〈u, v 〉 is a complex number, but 〈u, u 〉 is real. Note that 〈w, au + bv 〉 = imagesw, u 〉 + imagesw, v 〉 and 〈u, 0 〉 = 0. If 〈u, v 〉 = 0 for all v ∈ V, then u = 0. Given any inner product, images is a norm on V.

For a real inner product space V, the inner product provides angle information between two vectors u and v. We denote the angle θ between u and v as

images

We interpret the angle as follows.

1. If θ = 0, then 〈u, v 〉 = ||u|| ||v|| and v = au for some a > 0. The two vectors u, v are in the same direction.
2. If θ = π, then 〈u, v 〉 = − ||u|| ||v|| and v = au for some a < 0. The two vectors u, v are in opposite directions.
3. If θ = ± π/2, then 〈u, v 〉 = 0. The two vectors u, v are orthogonal.

Definition 2.1.12
The set {u1, …, un} in an inner product space V is said to be an orthonormal set ifuj, uk 〉 = 0 for j ≠ k and ||uj|| = 1. A basis is an orthonormal basis if it is an orthonormal set.

2.1.3 Hilbert Space

When we analyze a vector f in a vector space V having a basis images, we wish to represent f as

images

Computation of the coefficients aj could be very laborious when the vector space V is not equipped with an inner product and images is not an orthonomal set. A Hilbert space is a closed vector space equipped with an inner product.


Definition 2.1.13
A vector space H over images or images is a Hilbert space if:
1. H is an inner product space;
2. H = images (H is a closed vector space), that is, whenever limn→∞||unu|| = 0 for some sequence {un} ⊂ H, u belongs to H.

For u, v, wH, a Hilbert space, we have the following properties.

  • Cauchy–Schwarz inequality: |〈u, v 〉 | ≤ ||u|| ||v||.
  • Triangle inequality: ||u + v|| ≤ ||u|| + ||v||.
  • Parallelogram law: ||u + v||2 + ||uv||2 = 2||u||2 + 2||v||2.
  • Polarization identity: 4〈u, v 〉 = ||u + v||2 − ||uv||2 + i||u + iv||2 − i||u − iv||2.
  • Pythagorean theorem: ||u + v||2 = ||u||2 + ||v||2 if 〈u, v 〉 = 0.

Exercise 2.1.14 (Gram–Schmidt process)
Let H be a Hilbert space with a basis {v1, v2, …}. Assume that {u1, u2, …} is obtained from the following procedure depicted in Figure 2.2:
1. Set w1 = v1 and u1 = w1/||w1||;
2. Set w2 = v2 − 〈v2, u1u1 and u2 = w2/||w2||;
3. For n = 2, 3, …,

images

Figure 2.2 Illustration of the Gram–Schmidt process

2.2
Prove that {u1, u2, …} is an orthonormal basis of H, that is,

images


 


Theorem 2.1.15 (Projection theorem)
Let G be a closed convex subset of a Hilbert space H. For every uH, there exists a unique u*G such that ||uu*|| ≤ ||uv|| for all vG.

For the proof of the above theorem, see Rudin (1970). Let S be a closed subspace of a Hilbert space H. We define the orthogonal complement S of S as

images

According to the projection theorem, we can define a projection map PS:H → S such that the value PS(u) satisfies

images

This means that f(t) = ||u − PS(u) + tv||2 has its minimum at t = 0 for any v ∈ S and, therefore,

images

or

images

Hence, the projection theorem states that every u ∈ H can be uniquely decomposed as u = v + w with v ∈ S and w ∈ S and we can express the Hilbert space H as

images

From the Pythagorean theorem,

images

Figure 2.3 illustrate the projection of a vector u onto a subspace S.

