Fundamental course: Fourier Optics

Lecture notes
Revision date: August 27, 2018

Simulation of the Point-Spread Function of a 39 pupils

interferometer, with sub-apertures disposed on 3
concentric rings.
Contents

0 Reminders about Fourier analysis 4

0.1 Some useful functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.1.1 The rectangle function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.1.2 Dirac delta distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
0.1.3 Dirac comb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
0.2 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.2.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.3 Fourier transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
0.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
0.3.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
0.3.3 Table of Fourier transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1 Reminders about diffraction 12

1.1 Some particular kinds of waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.1.1 Monochromatic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.1.2 Plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.1.3 Spherical waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Huyghens-Fresnel principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.2 Paraxial approximation and Fresnel diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.3 Far field: Fraunhofer diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.4 Example of Fraunhofer diffraction patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 Fourier properties of converging lenses 22

2.1 Phase screens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.1 Transmission coefficient of a thin phase screen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.2 Example: prism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.1.3 Converging lens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Converging lenses and Fourier transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.1 Object in the same plane as the lens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.2 Object at the front focal plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3 Coherent optical filtering 27

3.1 Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Abbe-Porter experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 Object-image relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4 Low-pass and high-pass filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.1 Low-pass filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.2 High-pass filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5 Strioscopy and phase contrast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5.1 Strioscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5.2 Phase contrast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2
4 Image formation 35
4.1 Object-image convolution relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.1 Image created by a point-source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.2 Object-image relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Optical transfer function (OTF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3 Case of a circular pupil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3.1 PSF and resolving power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3.2 Optical transfer function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3
Chapter 0

Reminders about Fourier analysis

Also read:
• Bracewell, R. “The Fourier Transform and its applications”
• Goodman, J.W. “Introduction to Fourier Optics”, chap 2

• Roddier, F., “Distributions et transformation de Fourier” (in french)

0.1 Some useful functions

0.1.1 The rectangle function
The rectangle function is useful to describe objects like slits or diaphragms whose transmission is 0 or 1. It is defined as
1,0

0,75
1
Π(x) = 1 if |x| < 2

(1) 0,5
Π(x) = 0 otherwise

0,25

0,0
−1,0 −0,5 0,0 0,5 1,0
x

1
 (in particular for the value at x = ± 2 ). A rectangular function of
Some other definitions may be found in the literature
x−b
width a centered at x = b will express as Π .
a

2D rectangle function
We consider functions of 2 variables x and y. The quantity

f (x, y) = Π(x) = Π(x) 1(y)

describes a strip of width 1 parallel to the y axis : it is invariant by translation along y (see Fig. 1). The notation 1(y)
stands for a function which value is 1 whatever y.
A two dimensional rectangle function of width a in the x direction and b in the y direction expresses as
x y
f (x, y) = Π Π (2)
a b
We whall use this function throughout this course to express the transmission coefficient of rectangular slits.

4
Figure 1: 2D rectangle function f (x, y) = Π xa Π yb of width a in the x direction and b in the y. Left: perspective plot as

a function of x and y. Right: gray-level representation in the (x, y) plane.

ρ


Figure 2: 2D circular function f (x, y) = Π d of diameter d. Left: perspective plot as a function of x and y. Right:
gray-level representation in the (x, y) plane.

2D circular function
We consider the following quantity: ρ
f (x, y) = Π (3)
d
p
with ρ = x2 + y 2 . Its value is one for ρ < d2 , i.e. inside a disc of diameter d. This function will be used to describe
transmission coefficient of circular diaphragms.

0.1.2 Dirac delta distribution

The Dirac delta distribution (also known as “Dirac impulse”) δ(x) can be defined as
a function which is 0 if x 6= 0 and infinite for x = 0. Its integral is 1:
Z ∞
δ(x) dx = 1 (4)
−∞

and this last relation shows that δ(x) has the dimension of [x]−1 . The δ distribution
is often represented as a vertical arrow centered as x = 0, of height 1, as in the graph
on the right.
1
The Dirac δ is sometimes defined as the limit of a rectangular function of width  → 0 and height  so that its integral
remains 1:
1 x
δ(x) = lim Π (5)
→0  
An important property is:
f (x) δ(x − a) = f (a) δ(x − a) (6)
where δ(x − a) is the Dirac impulse centered at x = a. Some other properties are:

5
 δ (x−a,y−b)
δ (x,y)
y
y
b
x a
x
Figure 3: Graphic representation of the 2D Dirac impulses. Left: δ(x, y). Right: shifted impulse δ(x − a, y − b) centered at
x = a, y = b.

1
• δ(ax) = |a| δ(x) (for x 6= 0)
d
• δ(x) = dx H(x) where H(x) is the Heaviside distribution (1 if x > 0, 0 if x < 0).

2D Dirac impulse
We define the 2D Dirac distribution as
δ(x, y) = δ(x).δ(y) (7)
It is 0 in the whole plane (x, y) excepted at the origin where it is infinite. Its integral is 1:
ZZ ∞
δ(x, y) dx dy = 1 (8)
−∞

The 2D Dirac impulse can be considered as the limit of a rectangular function of width  in both directions x and y (the
surface of the rectangle being 2 ):
1 x y
δ(x, y) = lim 2 Π Π (9)
→0   
2
It is also the limit of a 2D circular function of diameter  (and of surface s = π 4 )

1 ρ
δ(x, y) = lim Π (10)
→0 s 
The 2D Dirac distribution is often used in optics to describe the amplitude of a point-source, or the transmission coefficient
of a pin-hole (diaphragm with very small diameter).

0.1.3 Dirac comb

The Dirac comb is a periodic succession of Dirac impulses:

Ø(x) =
∞
X 1
δ(x − n) (11)
n=−∞

its period is 1 in the standart form above, its graph (on the right) looks like a −3 −2 −1 0 1 2 3 x
comb, hence its name.

It is possible to define a comb of period a, having all δ impulses located at x = na (n integer) as

Ø(x) = Ø  xa 
∞
X 1
δ(x − na) = (12)
a
n=−∞
|a|

The comb is of fundamental importance in signal processing: it is the tool used to describe mathematically the operation of
sampling. In optics it is used to describe periodic structures such as diffraction gratings.

6
2D Dirac comb

The 2D Dirac comb (sometimes denoted as “Dirac brush”) is the product of 2

combs in directions x and y:

Ø(x). Ø(y) =
∞
X ∞
X
δ(x − na, y − pb) (13)
a b
n=−∞ p=−∞

with a and b the periods in x and y directions.

0.2 Convolution
0.2.1 Definition
The convolution product of two functions f and g is defined as
Z ∞
h(x) = (f ∗ g)(x) = f (x0 )g(x − x0 )dx0 (14)
−∞

It can be interpreted as a weighted moving average of the function f (g(−x) being the weighting function). In signal
processing, g is denoted as “impulse response”. If g is a function such as a Gaussian or a rectangle, the convolution will
result into a smoothing of the details of f .

2D convolution
The 2D convolution between two functions of (x, y) is
ZZ ∞
h(x, y) = (f ∗ g)(x, y) = f (x0 , y 0 )g(x − x0 , y − y 0 )dx0 dy 0 (15)
−∞

Note that we use the same symbol ∗ for 1D and 2D convolutions, but the two operations are different (single integral for 1D,
double integral for 2D). The 2D impulse response is sometimes denoted as “point-spread function”. The 2D convolution has
a lot of applications in image processing; for example convolving an image f (x, y) by a 2D rectangle function will blur the
image.

0.2.2 Properties
Here are some properties of the convolution:

• It is commutative (f ∗ g = g ∗ f ) and associative (f ∗ (g ∗ h) = (f ∗ g) ∗ h))

R
• The convolution by 1 gives the integral of the function: f (x) ∗ 1(x) = f (x)dx (this can be useful for certain types of
calculations).
• Dilatation: (f ∗ g)(λx) = |λ|f (λx) ∗ g(λx)

• Convolving by δ(x) has no effect (f (x) ∗ δ(x) = f (x)). This property is the origin of the term “impulse response” for
g in the relation (f ∗ g).
• Convolving by a shifted impulse δ(x − a) translates the function: f (x) ∗ δ(x − a) = f (x − a). This property is very
important and useful for Fourier optics calculations.

• Convolution by a comb:
Ø(x) =
∞
X
f (x) ∗ f (x − na) (16)
a
n=−∞

this is a periodization of the function f (each impulse of the comb is replaced by f , see Fig. 4).

7
Figure 4: The convolution of a function f by a comb results into a periodization of f (Eq 16). Left, the comb of period a.
Center: the function f (here a rectangular function). Right: result of the convolution.

Figure 5: Property of translation of the convolution (Eq. 17): a function f (x, y) (on the left) centered at the origin is
convolved by a Dirac impulse centered at (x = a, y = b). The result (on the right) is the shifted function f (x − a, y − b)
centered at (x = a, y = b).

2D convolution
Most of the above properties apply to 2D convolution. In particular this one:

f (x, y) ∗ δ(x − a, y − b) = f (x − a, y − b) (17)

which is illustrated by the Figure 5. A translation of a function inside the (x, y) plane can be expressed as a convolution by
a shifted 2D Dirac impulse. This is the origin of the name “point-spread function” (PSF) for the impulse response at 2D
(a 2D Dirac impulse is a infinitely sharp point in the (x, y) plane, and the convolution transforms this point into a larger
function f ).
A corollar of this property is:
X X
f (x, y) ∗ An δ(x − xn , y − yn ) = An f (x − xn , y − yn ) (18)
n n

which is illustrated by Figs. 6 and 7.

Figure 6: Illustration of Eq. 18. (a) gray-scale plot of a sum of 2D Dirac impulses with different amplitudes. (b) gray-
scale plot of the point-spread function f (x, y) (f (x, y) = 1 inside an octogonal domain, 0 elsewhere). (c) result of the 2D
convolution of the two functions. As predicted by Eq. 18, the result is a sum of shifted PSFs (each impulse of the sum is
replaced by the PSF, with the same amplitude An ) When PSFs overlap, the result is the sum of overlapping terms.

8
Figure 7: Image of portion of sky with a defocused optics: each star has the shape of a small disc, with a central obstruction.
This is a typical illustration of a 2D convolution as in Fig. 7. The perfect image f (x, y) is composed of a sum of 2D impulses
(ideal image of a point-source). It is convolved by a Point-Spread function g(x, y) which is the small disc (two examples are
in the red boxes). The fuzzy objet on the top right is the Dumbbell nebula, which is also convolved by the PSF (so that
every point of the nebula is replaced by the PSF, resulting in a blurred image).

0.3 Fourier transform

0.3.1 Definition
The Fourier transform of a function f (x) is defined as
Z ∞
fˆ(u) = f (x) e−2iπux dx (19)
−∞

it can be noted as fˆ(u) or F[f ]. It is sometimes denoted as frequency spectrum, since it comes from the idea that a function
can be developped as a weighted sum of complex sinusoids. The value fˆ(u) represents the wheight of the sinusoid of frequency
u in the sum.
If x is a length, the variable u is a spatial frequency, having the dimension of [x]−1 . The spatial frequency of a space-
dependent sinusoidal function (for ex. cos(2πux)) plays the same role as the temporal frequency for a time-dependent
function, it represents the number of periods per unit of length. The dimension of fˆ(u) is

[fˆ(u)] = [f ].[x] (20)

A very important property is

F[exp(2iπu0 x)] = δ(u − u0 ) (21)
i.e. the Fourier transform of a complex sinusoid of frequency u0 is a Dirac impulse centered at u0 . This emphasizes the fact
that the complex sinusoid has only one frequency in its spectrum.
The inverse Fourier transform is Z ∞
f (x) = fˆ(u) e+2iπux du = F −1 [fˆ] (22)
−∞

2D Fourier transform
The 2D Fourier transform of a function of two variables f (x, y) is defined as
ZZ ∞
ˆ
f (u, v) = f (x, y) e−2iπ(ux+vy) dx dy (23)
−∞

The variables u and v are spatial frequencies associated to the space variables x and y. They define, in the (u, v) plane, a
 u
“spatial frequency vector” ~σ =  (see Fig. 8). As for the 1D Fourier transform, the idea is that a function f (x, y) can be
v
expressed as a sum of 2D complex sinusoids of any period and any orientation.
The 2D inverse Fourier transform is ZZ ∞
f (x, y) = fˆ(u, v) e+2iπ(ux+vy) dx dy (24)
−∞

9
Figure 8: Representation of a 2D sinudoidal function f (x, y) = cos(2π~σ .~ρ) = cos(2π(ux + vy)). (a): perpective plot. (b):
grayscale plot. The frequency vector ~σ = (u, v) has been drawn on the right plot; it is perpendicular to the ridge lines of the
function, its modulus is the frequency of the oscillations measured along the unit vector σ̂. Its components are u = a1 , v = 1b
with a and b the periods in the x and y directions.

0.3.2 Properties
Here is a short list of useful properties of the 1D Fourier transform
• Dilatation: a function which is large in the direct plane will be narrow in the Fourier plane
x
F
f −→ |a| fˆ(au) (25)
a

• Convolution and product:

F
f (x) . g(x) −→ fˆ(u) ∗ ĝ(u)

F
f (x) ∗ g(x) −→ fˆ(u) . ĝ(u) (26)

• Derivation: the Fourier transform of the derivative of a function is a high-pass filtering in the frequency plane (product
by u which strengthens the high frequencies)

df F
−→ 2iπufˆ(u)
dx

F 1 dfˆ
x . f (x) −→ − (27)
2iπ du

• Value at frequency u = 0: it is the integral of the function (this property in sometimes interesting to calculate integrals)
Z ∞
fˆ(0) = f (x) dx (28)
−∞

• Sign change for the variable x:

F
f (−x) −→ fˆ(−u) (29)

• Complex conjugate:
F
f (x) −→ fˆ(−u) (30)

• Double Fourier transform:

F F ˆ
f (x) −→ fˆ(u) −→ f (−x) (or: fˆ(x) = f (−x)) (31)

• Fourier transform of real-valued functions: they are Hermitian, with an even real part and an odd imaginary part. It
can be summarized by
is f (x) real , then fˆ(u) = fˆ(−u) (32)

10
 • Real/even functions: if a function f is real and even, then
– Its Fourier transform is also real and even (no imaginary part)
– Its inverse transform is equal to its direct transform: F −1 [f ](u) = F[f ](u)
• Parseval theorem: Z ∞ Z ∞
|fˆ(x)|2 dx = |fˆ(u)|2 du (33)
−∞ −∞

Specific properties for 2D Fourier transform

Separable functions: if a function h(x, y) is the product of two functions of one variable f (x) and g(y), then its 2D Fourier
transform is also a separable function, i.e.
F
h(x, y) = f (x) . g(y) −→ ĥ(u, v) = fˆ(u) . ĝ(v) (34)

This property must not to be confused with the Fourier transform of a product of functions of the same variables
(Eq. 26): here the variables for f and g are different, and the 2D transform is a double integral.
ˆ
p
Radial functions of 2 2
√ the type f (x, y) = f (ρ) with ρ = x + y : the 2D Fourier transform f (u, v) is also a radial function
2 2
F (q) with q = u + v . It takes the following form known as Hankel transform:
Z ∞
ˆ
F (q) = f (u, v) = 2πρ f (ρ) J0 (2πqρ) dρ (35)
0

where J0 (x) is the zero order Bessel function. The Hankel transform f (ρ) −→ F (q) is not to be confused with the 1D
Fourier transform (Eq. 19).