Figure 2.3 Illustration of a projection of a vector u onto a subspace S

2.3

Example 2.1.16 (Euclidean space images is a Hilbert space)
A Hilbert space is a generalization of the Euclidean space images. For images, we define the inner product and norm as

images

The distance between x and y is defined by ||xy||, and ||xy|| = 0 implies x = y. Ifx, y 〉 = 0, then x and y are orthogonal and the following Pythagorean theorem holds:

images

If {e1, e2, …, en} is an orthonormal basis of images, every images can be represented uniquely by

images

For example, we let e1 = (1, 0, …, 0), e2 = (0, 1, 0, …, 0), …. If Vm = span{e1, …, em} for m < n, the vector images is

images

with the distance images We can generalize this dot product property in Euclidean space images to an infinite-dimensional Hilbert space.

 


Example 2.1.17 (L2-space)
Let I be the interval [0, 1]. We denote the set of all square-integrable complex functions by L2(I), that is,

images

For f, g ∈ L2(I), we define the inner product as

images

where images denotes the complex conjugate of g(x). Dividing the interval [0, 1] into N subintervals with endpoints x0 = 0, x1 = Δx, …, xN = NΔx = 1 and equal gap Δx = 1/N, the inner product 〈f, g 〉 in L2(I) can be viewed approximately as the inner product in Euclidean space images:

images

The vector space L2(I) with the above inner product is a Hilbert space and retains features of Euclidean space. Indeed, in order for the vector space L2(I) to be a Hilbert space, we need Lebesgue measure theory because there are infinitely many f with ||f|| = 0 but f ≠ 0 in the pointwise sense. In Lebesgue measure theory, ||f|| = 0 implies f = 0 almost everywhere in the sense of the measure. In L2-space, f = g means that f = g almost everywhere. For the details of Lebesgue measure theory, please refer to Rudin (1970).

 


Exercise 2.1.18
Prove that

images

is an orthonormal set in L2([0, 1]).

2.2 Vector Calculus

Based on the underlying physical principles, we need to express interrelations among system parameters, variables, excitations and measurements. A scalar-valued function f(x) defined in images represents a numerical quantity at a point images. Examples may include temperature, voltage, pressure, altitude and so on. A vector-valued function F(x) is a vector quantity at images. It represents a vector field such as displacement, velocity, force, electric field intensity, magnetic flux density and so on. We now review the vector calculus of gradient, divergence and curl to handle basic dynamics among variables and parameters.

2.2.1 Gradient

Let images. For a given unit vector images, the directional derivative of f at x in the direction b is denoted by ∂bf(x). It represents the rate of increase in f at x in the direction of b:

images

If b = ej, the jth unit vector in the Cartesian coordinate system, we simply write images. The gradient of f, denoted as ∇f, is a vector-valued function that points in the direction of maximum increase of f:

images


Remark 2.2.1
Let images. Suppose that f(x0) = λ and ∇f(x0) ≠ 0. The vector ∇f(x0) is perpendicular to the level set images because there is no increase in f along the level set imagesλ. If images is a domain enclosed by the level surface imagesλ, the unit outward normal vector n(x0) at x0 on the boundary ∂Ωλ is

images

which points in the steepest ascending direction. The curvature along the level set imagesλ is given by

images


 


Proof.
If images is a smooth curve lying on a level surface

images

then

images

The vector images is perpendicular to the tangent direction images of the level set imagesλ. Since images has the same direction as the gradient ∇f(x), we can write

images


 


Exercise 2.2.2
Prove that ∇f = (∂1f, …, ∂nf).

2.2.2 Divergence

The divergence of F(r) at a point r, written as div F, is the net outward flux of F per unit volume of a ball centered at r as the ball shrinks to zero:

images

where dS is the surface element, Br(r) is the ball with radius r and center r, and ∂B is the boundary of B, which is a sphere.


Theorem 2.2.3 (Divergence theorem)
Let Ω be a bounded smooth domain in images. The volume integral of the divergence of a C1-vector field F = (F1, F2, F3) equals the total outward flux of the vector F through the boundary of Ω:

images


 


Exercise 2.2.4
Prove div F = ∂1F1 + ∂2F2 + ∂3F3 and the divergence theorem.