0.3.3 Table of Fourier transforms

1D transforms
Function Fourier transform Function Fourier transform
δ(x) 1(u) 1(x) δ(u)
δ(x − a) exp(−2iπua) exp(2iπmx) δ(u − m)
 
1 1 x
Heaviside H(x) δ(u) + VP Π |a| sinc(πua)
H(x) = 1 if x > 0, 0 otherwise
2 2iπu a with sinc(x) =
sin(x)
x

Ø (t)
a Ø(au) Triangle Λ
x
a
|a| sinc2 (πua)
Λ(x) = 1 − |x| if |x| ≤ 1, 0 otherwise
1 1 1 1
cos(2πmx) δ(u − m) + δ(u + m) sin(2πmx) − δ(u − m) + δ(u + m)
  x  2 2 2 2
2|a| 1
exp −   π|a| exp (−2π|au|)
 

2
a 1 + 4π 2 a2 u2 1+ xa
  x 2   x 2  √
exp −π |a| exp(−πa2 u2 ) exp iπ ia2 exp(−iπa2 u2 )
a a

2D transforms (radial functions)

Function Fourier transform Function Fourier transform
1 exp(−2πaq)
δ(ρ − a) 2πaJ0 (2πaq) p
ρ + a2
2 q
ρ ρ
Π 2S jinc(πdq) 4S jinc(πdq)2
d jinc(x) =
J1 (x)
S= πd2 d
x 4 (ρ) =Π(ρ) ∗ Π(ρ)
2

2 2 ρ
exp(−πρ ) exp(−πq ) exp iπ ia2 exp(−iπa2 q 2 )
a2
p √
with ρ = x2 + y 2 and q = u2 + v 2

11
Chapter 1

Reminders about diffraction

Also read : Goodman, ‘Introduction to Fourier Optics”, chap. 3, 4

1.1 Some particular kinds of waves

1.1.1 Monochromatic waves
Electromagnetic waves are created by the propagation of an electromagnetic field (E, ~ B).
~ Both are vectorial quantities, but
the majority of diffraction phenomena can be explained by considering the scalar quantity E(x, y, z, t) (either the electric or
magnetic field, without vectors).
A monochromatic wave is characterised by a time dependence in e−iωt . It has only one pulsation ω (with ω = 2πν where
ν is the temporal frequency, typically 1015 Hz for visible light). It also has a unique wavelength λ = νc (or λ = nν c
if the
refraction index is n) whose dimension is a length (in m).
The field E(x, y, z, t) takes the following form of a product of a spatial term and a temporal term:

E(x, y, z, t) = ψ(x, y, z) e−iωt (1.1)

where ψ(x, y, z) (or ψ(~r)), the spatial part of the field, is called the complex amplitude of the vibration. In what follows we
will consider that all the waves are monochromatic, and will deal with complex amplitudes.

1.1.2 Plane waves

Wavefronts (i.e. surfaces where the electric field value is constant at a given time) are planes. The distance between two
consecutive wavefonts is the wavelength λ (two consecutive wavefronts are characterized by a 2π phase difference between
complex amplitudes in both planes).
We define the wave vector ~k = 2π 2π
λ k̂. Its norm is λ and its unit vector k̂ is parallel to the direction of propagation (see
Fig. 1.1).

plane wavefront

wave vector

wavelength image : wikipedia

Figure 1.1: Structure of a plane wave

12
 ~ of a monochromatic plane wave takes the form
The electric field E

~ = ψ0 ei(~k.~r−ωt) Ê
E (1.2)

with ψ0 a constant [unit: V/m] and Ê a unit vector, perpendicular to ~k. Note that E
~ is constant in a plane perpendicular
~ ~
to k (i.e. wave planes are perpendicular to k, see Fig. 1.1).
We denote as α, β and γ the three components of the unit
vector k̂ (projections of k̂ onto the 3 axes as in the scheme
on the right). The wave vector expresses as

~k = 2π (αx̂ + β ŷ + γ ẑ) (1.3)

λ
p
with the condition α2 + β 2 + γ 2 = 1 (k̂ is a unit vector).
The case (α = β = 0, γ = 1) corresponds to a propagation
parallel to the z axis.
With these notations, the complex amplitude of a monochromatic plane wave takes the form
 
~ 2iπ
ψ(~r) = ψ0 eik.~r = ψ0 exp (αx + βy + γz) (1.4)
λ

It exhibits a phase term which is a linear function of coordinates (x, y, z).

Intensity It is the electromagnetic power per surface unit carried by the electromagnetic wave. It is proportional to the
square modulus of the complex amplitude
I = (C te ) |ψ(~r)|2 (1.5)
the multiplicative constant is generally taken as unity for a sake of simplicity.

1.1.3 Spherical waves

Wavefronts are concentric spheres (Fig. 1.2). If the center of these spheres is at the point (0, 0, 0), the complex amplitude
expresses as
S0 ik.r S0 −ik.r
ψ(~r) = e or ψ(~r) = e (1.6)
r r
without vectors in the complex exponential. The sign + (resp. −) denotes a diverging (resp. converging) spherical wave:
the radius of the wave-spheres increase (resp decrease) with time. Note that
• The point r = 0 (center of the spheres) is singular, the field value diverges. It can be a point-source (case of a diverging
wave) or a point of focalisation (converging wave).
• The dimension of S0 is not that of an amplitude ([S0 ] = [ψ].[r], in V ).

Case of a spherical wave not centered at the origin: if we call ~r0 = (x0 , y0 , z0 ) the center of wave-spheres, a simple
translation allows to write the complex amplitude as:
S0
ψ(~r) = e±ik.|~r−~r0 | (1.7)
|~r − ~r0 |
p
with |~r − ~r0 | = (x − x0 )2 + (y − y0 )2 + (z − z0 )2 .

13
 Figure 1.2: Structure of a spherical wave.

Paraxial approximation

It is an important approximation in Fourier optics, the majority of

relations will be derived within this approximation. We consider
that the segment joining the source and the point at ~r is nearly
parallel to the z-axis, as illustrated by the scheme on the right. If
the source is at the origin, we have |x|  |z| and |y|  |z|, and
make the following approximation for r:

ρ2
r ' |z| + with ρ2 = x2 + y 2 (1.8)
2|z|

The complex amplitude in paraxial approximation (if the source is at the origin) becomes
iπρ2
 
S0 ±ik|z|
ψ(~r) ' e . exp ± (1.9)
|z| λ|z|
| {z } | {z }
plane wave phase curvature
It is the product of two terms
S0
• A plane wave e±ik|z| with wave-planes perpendicular to z,
|z|
 2

• A quadratic phase term exp ± iπρ
λ|z| which can be interpreted as a deviation from the plane wave. The wavefronts
associated to this term are paraboloı̈ds (a paraboloı̈d is indeed the 2nd order approximation of a sphere). This term
in sometimes denoted as phase curvature because it is a pure phase term and in reference to the curvature of the
wavefronts (see Fig. 1.3).
Plots of the paraxial form of the spherical wave amplitude are shown in Fig. 1.4. In a√plane z = C te , it is a centro-symmetric
function whose real and imaginary part show concentric rings of characteristic size λz (for z > 0).
For large z, the phase curvature vanishes and the spherical wave becomes plane. This happens when |z|  ρ2 /λ (1m for
λ = 1µm and ρ = 1mm).

Validity of the approximation

The approximate expression for r in Eq. 1.8 is the leading terms of a Taylor expansion. The approximation is valid if the
following term is small:
ρ2 ρ4
r ' |z| + + + ... (1.10)
2|z| 8|z|3
The corresponding complex amplitude is
iπρ2 iπρ4
   
S0 ±ik|z|
ψ(~r) ' e exp ± exp ± (1.11)
|z| λ|z| 4λ|z|3

14
Figure 1.3: Paraxial approximation of a spherical wave, in the case of z  ρ. In this example, z is positive and the wave is
diverging (+ sign in the complex exponentials). The spherical wave is the product of a plane wave under normal incidence
ρ2
(blue wave-plane) and a correcting term (phase curvature eik∆ , red wavefront) as written in Eq. 1.9. The quantity ∆ = 2z
is the path difference between blue and red wavefronts.

 
iπρ2
Figure 1.4: Plot of the complex amplitude ψ(x, y, z) = C te exp of a spherical wave in paraxial approximation for fixed
λz
√
z. (a) and (c): Real and imaginary parts as a function of ρ (in units of λz). (b) and (d): Grey-level plot of the real and
imaginary parts in the (x, y) plane.

15
The paraxial approximation (Eq. 1.9) is valid if the last term is close to unity, i.e. if
 13
ρ4

|z|  (1.12)
λ

In visible light (λ = 0.5µm) and for ρ=1mm this gives a few centimeters.

Case of a source at a position ~r0 = (x0 , y0 , z0 ): Making the variable change x → x − x0 , y → y − y0 , z → z − z0 , we

obtain
iπ[(x − x0 )2 + (y − y0 )2 ]
 
S0 ±ik|z−z0 |
ψ(~r) ' e exp ± (1.13)
|z − z0 | λ|z − z0 |

1.2 Huyghens-Fresnel principle

Also read : Goodman, “Introduction to Fourier Optics”, chap 4.

1.2.1 Introduction
The Huygens-Fresnel principle describes the way a monochromatic wave is modified when it propagates into space (this
phenomenom is called diffraction). We shall consider a monochromatic wave of amplitude f0 (x0 , y 0 ) in a plane taken as
origin, z = 0. This can be obtained, for example, by placing a screen of transmission t(x0 , y 0 ) (for example a slit or a
diaphragm) in the path of a monochromatic plane wave of complex amplitude A = Aeikz . In this case we simply have

f0 (x0 , y 0 ) = A t(x0 , y 0 ) (1.14)

The observation is made at in a plane a distance z from the plane z = 0, as in the scheme below. Note that we use the
notations (x0 , y 0 ) for coordinates in the plane z = 0 and (x, y) in the observation plane. We suppose that the propagation is
made towards z > 0, which will be assumed for the rest of this course.

In the observation plane at a point P of coordinates (x, y, z), we denote as fz (x, y) the complex amplitude. This notation
emphasises the fact that is is a 2D structure in x and y, and that z is here a parameter. The Huygens-Fresnel principle show
that fz (x, y) expresses as a sum of complex amplitudes produced by all the points of f0 (x0 , y 0 )
ZZ ∞
eikz exp (ik|~r − ~r0 |) 0 0
fz (x, y) = f0 (x0 , y 0 ) dx dy (1.15)
iλ −∞ |~r − ~r0 |

where ~r = (x, y, z) and ~r0 = (x0 , y 0 , 0). This integral is indeed a continuous sum of spherical waves centered at every point ~r0
on the plane z = 0 (Huyghens wavelets).

1.2.2 Paraxial approximation and Fresnel diffraction

When the paraxial approximation is valid, the above integral becomes
ZZ ∞
eikz (x − x0 )2 + (y − y 0 )2
 
fz (x, y) = f0 (x0 , y 0 ) exp iπ dx0 dy 0 (1.16)
iλz −∞ λz
  13
d4
This paraxial form of the Huyghens-Fresnel principle is known as Fresnel diffraction and is valid for z  λ with d the
0 0
size of the diffracting aperture (width of the function f0 (x , y )).

16
Figure 1.5: Geometry for the Fraunhofer diffraction by an aperture in the plane z = 0. x and y are coordinates in the
observation plane at large distance z from aperture plane. d is here the aperture diameter.

The equation 1.16 is a convolution relation, which we can note as:

fz (x, y) = eikz f (x, y) ∗ Dz (x, y) (1.17)
ikz
where the term e expresses the propagation of a plane wave on the distance z, and
ρ2
 
1
Dz (x, y) = exp iπ with ρ2 = x2 + y 2
iλz λz
is the complex amplitude of a spherical wave of center O. This convolution relation expresses fz (x, y) as the sum of spherical
waves produced by point-sources in the plane z = 0.
The function Dz is the “amplitude point-spread function” of the Fresnel diffraction. It is a normalized function:
ZZ ∞
Dz (x, y) dx dy = 1
−∞

1.2.3 Far field: Fraunhofer diffraction

Fraunhofer diffraction (or diffraction at infinity) is the limit of the Fresnel diffraction when the distance z tends towards
infinity (far-field hypothesis). The equation (1.16) writes as:
ZZ ∞
eikz iπz ρ2 ρ02 xx0 + yy 0
  
fz (x, y) = f0 (x0 , y 0 ) exp + − 2 dx0 dy 0 (1.18)
iλz −∞ λ z2 z2 z2
with ρ2 = x2 + y 2 and ρ02 = x02 + y 02 . Since z −→ ∞, we assume ρ  z and ρ0  z and neglect the second order terms in
ρ2 ρ02
z 2 and z 2 in the exponential term. The above equation simplifies in
ZZ ∞
eikz  h x y i
fz (x, y) = f0 (x0 , y 0 ) exp −2iπ x0 + y0 dx0 dy 0 (1.19)
iλz −∞ λz λz
We recognise the expression of a Fourier transform

eikz ˆ  x y 
fz (x, y) = f0 , (1.20)
iλz λz λz
where the symbol ˆ stands for the Fourier transform. We can introduce the quantities α = xz and β = yz , which are angular
coordinates of a point in the observation plane as seen from the origin (see Fig. 1.5). We obtain the agular form of the
Fraunhofer diffraction formula:
eikz ˆ α β
 
fz (α, β) = f0 , (1.21)
iλz λ λ
In the direction α and β, the diffracted amplitude is proportional  to the Fourier transform of the screen. The intensity is
 ˆ α β 2

thus proportional to the power spectrum of f (x, y), i.e. f ( , ) .
λ λ

This kind of calculation is often met in the field of signal processing; it is possible to use the phenomenom of Fraunhofer
diffraction to realise the 2-dimensional Fourier transform. Experimentally, Fraunhofer conditions can be obtained by diffrac-
tion on a distance of several meters or tens of meters. Such an optical setup is somewhat cumbersome and needs a very
bright light source (fz is proportional to 1/z). But in next chapter we shall see that the use of a converging lens allow to
observe Fraunhofer diffraction at finite distance.