2.2.3 Curl

The circulation of a vector field F = (F1, F2, F3) around a closed path C in images is defined as a scalar line integral of the vector F over the path C:

images

If F represents an electric field intensity, the circulation will be an electromotive force around the path C.

The curl of a vector field F, denoted by curl F, is a vector whose magnitude is the maximum net circulation of F per unit area as the area shrinks to zero and whose direction is the normal direction of the area when the area is oriented to make the net circulation maximum. We can define the b-directional net circulation of F at r precisely by

images

where images is the disk centered at r with radius r and normal to d. Then, curl F is its maximum net circulation:

images


Theorem 2.2.5 (Stokes's theorem)
Let Carea be an open smooth surface with its boundary as a smooth contour C. The surface integral of the curl of a C1-vector field F over the surface Carea is equal to the closed line integral of the vector F along the contour C:

images


 


Exercise 2.2.6
Prove that the expression for curl F in Cartesian coordinates is

images


 


Exercise 2.2.7
Let images and images. Prove the following vector identities.
1. A × (B × C) = (A · C)B − (A · B)C.
2. ∇ × ∇ × U = ∇(∇ · U) − ΔU.
3. ∇ × ∇u = 0.
4. If ∇ × U = 0, then there exists images such that ∇v = U in images.
5. ∇ · (∇ × U) = 0.
6. ∇ · (A × B) = B · (∇ × A) − A · (∇ × B).

 


Exercise 2.2.8
Let images. Let C be a closed curve and let Carea be a surface enclosed by C. Let Ω be a bounded domain in images. Prove the following:
1. images
2. images
3. images.

2.2.4 Curve

A curve images in images and images is represented, respectively, by

images

and

images

The curve images is said to be regular if r′(t) ≠ 0 for all t. The arc length s(t) of images between r(t0) to r(t1) is given by

images

The unit tangent vector T of the curve images is given by

images

The unit normal vector n of the curve images is determined by

images

where κ is the curvature given by

images

Here, we use |r′′ × r′| = |(s′)2κ T × (s′n)| = |(s′)3κ|. Note that dn/ds = − κ T since T · n = 0 and Ts · n + T · ns = 0.

2.2.5 Curvature

Consider a plane curve r(s) = (x(s), y(s)) in images, where s is the length parameter. If θ(s) stands for the angle between T(s) and the x-axis, then κ(s) = dθ/ds because

images

When the curve is represented as r(t) = (x(t), y(t)), then

images

Now, consider a curve images given implicitly by the level set of images:

images

Then, the normal and tangent vectors to the level curve are

images

The curvature κ is

images

To see this, assume that ϕ(r(t)) = 0 and set y′ = − ϕx and x′ = ϕy because ∇ϕ(r) · r′(t) = 0. Then,

images

imply

images

Hence,

images

The curvature κ at a given point is the inverse of the radius of a disk that best fits the curve images at that point.

Next, we consider a space curve images in images represented by

images

Then

images

The curvature κ and torsion τ are computed by

images

where [ ] stands for the triple scalar product. If the relation between the moving coordinate system {T, N, B} and the fixed coordinates {x, y, z} is

images

then

images

We consider a regular surface images:

images

Note that images. The condition corresponding to the regularity of a space curve is that

images

or

images

If |ru × rv| ≠ 0, the tangent vectors ru and rv generate a tangent plane that is spanned by ru and rv. The surface normal vector is expressed as

images

For example, if U = [0, 2π] × [0, π] and images, then images represents the unit sphere. If the surface images is written as z = f(x, y), then

images

and the unit normal vector n is

images

If the surface images is given by ϕ(r) = 0, then

images

2.3 Taylor's Expansion

The dynamics among system parameters, variables, excitations and measurements are often expressed as complicated nonlinear functions. In a nonlinear inverse problem, we may adopt a linearization approach in its solution-seeking process. Depending on the given problem, one may need to adopt a specific linearization process. In this section, we review Taylor's expansion as a tool to perform such an approximation.