17
 Figure 1.6: Fraunhofer diffraction by a rectangular slit in the plane z = 0.

Validity of the approximation

2 02
Fraunhofer diffraction is valid if both the terms ρz2 and ρz2 in the exponential term of Eq. 1.18 can be neglected. This
exponential term can be developed as:
ρ02 xx0 + yy 0 iπρ02
 0
iπz ρ2 iπρ2 x x y0 y
        
exp + − 2 = exp . exp . exp −2iπ + (1.22)
λ z2 z2 z2 λz λz λz λz
 02 
The term in ρ02 can be neglected if exp iπρ λz ' 1, thus if ρ02  λz. If the size of the diffracting aperture is d (so that
ρ0 ≤ d in the integral of Eq. 1.18), we obtain a condition on z to apply Fraunhofer diffraction:

d2
z (1.23)
λ
In the visible domain (λ = 500nm), this gives z > 2m for d = 1mm, and z > 200m for d = 1cm.

1.2.4 Example of Fraunhofer diffraction patterns

Rectangular slit
The object is a rectangular slit of width a in the x direction and b in the y direction (see Fig. 1.6). Its transmission can be
written as x y
t(x, y) = Π Π (1.24)
a b
We suppose that this slit is lit by a plane wave under normal incidence whose amplitude is A in the plane of the slit. Applying
Eq. 1.20, the diffracted amplitude at distance z under Fraunhofer approximation is
 
ikz ab
 πax  πby
fz (x, y) = A e sinc sinc (1.25)
iλz λz λz
The graph of the corresponding intensity is displayed in Fig. 1.7. It exhibits, in both directions, a central lobe surrounded
by secondary maxima. The brightest secondary maximum has an intensity of 4.5% of the maximum. The half-size of the
central lobe in the x (resp y) direction is λz/a (resp. λz/b), inversely proportional to the slit width. The position of the
secondary minima (in the x direction) is periodically distributed at pλz/a (p 6= 0).

Circular diaphragm
The object is a circular aperture of diameter a. Its transmission is
ρ p
t(x, y) = Π with ρ = x2 + y 2 (1.26)
a

18
Figure 1.7: Intensity diffracted by a rectangular slit in Fraunhofer approximation. (a) Plot of the normalised intensity as
a function of x. (b) Perspective plot of the 2D intensity in the (x, y) plane. (c) Gray level plot showing the aspect of the
image.

19
Figure 1.8: Airy disc: intensity diffracted by a circular aperture. (a) Plot of the normalised intensity as a function of x. (b)
Perspective plot of the 2D intensity in the (x, y) plane. (c) Gray level plot showing the aspect of the image.

As for the previous example, the incident wave is plane with normal incidence and amplitude A in the aperture plane. Note
that the wave in the plane z = 0 is invariant by rotation around the z-axis: its diffraction pattern has the same symmetry.
The diffracted amplitude at distance z is

S  πaρ  πa2
fz (x, y) = A eikz 2 jinc with S = (aperture surface) (1.27)
iλz λz 4
which looks very similar to the amplitude diffrated by a slit of same width a, excepted that the sinc function is replaced by
a jinc (preceeded by a multiplicative factor 2 because jinc(0)= 21 ). The intensity is the well-known Airy function:

|A|2 S 2  πaρ 
I(x, y) = 4 jinc2 (1.28)
λz λz
its graph is displayed in Fig. 1.8. It has the appearence of a bright disc surrounded by faint rings (Airy disc). The radius
of the central lobe is 1.22 λz/a. The first secondary ring has an intensity of 1.7% of the maximum. Note that radii of dark
rings is not periodic (as it was the case for the slit).
It it interesting to compare the diffraction patterns (intensity) produced by a disc of diameter a and a square of same side a
(Fig. 1.9). Some differences may be noticed:
• the size of the central lobe is larger in the case of the disc (by 22%),
• the central lobe contains more energy in the case of the disc (84% of the integrated intensity for the disc, 81% for the
square),

• secondary maxima are fainter for the disc (1st maximum at 1.7% of the central intensity for the disc, 4.5% for the
square),
• secondary minima are periodically distributed for the square, not for the disc.

20
Figure 1.9: Comparison between Fraunhofer diffraction patterns of a disc of diameter a and a square aperture of same side
a. Left: graph of the intensities (normalised so that the maximum is 1). Right: same in semi-logarithmic scale.

21
Chapter 2

Fourier properties of converging lenses

Also read:
• Goodman, J.W., “Introduction to Fourier Optics”, chap 5

2.1 Phase screens

2.1.1 Transmission coefficient of a thin phase screen

It can be considered that transparent objects with variable thickness e(x, y) (e.g. prisms,
lenses, glass plates) and/or variable refraction index n(x, y) (for example a layer of gas
with inhomogeneous temperature) act as screens introducing a phase shift. The associated
transmission coefficient t(x, y) is a complex phase term which takes the form :

t(x, y) = eik∆ exp [ik(n − 1) e(x, y)] (2.1)

with k the wave number of the incoming light and ∆ the maximal thickness of the plate.
Paraxial approximation is assumed here so that we consider that the traject in the material
is e(x, y) whatever the incidence of the wave, neglecting effects of inclinations. This relation
is true only if the phase screens are thin.

Proof: The plate is enclosed between two parallel planes (see graph above) Σ (corresponding to z = 0) and Σ0 (at z = ∆).
We consider a plane incident wave with normal incidence. Its complex amplitude at the entrance plane Σ is a constant

f0 (x, y) = A

In the exit plane z = ∆, the light ray crossing the point (x, y) has travelled a distance e(x, y) in the material1 , and ∆ − e(x, y)
in the vacuum, so that the complex amplitude is

f∆ (x, y) = A eik n e(x,y) eik (∆−e(x,y))

which is the product of two terms:

f∆ (x, y) = f0 (x, y) eik∆ exp[ik(n − 1)e(x, y)] (2.2)

the influence of the plate is a multiplicative term t(x, y) = eik∆ exp[ik(n − 1)e(x, y)].

We generally neglect the constant term eik∆ and write the transmission coefficient of the plate as

t(x, y) = exp [ik(n − 1)e(x, y)] (2.3)

1 assuming that the ray goes almost parallel to the z axis so that it hits the planes Σ and Σ0 at the same transverse coordinates (x, y): this

approximation is possible only for thin plates.

22
Figure 2.1: (a): glass prism of refraction index n. The thickness function is e(x, y) = Ax with A  1. (b): illustration of
the deviation of a plane wave with normal incidence by the prism. The new incidence is α = (n − 1)A.

2.1.2 Example: prism

We consider the glass prism of Fig. 2.1a with a thickness e(x, y) = Ax and A  1. The transmission of this prism is

t(x, y) = exp[ik(n − 1)Ax] (2.4)

Note that we have neglected the edges of the prism, considering that is has an infinite extension in the plane (x, y). The
above transmission coefficient is indeed to be multiplied by an adequate rectangular function. If the prism is lit by a plane
wave under normal incidence and complex amplitude ψ0 at the entrance plane of the prism (taken as z = 0), then the exit
amplitude (in the plane z = 0)2 is
f (x, y) = ψ0 eik(n−1)Ax (2.5)
which is,
 in paraxial approximation, the amplitude of a plane wave (linear phase in x, see Section 1.1.2) of wave vector
 α = (n − 1)A
~k = 2π  β = 0

λ  the coefficient γ does not appear explicitely here since the amplitude is written in the plane z = 0.
 γ
It can be calculated using the relation α2 + β 2 + γ 2 = 1. Note that the tilt angle of this wave is (n − 1)A, a result well-known
in geometric optics (see Fig. 2.1b).