Taylor polynomials are often used to approximate a complicated function. Taylor's expansion for images about x is

2.2 2.1

where O(|h|m+1) is the remainder term containing the (m + 1)th order of h. Precisely, the remainder term is

images

This expansion leads to numerical differential formulas of f in various ways, for example:

1. images (forward difference);
2. images (backward difference);
3. images (centered difference);
4. images.

The Newton–Raphson method to find a root of f(x) = 0 can be explained from the first-order Taylor's approximation f(x + h) = f(x) + hf′(x) + O(h2) ignoring the term O(h2), which is negligible when h is small. The method is based on the approximation

images

It starts with an initial guess x0 and generates a sequence {xn} by the formula

images

It may not converge to a solution in general. The convergence issue will be discussed later.

We turn our attention to the second derivative f′′(x). By Taylor's expansion, we approximate f′′(x) by

images

The sign of f′′(x) gives local information about f for a sufficiently small positive h:

1. images (mean value property, MVP);
2. images (sub-MVP);
3. images (super-MVP).

We consider a multi-dimensional case images. The following fourth-order Taylor's theorem in n variables will be used later.


Theorem 2.3.1 (Taylor's approximation)
For images,

images

where x = (x1, x2, …, xn), h = (h1, …, hn) and

images

which is called the Hessian matrix.

This leads to

images

If D2f(x) is a positive definite matrix, then, for a sufficiently small r,

images

which leads to the sub-MVP

images

Similarly, the super-MVP can be derived for a negative definite matrix D2f(x).


Theorem 2.3.2
Suppose images is a C3 function and ∇f(x0) = 0.
1. If f has a local maximum (minimum) at x0, then the Hessian matrix D2f(x0) is negative (positive) semi-definite.
2. If D2f(x0) is negative (positive) definite, then f has a local maximum (minimum) at x0.

 


Example 2.3.3
If images satisfies the Laplace equation images, then, for a small h > 0, we have

images

Hence, f(x) can be viewed approximately as the average of neighboring points:

images

This type of mean value property will be discussed later.

2.4 Linear System of Equations

We consider a linear system of equations including system parameters, variables, excitations and measurements. We may derive it directly from a linear physical system or through a linearization of a nonlinear system. Once we express a forward problem as a linear transform of inputs or system parameters to measured outputs, we can adopt numerous linear methods to seek solutions of the associated inverse problem. For the details of linear system of equations, please refer to Strang (2005, 2007).

2.4.1 Linear System and Transform

We consider a linear system in Figure 2.4. We express its outputs yi for i = 1, …, m as a linear system of equations:

2.3 2.2

Using two vectors images, images and a matrix images, we can express the linear system as

2.4 2.3

or

images

Figure 2.4 Linear system with multiple inputs and multiple outputs

2.4

In most problems, the output vector y consists of measured data. If the input vector x includes external excitations or internal sources, the matrix images is derived from the system transfer function or gain determined by the system structure and parameters. When the vector x contains system parameters, the linear system of equations in (2.4) is formulated subject to certain external excitations or internal sources, information about which is embedded in the matrix images.

For the simplest case of m = n = 1, it is trivial to find x1 = y1/a11. If m = 1 and n > 1, that is, images, it is not possible in general to determine all xj uniquely since they are summed to result in a single value y1. This requires us to increase the number of measurements so that mn. However, for certain cases, we will have to deal with the situation of m < n. Figure 2.5 illustrates three different cases of the linear system of equations. To obtain an optimal solution x by solving (2.4) for x, it is desirable to understand the structure of the matrix images in the context of vector spaces.

Figure 2.5 Three cases of linear systems of equations

2.5

2.4.2 Vector Space of Matrix

We denote by images a set of linear transforms from images to images, that is, images is the vector space consisting of m × n matrices

images

We call an m × 1 matrix a column vector and an 1 × n matrix a row vector. For the matrix A, we denote its ith row vector as

images

and its jth column vector as

images

For two vectors images, we define their inner product by

images

and define the outer product as

images

For images and images, we consider a linear transform or a linear system of equations:

images

We can understand it as either

2.5 2.4

or

2.6 2.5

Figure 2.6 shows these two different representations of a linear system of equations images for the case of m = n. In (2.5), each output yi is a weighted sum of all images, and the row vector, row(A;i), provides weights for yi. In (2.6), the output vector y is expressed as a linear combination of n column vectors, images with weights images. It is very useful to have these two different views about the linear transform in (2.4) to better understand a solution of an inverse problem as well as the forward problem itself.

Figure 2.6 Two different representations of a linear system of equations y = Ax

2.6

We now summarize the null space, range and rank of a matrix images before we describe how to solve images for x.


Definition 2.4.1
We have the following:
  • The null space of a matrix images is

images

  • The range space of a matrix images is

images

  • The rank of a matrix images, denoted by rank(A), is the maximum number of independent columns or rows.

From (2.5) and (2.6), we have

2.7 2.6

and

2.8 2.7

If N(A) = {0}, the following are true:

1. col(A;1), …, col(A;n) are linearly independent.
2. images is invertible, that is, Ax = y uniquely determines x.
3. Every yR(A) can be decomposed uniquely as images.
4. A has a left inverse, that is, there exists images such that 〈row(B;j), col(A;k) 〉 = 0 for jk and 〈row(B;j), col(A;j) 〉 = 1.
5. det(ATA) ≠ 0.

If images, the following hold:

1. images.
2. images is linearly independent and N(AT) = {0}.
3. A has a right inverse, that is, there exists images such that 〈row(A;j), col(B;k) 〉 = 0 for jk and 〈row(A;j), col(B;j) 〉 = 1.
4. det(AAT) ≠ 0.

We also note that

images

2.4.3 Least-Squares Solution

We consider a linear system of equations y = Ax. If there are more equations than unknowns, that is, m > n, the system is over-determined and may not have any solution. On the other hand, if there are fewer equations than unknowns, that is, m < n, the system is under-determined and has infinitely many solutions. In these cases, we need to seek a best solution of y = Ax in an appropriate sense.


Definition 2.4.2
Let images. Then
  • x* is called the leastsquares solution of y = Ax if

images

  • x is called the minimumnorm solution of y = Ax if x is a least-squares solution of y = Ax and

images


If x* is the least-squares solution of y = Ax, then Ax* is the projection of y on R(A), and the orthogonality principle yields

images

If ATA is invertible, then

images

and the projection matrix on R(A) can be expressed as

2.9 2.8

Since ||x||2 ≥ ||x||2 for all x such that y = Ax,

images

that is, x is orthogonal to N(A).


Exercise 2.4.3
Considering Figure 2.7, explain the least-squares and minimum-norm solutions of images.

Figure 2.7 Least-squares and minimum-norm solutions of y = Ax

2.7

2.4.4 Singular Value Decomposition (SVD)

Let A be an m × n matrix. Then ATA can be decomposed into

images

where v1, …, vn are orthonormal eigenvectors of ATA. Since V is an orthogonal matrix, V−1 = VT.

If we choose ui = (1/σi)Avi, then

2.10 2.9

and

2.11 2.10

Since V−1 = VT, we have the singular value decomposition (SVD) of A:

2.12 2.11

where

images

Suppose we number the ui, vi and σi so that σ1 ≥ σ2 ≥ ··· ≥ σr > 0 = σr+1 = ··· = σn. Then, we can express the singular value decomposition of A as

2.13 2.12

This has the very useful property of splitting any matrix A into rank-one pieces ordered by their sizes:

2.14 2.13

Figure 2.8 shows two graphical representations of the SVD. If σt+1, …, σr are negligibly small, we may approximate images by the truncated SVD as

2.15 2.14

with t < r.