2.1.3 Converging lens

We consider a plano-convex lens of refraction index n and radius of curvature R (Fig. 2.2a). The curved surface is a portion
of sphere which can be approximated by a paraboloid if the lens is thin (e.g. R  ∆). In this case the thickness function
expresses as
ρ2
e(x, y) ' ∆ − (2.6)
2R
with ρ2 = x2 + y 2 . The transmission coefficient (neglecting the constant term in ∆) is

ρ2
 
t(x, y) = exp −ik(n − 1) (2.7)
2R
let’s introduce the focal length F of the lens
R
F = (2.8)
n−1
the transmission of the lens is a quadratic phase term which will now be denoted using the notation LF (x, y):

iπρ2
 
LF (x, y) = exp − (2.9)
λF

As for the example of the prism above, we have neglected the edges of the lens, considering an infinite extension if the (x, y)
plane. The complete transmission coefficient is obtained by multiplying LF (x, y) by a a 2D circular function.
2 the exit amplitude should be written in the exit plane z = ∆ with ∆ the maximum thickness of the prism. However we neglected the term

eik∆ in the transmission coefficient: that is similar to write the exit amplitude back in the plane z = 0.

23
Figure 2.2: (a): Plano-convex lens of refraction index n. The exit surface is a portion of sphere of radius of curvature R and
maximum thickness ∆. (b): Transformation of an incident plane wave into a converging spherical wave at the exit of the
lens. The center of this converging wave is at z = F (focal length of the lens).

Figure 2.3: Optical scheme for the observation of the 2D Fourier transform of the transmission of an objet using a converging
lens. (a): Object in the same plane as the lens. (b): Objet at the front focal plane of the lens.

It the lens is lit by a plane wave under normal incidence and complex amplitude ψ0 at the entrance plane of the lens (plane
z = 0), then the exit amplitude (in the plane z = 0+ ) is
iπρ2
f0+ (x, y) = ψ0 e− λF (2.10)

the quadratic phase term in x and y is the signature of a spherical wave under paraxial approximation (see Section 1.1.3)
of center (0, 0, F ), and the minus sign in the exponential shows that this is a converging wave. This is illustrated by the
figure 2.2b. The plane z = F is called back focal plane (or just “focal plane”) of the lens.

Case of a diverging lens: a simular calculation shows that the transmission of a diverging lens of focal F is obtained by
changing F into −F in the equation 2.9, so that it can be denoted as “L−F (x, y)”:

iπρ2
 
L−F (x, y) = exp + (2.11)
λF

It it thus possible to define a generalized focal length, positive for a converging lens and negative for a diverging lens, and
use Eq. 2.9 for both cases.

2.2 Converging lenses and Fourier transform

In this section we consider an ensemble formed by an object of transmission t(x, y) and a converging lens. This ensemble is
lit under normal incidence by a plane wave. The aim is to write down the complex amplitude at the focal plane of the lens.
Two cases will be considered:

• the object is attached to the lens (they are in the same plane),
• the object is in the front focal plane of the lens (distance z = −F from the lens)
The figure 2.3 shows the optical configuration for the two cases.

24
2.2.1 Object in the same plane as the lens
We use the following conventions:
• the plane of the objet/lens is at z = 0
• coordinates in the plane z = 0 are denoted (x0 , y 0 ), and (x, y) in the focal plane z = F

• the complex amplitude of the incident wave (for z < 0) is ψ0 eikz (normal incidence, ~k k ẑ)
The optical configuration is shown in Fig. 2.3a. To calculate the complex amplitude in the plane z = F , we have to write
the amplitude just after the lens (z = 0+ ), then do a Fresnel diffraction to the plane z = F .
The complex amplitude just before the lens (z = 0− ) is the product

f0 (x0 , y 0 ) = ψ0 t(x0 , y 0 ) (2.12)

After the lens, it becomes

f0+ (x0 , y 0 ) = f0 (x0 , y 0 ) LF (x0 , y 0 ) (2.13)
The Fresnel diffraction to the plane z = F is given by Eq 1.17:

fF (x, y) = eikF f0+ (x, y) ∗ DF (x, y) (2.14)

iπρ2
1
with DF (x, y) = iλF e λF . By expanding the convolution product, it can be put into the following form (known as Fresnel
transform):
eikF iπρ2
 
iπρ02
fF (x, y) = e λF F f0+ (x0 , y 0 )e λF (2.15)
iλF u= x y
,v= λF
λF

iπρ02
with ρ02 = x02 + y 02 and F the Fourier transform. The curvature phase term e λF in the brackets simplifies with the
coefficient LF (x0 , y 0 ) so that the complex amplitude in the focal plane is

eikF iπρ2 ˆ  x y 
fF (x, y) = e λF f0 , (2.16)
iλF λF λF

Hence, in the focal plane the amplitude is proportional to the Fourier transform of the amplitude in the plane z = 0. The
intensity is
1   x y 2
I(x, y) = 2 2 fˆ0 , (2.17)
λ F λF λF


As for a Fraunhofer diffraction (see Section 1.2.3), it is possible to observe the 2D power spectrum |fˆ0 (u, v)|2 of the complex
amplitude at the entrance of the lens. Note that this observation is valid only in the focal plane of the lens (otherwise a
Fresnel diffraction between the focal plane and the observation plane applies).
An astrophysical application of this Fourier property was made by Labeyrie in 1970 in his historical paper about speckle
interferometry (Labeyrie A., 1970, “Attainment of Diffraction Limited Resolution in Large Telescopes by Fourier Analysing
Speckle Patterns in Star Images”, A&A 6, 85). Labeyrie needed to accumulate power spectra of short exposure photographs
of stars, and he did it using an optical bench (a laser beam passing trough the photographic film, a lens and a photographic
plate recording the power spectrum).
In the absence of object (t(x0 , y 0 ) = 1) the diffracted amplitude is proportional to a Dirac function δ(x, y), which explicits
the focusing effect of a converging lens on the axis.
iπρ2
The multiplicative phase curvature e λF in Eq. 2.16 may be cancelled by adding a second converging lens of focal F in the
focal plane. But we shall see hereafter that this phase term vanishes if the object is placed at the distance F if front of the
lens.

2.2.2 Object at the front focal plane

The optical scheme is shown in Fig. 2.3b. The object is at a distance F in front of the lens (front focal plane). To calculate the
amplitude at the back focal plane z = 2F we use a plane by plane approach, and successively calculate complex amplitudes
in the 3 planes of interest (object, lens, focal).
Object plane z = 0: the complex amplitude just after the object is denoted as f0 (x0 , y 0 ) as in the paragraph before.
Lens plane, just before the lens (z = F − ): we apply a Fresnel diffraction (Eq 1.17) between the planes z = 0 and z = F :

fF (x, y) = eikF f0 (x, y) ∗ DF (x, y) (2.18)

25
Focal plane, z = 2F : we can apply the result of the paragraph above (Eq. 2.16), the amplitude fF (x, y) plays here the
same role as f0 in Eq. 2.16:
eikF iπρ2 ˆ  x y 
f2F (x, y) = e λF fF , (2.19)
iλF λF λF
using the expression of fF (x, y) we have

eikF iπρ2  ikF 

f2F (x, y) = e λF F e f0 (x, y) ∗ DF (x, y) u= x ,v= y (2.20)
iλF λF λF

which gives
e2ikF iπρ2 ˆ  x y   x y 
f2F (x, y) = e λF f0 , . D̂F , (2.21)
iλF λF λF λF λF
iπρ2
The Fourier transform of DF is D̂F (u, v) = exp[−iπλF (u2 + v 2 )]. It simplifies with the curvature phase term e λF ,
and we finally obtain
e2ikF ˆ  x y 
f2F (x, y) = f0 , (2.22)
iλF λF λF
iπρ2
The phase curvature e λF has now been cancelled and we obtain the exact Fourier transform of the amplitude f0 of the
1
object, scaled by a factor λF in both directions x and y. The relation between f0 and f2F will be denoted as “optical
Fourier transform”.
It can be shown that these results remain valid whatever the position of the object before the lens, providing that the lens
is large enough to collect all the light diffracted by the object. Moving the object along the z axis changes only the phase
curvature term in Eq. 2.16 but the intensity remains the same.
In the case where the object if placed after the lens, in the converging light beam, one still observes its 2D Fourier transform
but there is a scale factor depending on the distance between the lens and the object. We then have a “zoom” effect on the
power spectrum by varying the distance.