Figure 2.8 Graphical representations of the SVD, images

2.8

We may interpret the linear transform images as shown in Figure 2.9. First, VTx provides coefficients of x along the input directions of images. Second, ΣVTx scales VTx by images. Third, UΣVTx reconstructs the output y in the directions of images. From the relation Avi = σiui, we can interpret v1 as the most sensitive input direction and u1 as the most sensitive output direction with the largest gain of σ1.

Figure 2.9 Interpretation of images

2.9

2.4.5 Pseudo-inverse

The pseudo-inverse of A is

2.16 2.15

Here, in the pseudo-inverse Σ of the diagonal matrix Σ, each σ ≠ 0 is replaced by 1/σ.

Since

2.17 2.16

the products AA and AA can be viewed as projection matrices:

2.18 2.17

We also know that Ay is in the row space of A = span{u1, …, ur} and Ax = y is solvable only when y is in the column space of A.

The least-squares solution x = Ay minimizes

2.19 2.18

Hence, x satisfies the normal equation

2.20 2.19

Moreover, x is the smallest solution of ATAx = ATy because it has no null components.

2.5 Fourier Transform

Joseph Fourier (1768–1830) introduced the Fourier series to solve the heat equation. Since then, Fourier analysis has been widely used in many branches of science and engineering. It decomposes a general function f(x) into a linear combination of basic harmonics, sines or cosines, that is easier to analyze. For details of the theory, refer to Bracewell (1999) and Gasquet and Witomski (1998). We introduce the Fourier transform as a linear transform since it is also widely used in the inverse problem area.

2.5.1 Series Expansion

Assume that a set {ϕ0, ϕ1, …} is an orthonormal basis of a Hilbert space H, that is:

1.0, ϕ1, … } is orthonormal;
2. every f ∈ H can be expressed as f = ∑cjϕj.

If we denote by Pn the projection map from H to Vn, then

images

If we denote images, then images. By the Pythagorean theorem, we have

images

and we obtain the Bessel inequality

2.21 2.20

By the Bessel inequality, the series

images

Moreover,

images

or

images

Hence,

images

This gives us the following series expansion:

images


Exercise 2.5.1
Let VN = span{ 1, cos2πx, sin2πx, …, cos2πNx, sin2πNx}. Prove that f ∈ VN can be expanded as the following trigonometric series:

images

where

images



Example 2.5.2
Let VN = span{e2πikx | k = 0, ± 1, …, ±N} be the subspace of L2([0, 1]) and let H be the closure of images. Then, any complex-valued function f ∈ H can be uniquely represented by

images

To see this, note that

images

The projection map PN from H to VN can be expressed as

images

From the Bessel inequality, we have

images

Hence, we have the Riemann–Lebesgue lemma:

images

and

images


2.5.2 Fourier Transform

For each p with 1 ≤ p < ∞, we denote the class of measurable functions f on images such that images by images:

images

where the Lp-norm of f is

images


Definition 2.5.3
We have the following:
1. If images, the Fourier transform of f is images defined by

images

2. The convolution of two functions f and g is given by

images

3. The Dirac delta function δ is a distribution satisfying

images

4. The support of f is the set images.
5. The function f is bandlimited if images is bounded. The function f is said to be band-limited with width L if images.
6. The Gaussian function with a mean value b and a standard deviation σ is

images

The Gaussian function has the property that images and limσ→0Gσ, b(x) = δ(x − b).

Let us summarize the Fourier transform pairs of some important functions.

  • Scaling property:

images

  • Shifting property:

images

  • Dirac delta function:

images

  • Dirac comb:

images

  • Indicator function of the interval [a, b], denoted by χ[a, b](x), is

images

We have

images

  • The Fourier transform of Gaussian function images is given by

images

For the above identity images, we use the Cauchy integral formula that images for any closed curve C.
  • The Fourier transform of f * g(x) = ∫f(x − y)g(y) dy is

images

  • Modulation: if g(x) = f(x)cos(xi0x), then

images

  • Derivative: if images, then integrating by parts leads to

images

The following Fourier inversion provides that f can be recovered from its Fourier transform images.


Theorem 2.5.4 (Fourier inversion formula)
If images, then

2.22 2.21


 


Proof.
The inverse Fourier transform of images can be expressed as

2.23 2.22

The last identity can be obtained by interchanging the order of the integration. From the identity images and the scaling property images, the last quantity in (2.23) can be expressed as

images

Then, the identity (2.23) can be simplified into

images

Since limimages→0ϕimages(x) = 0 for all x ≠ 0 and images, we have

images


Indeed, the Fourier inversion formula holds for general images because images is dense in images and images. We omit the proof because it requires some time-consuming arguments of Lebesgue measure theory and some limiting process in L2. For a rigorous proof, please refer to Gasquet and Witomski (1998).

The following Poisson summation formula indicates that the T-periodic summation of f is expressed as discrete samples of its Fourier transform images with the sampling distance 1/T.


Theorem 2.5.5 (Poisson summation formula)
For images, we have

2.24 2.23

In particular,

images


 


Proof.
Denoting images, we have

images

The T-periodic function f * combT(x) can be expressed as

images

where

images

This an is given by

images


 


Theorem 2.5.6
If images, then

images

From this, we again derive the Poisson summation formula:

2.25 2.24


 


Proof.
Since images is 1/T-periodic, it suffices to prove

images

The above identity can be derived from the following facts.
1. images
2. images
3. For any images with ϕ(0) = 0,

images


The band-limited function f with bandwidth B, that is, images, does not contain sinusoidal waves at frequencies higher than B. This means that f cannot oscillate rapidly within a distance less than 1/(2B). Hence, the band-limited f can be represented by means of its uniformly spaced discrete sample {f(nΔx):n = 0, ± 1, ± 2, …} provided that the sampling interval Δx is sufficiently small. Indeed, the following sampling theorem states that, if Δx ≤ 1/(2B), then the discrete sample {f(nΔx):n = 0, ± 1, ± 2, …} contains the complete information about f. This 2B is called the Nyquist rate.


Theorem 2.5.7 (Whittaker–Shannon sampling theorem)
Suppose images and images. The original data f can be reconstructed by the interpolation formula

images

where sinc(x) = sin(πx)/(πx).

 


Proof.
It is enough to prove the above sampling theorem for images. Denoting

images

the sampled version f(x) comb(x) is

images

Taking the Fourier transforms of the above identity, we obtain

images

The last equality comes from the Poisson summation formula (2.24). From the assumption images,

images

This means that

images

Since images, we get

images

This gives

images


 


Remark 2.5.8
When we need to analyze an analog signal, it must be converted into digital form by an analog-to-digital converter (ADC). Assume that images is the analog signal to be analyzed and 1/T is the sampling interval. According to the Poisson summation formula, the T-periodic function f * combT is reconstructed from the samples images. This result is based on the fact that the Fourier transform of images is (1/T) comb1/T(xi). Hence, if f(x) ≠ 0 for images, the formula (2.25) produces aliasing or wrap-around. To avoid aliasing, we require that supp(f) ⊂ [ − T/2, T/2], that is, the spatial frequency of f should be less than T/2.

2.5.3 Discrete Fourier Transform (DFT)

Assume that f is a continuous signal such that

images

Choose a number N so that

images

With the condition N ≥ TΞ, the original signals f and images are approximately recovered from the samples {f(kT/NT/2):k = 0, 1, …, N − 1} and images based on either the Poisson summation formula or the Whittaker–Shannon sampling theorem.

Let us convert the continuous signal f into the digital signal {f(kΔx − T/2):k = 0, 1, …, N − 1} with the sampling spacing Δx = T/N. The points xk = kΔx − T/2, with k = 0, 1, …, N − 1, are called the sampling points.