26
Chapter 3

Coherent optical filtering

Also read:
• Goodman, J.W., “Introduction to Fourier Optics”, chap 8

3.1 Principle
In electronics, or signal processing, the frequency filtering is the operation which consists to multiply the frequency spectrum
of a given temporal signal by a function called “filter”. The result is a convolution of the signal by an impulse response,
which is the inverse Fourier transform of the filter. In optics, this operation is made on 2D functions, and is a filtering of
spatial frequencies.
It is very easy to perform:
we showed in section 2.2.2 that a converging lens of focal F forms, in its focal plane, the 2D
x y
Fourier transform t̂ λF , λF of an object of transmission t(x, y). It is therefore possible to multiply this Fourier transform
by placing objects in the focal plane of the lens. We will denote this focal plane as “filtering plane”, since the filtering is
performed in this particular plane. Objects such as slits or diaphragms will act on the modulus of t̂. Transparent objects
with a given index of refraction and thickness will act on the phase of t̂.
To observe the result of the filtering in the direct plane, one needs to perform an inverse Fourier transform. In signal
processing, this is done by a dedicated software. In optics, one can take advantage of the following remarkable Fourier
property:
F
fˆ(u, v) −→ f (−x, −y) (3.1)
hence, a direct Fourier transform is similar to an inverse Fourier transform, with a change of the sign of the variables x and
y (resulting in a 180◦ rotation of the function f ). This property allows to observe the result of an optical filtering by placing
a second lens after the filtering plane: in the focal plane of this second lens one observes the 2D inverse Fourier transform of
the filtered spectrum t̂, i.e. the filtered object, rotated by 180◦ .
This two-lens system is also known as “double diffraction setup” (see Fig. 3.1), each lens performing an optical Fourier
transform (analogous to a Fraunhofer diffraction). Note that the two lenses may have different focal lengths, but the filtering
plane has to be at the front focal plane of the second lens (conjugated lenses) in order to avoid unwanted phase curvature
terms in the optical Fourier transforms.

3.2 Abbe-Porter experiments

Pioneering experiments were made in the early 1900’s and give a spectacular illustration of image formation and Fourier
optics fundamentals. They consist in simple filtering of spatial frequencies, as described in the previous section. The principle
is the following: a fine 2D metallic grid is lit by a laser beam (Fig. 3.2). This object is placed a the front focal plane of the
first lens (plane P0 ). In the filtering plane (P1 ) one observes the Fourier transform of the object. It is composed of bright
spots (since the object is periodic), every spot being an Airy disc if the grid is limited by a circular diaphragm. In the focal
plane of the second lens (observation plane, P2 ), one observes a replica of the object, convoluted by the PSF of the filter (see
section 3.3 for details). If no filter is placed in the filtering plane, the image is identical to the object (rotated by 180◦ ).
In the filtering plane, bright spots along the horizontal (resp. vertical) axis correspond to horizontal (resp. vertical) spatial
frequencies composing the object. We suppose that the filter is here a slit which selects one row of bright spots. If the slit is
vertical (Fig. 3.3a) it selects only the vertical frequencies of the object: the image in the plane P2 is a grid with horizontal
parallel strips. If the slit is horizontal, (Fig. 3.3b) horizontal frequencies are selected and the image is composed of vertical
parallel strips.

27
Figure 3.1: Double diffraction setup for optical filtering. The object is placed at the front focal plane of the first lens of focal
F1 and lit by a plane wave. The second lens (focal F2 ) is placed at a distance z = F1 + F2 from the first one, so that the
filtering plane is at the front focal plane of the second lens. The observation is made in the back focal plane of the second
lens.

Figure 3.2: Illustration of the Abbe-Porter experiment. A 2D grid is placed in the front object plane P0 of the first lens. In
the focal plane P1 , one finds the spectrum of the grid, composed of a 2D arrangement of bright spots. If no filter is placed
in P1 , then the second lens forms; at its focal plane P2 , a replica of the object, rotated by 180◦ (Fourier transform of its
spectrum).

28
Figure 3.3: Abbe-Porter experiment similar to Fig. 3.2, with a slit in the filtering plane P1 . (a) vertical slit, selecting the
central column of bright spots. The corresponding image in the plane P2 is a series of horizontal strips, Fourier Transform
of the truncated spectrum. (b) horizontal slit, producing in P2 a set of vertical strips.

3.3 Object-image relation

We consider the double diffraction experiment of Fig. 3.1. The aim of this section is to explicit the relations between the
complex amplitudes of the object (plane P0 ) and the image in the plane P2 . We denote as
• ψ0 the complex amplitude of the incident plane wave in the plane P0 (normal incidence is assumed)
• F1 and F2 the focal length of the two lenses
• (x0 , y 0 ), (x1 , y1 ) and (x, y) the coordinates of a point in the planes P0 (object plane), P1 (filtering plane) and P2
(observation plane)
• t(x0 , y 0 ) the transmission coefficient of the object (plane P0 )
• P (x1 , y1 ) the transmission coefficient of the filter (plane P1 )
Let’s calculate step by step the propagation from planes P0 to P2 : at the output the plane P0 , the complex amplitude is
simply proportional to the transmission of the object:

f0 (x0 , y 0 ) = ψ0 t(x0 , y 0 ) (3.2)

We can apply Eq. 2.22 to write the amplitude in the plane P1 , taking advantage of the Fourier property of the converging
lens:
e2ikF1 ˆ
 
x1 y1
f1 (x1 , y1 ) = f0 , (3.3)
iλF1 λF1 λF1
In the plane P1 , the filtering operation is performed by multiplying f1 (x1 , y1 ) by the transmission of the filter:

e2ikF1 ˆ
 
x1 y1
f1+ (x1 , y1 ) = f0 , P (x1 , y1 ) (3.4)
iλF1 λF1 λF1
And to obtain the amplitude in the plane P2 , we make use, once again, of Eq. 2.22:
e2ikF2 ˆ
 
x y
f2 (x, y) = f1+ , (3.5)
iλF2 λF2 λF2

29
with
e2ikF1
     
x y x1 y1
fˆ1+ , = ˆ
F f0 , P (x1 , y1 ) (3.6)
λF2 λF2 iλF1 λF1 λF1 u= x y
,v= λF
λF2 2

We eventually obtain a convolution relation:

 
1  x y 1 x y
f2 (x, y) = − e2ik(F1 +F2 ) f0 − , − ∗ P̂ , (3.7)
G G G (λF2 )2 λF2 λF2

F2
where G = F1 is a magnification factor. The amplitude of the final image in the plane P2 is thus a convolution between
 
x y ◦ 1 x y


f0 − G , − G (the object magnified by a factor G and rotated 180 ) and the point-spread function (λF2 )2 P̂ λF2 , λF2 . The
magnification applies before the convolution by the impulse response. The case of identical lenses (F1 = F2 ) corresponds to
G = 1: no magnification, the object and image have the same size.