Writing fk = f(kΔx − T/2), the digital signal corresponding to the continuous signal f can be expressed in the following vector form:

images

Its Fourier transform images for xi ∈ [ − Ξ/2, Ξ/2] is expressed approximately by

images

Similarly, we denote images where xij = jΔxi − Ξ/2 and Δxi = Ξ/N. Then, f(x) for x ∈ [ − T/2, T/2] is expressed approximately by

images

In particular, we have

images

Since

images

we have the following approximations:

images

From the above approximations, we can define the discrete Fourier transform.


Definition 2.5.9
The discrete Fourier transform (DFT) on images is a linear transform images given by

images

where γN: = exp( − 2πi/N) is the Nth principal root of 1. Here, the DFT linear transform images can be viewed as an N × N matrix.

The DFT matrix images has the following interesting properties.

  • Let aj−1 be the jth row (or column) of the Fourier transform matrix images. Then,

images

  • The column vectors a0, …, aN−1 are eigenvectors of the following matrix corresponding to the Laplace operator:

images

  • Hence, {a0, …, aN−1} forms an orthogonal basis over the N-dimensional vector space images and

images

where A* is the complex conjugate of the transpose of the matrix A. The inverse of the DFT linear transform images is simply its transpose:

images

  • If images and images are the DFTs of f and g, then we have Parseval's identity:

images

where the inner product is defined by images.

2.5.4 Fast Fourier Transform (FFT)

The fast Fourier transform (FFT) is a DFT algorithm that reduces the number of computations from something on the order of N2 to NlogN. Let N = 2M and

images

From the definition of the DFT, we have

images

Since images and images, the (M + j, 2k)-component of the matrix images is

images

and the (M + j, 2k − 1)-component of the matrix images is

images

The FFT is based on the following key identity:

2.26 2.25

where images is the N × N DFT matrix defined in the previous section and

images


Remark 2.5.10
Assume that f is supported in images and that f = (fN/2, …, f(N/2)−1) is a digital image of f(x), where fn = f(nΔx) and Δx is the sampling interval (or pixel size). The interval images is referred to as the field of view (FOV). For simplicity, assume NΔx = 1, that is, images. The number N may be regarded as the number of pixels. Assume that f is band-limited with images, that is, images and denote images. We can recover f from f without any loss of information provided that it meets the Nyquist criterion:

images

Hence, if Δxi = 1/(NΔx), the DFT gives

images

Extending the sequence f = (fN/2, …, f(N/2)−1) to the N-periodic sequence in such a way that fn+mN = fn, the discrete version of the Poisson summation formula is

images


2.5.5 Two-Dimensional Fourier Transform

The one-dimensional definition of the Fourier transform can be extended to higher dimensions. The Fourier transform of a two-dimensional function ρ(x, y), denoted by S(kx, ky), is defined by

images

We can generalize all the results in the one-dimensional Fourier transform to the two-dimensional case because the two-dimensional Fourier transform can be expressed as two one-dimensional Fourier transforms along x and y variables:

images

If ρ(x, y) and S(kx, ky) are a Fourier transform pair, we have the following properties.

  • Scaling property:

images

  • Shifting property:

images

  • Modulation:

images

  • Fourier inversion formula:

images

  • Sampling theorem: if supp(S) ⊂ [ − B, B] × [ − B, B], then the original data ρ(x, y) can be reconstructed by the interpolation formula

images

References

Bracewell R 1999 The Fourier Transform and Its Applications, 3rd edn. McGraw-Hill, New York.

Gasquet C and Witomski P 1998 Fourier Analysis and Applications: Filtering, Numerical Computation, Wavelets. Springer, New York.

Rudin W 1970 Real and Complex Analysis. McGraw-Hill, New York.

Strang G 2005 Linear Algebra and Its Applications, 4th edn. Thomson Learning, London.

Strang G 2007 Computational Science and Engineering. Wellesley-Cambridge, Wellesley, MA.

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
3.15.219.174