3.4 Low-pass and high-pass filters

In signal processing, low-pass or high-pass filters attenuate low or high frequencies. The spectrum of the signal is multiplied
by a function (denoted as transfer function) which vanish for high (resp. low) frequencies in the case of a low-pass (resp.
high-pass) filter.
In optics, with a double diffraction experiment, the product between the spectrum of the object and a transfer function
appears in Eq. 3.4. For a sake of simplicity, we shall consider the case of two identical lenses (F1 = F2 = F ). Equation 3.4
x1 y1
can be rewritten using the variable change u = λF , v = λF :

e2ikF1 ˆ
f1+ (λF u, λF v) = f0 (u, v) P (λF u, λF v) (3.8)
iλF1

The right-hand side of this equation lets appear the spectrum fˆ0 (u, v) of the object, and the product by the following transfer
function:
h(u, v) = P (λF u, λF v) (3.9)
1
which is simply a replica of the pupil function, magnified by a factor λF .

3.4.1 Low-pass filters

A low-pass filter has a transfer function which vanish for high frequencies. A circular diaphragm of diameter d, centered on
the optical axis in the filtering plane, is an example of low-pass filter. The corresponding transfer function is
Y  λF q 
h(u, v) = with q 2 = u2 + v 2 (3.10)
d
d
This transfer function is isotropic, and vanishes for spatial frequencies q > fc with fc = 2λF the cutoff frequency.
Let-s consider the case where the object is a sinusoı̈dal grid of spatial frequency m whose transmission (between 0 and 1) is
1 1
t(x, y) = + cos(2πmx) (3.11)
2 2
its Fourier transform is composed of 3 Dirac peaks centered at frequencies (0, 0), (m, 0) and (−m, 0)
1 1 1
t̂(u, v) = δ(u, v) + δ(u − m, v) + δ(u + m, v) (3.12)
2 4 4
Multiplying this spectrum by the transfer function of Eq. 3.10 will cut the lateral peaks if m > fc . In this case, the image
in the plane P2 will be uniform. This is illustrated by Fig. 3.4. For a more general object t(x, y) having a continuous
frequency spectrum t̂(u, v), a low-pass filter will remove high frequencies, resulting into a fuzzy image with a loss of details,
as illustrated by Fig. 3.5.

30
Figure 3.4: Low-pass filtering of a sinusoı̈dal grid of spatial frequency m by a circular transfer function of cutoff frequency
fc < m (see § 3.4.1). (a) grayscale plot of the transmission of the object. (b) Perspective plot showing the transfer function
(Eq. 3.10) and the 3 peaks of the Fourier transform of the object (Eq. 3.12). Since m > fc the two lateral peaks are cut by
the transfer function. (c) Resulting image (uniform) in the observation plane P2 .

Figure 3.5: Low-pass filtering of an object (Mickey shape) for 3 different filters. (a) Object. (b) Spectrum of the object
(modulus). (c) Spectrum of the object (modulus) multiplied by the transfer function of the filter. (d) Resulting image in
the observation plane (without the 180◦ rotation). 1st row: the filter is a circular pupil. 2nd row: the filter is also a circular
pupil, but with lower diameter (resulting in a more fuzzy resulting image). 3rd row: the filter is an horizontal slit; it is still
a low-pass filter, but in the vertical direction only.

31
Figure 3.6: High-pass filtering of an object (Mickey shape). (a) Object. (b) Spectrum of the object (modulus). (c) Spectrum
of the object (modulus) multiplied by the transfer function of the filter. (d) Resulting image in the observation plane
(without the 180◦ rotation). The filter is here a circular occulter having the same diameter as the circular hole of the 1st
row of Fig. 3.5.

3.4.2 High-pass filters

A high-pass filter has a transfer function which vanish for low frequencies. An example is a on-axis circular occulter of placed
in the filtering plane P1 which blocks the light up to a certain distance from the center. This kind of filter tends to reinforce
high spatial frequencies in the image and to emphasize fine details – exactly the opposite of the low-pass filter.
Fig. 3.6 shows the effet of a high-pass filtering by a circular occulter on a “Mickey” object similar to the figure 3.5. The
filtered image displays only the contours of Mickey, everything else has been removed by the filter. This reinforcement of
the contours is easy to understand if one remarks that the filter (circular occulter) is the complementary screen of a circular
diaphragm. Its transmission expresses as
Po (x, y) = 1 − P (x, y) (3.13)
ρ


with P (x, y) = Π d the transmission of the circular diaphragm. Therefore the point-spread function of the filtering
(appearing in Eq. 3.7) is  x y   x y 
P̂o , = (λF )2 δ(x, y) − P̂ , (3.14)
λF λF λF λF
and the amplitude in the observing plane, given by Eq. 3.7, expresses as the difference of two terms
 
 1  x y 
f2 (x, y) = −e4ikF  f0 (−x, −y) − f0 (−x, −y) ∗ P̂ , (3.15)
 
(λF )2 λF λF 

 | {z }
amplitude of the unfiltered object | {z }
low-pass filtered object

The first term is the amplitude of the unfiltered object, i.e. the “Mickey” object (Fig. 3.5, column a). The second term is
the amplitude of object filtered by the circular diaphragm, it is a smoothed version of “Mickey” (Fig. 3.5, 1st row, column
d). The first term has sharpest countours than the second, this is why their difference let only appear only the countours of
the object.
Other spectacular illustrations of optical filtering of 2D objects can be found in the book “Atlas of Optical Transforms” by
Harburn.

3.5 Strioscopy and phase contrast

Optical filtering techniques can be applied to phase objects, i.e. transparent objects with variable thickness or refraction
index (for example a fingerprint on a microscope plate). Phase variations of the object can be made visible by placing an
appropriate mask in the filtering plane of a double-diffraction setup.
For the whole paragraph, we consider the optical setup of Fig. 3.1, with identical lenses (F1 = F2 = F ). The object in the
plane P0 is a pure phase object of transmission
t(x, y) = eiφ(x,y) (3.16)
lit by an incident plane wave of amplitude ψ0 . With the condition φ(x, y)  1 (small phase variations), the transmission
can be approximated to the first-order development

t(x, y) ' 1 + iφ(x, y) (3.17)

32
Figure 3.7: Optical Fourier transform of a phase object with small phase variations. The complex amplitude in the focal
plane of the lens (P1 ) shows a bright central spot surrounded by a halo of light diffracted by phase variations of the object
(see Eq. 3.18).

The complex amplitude in the plane P1 , given by Eq. 3.3, is composed of two terms:
 
e2ikF 
−i(λF )2 δ(x1 , y1 )
x
1 y1  
f1 (x1 , y1 ) = ψ0 + φ̂ ,  (3.18)
λF | {z } | λF{z λF }

bright central spot
light diffracted by phase variations

There is a bright spot at the origin, surrounded by a halo of light diffracted by the phase variations φ(x, y) of the object.
This halo of light is faint because φ(x, y)  1, and disappears if φ = 0, i.e. is there is no object in the plane P0 (in this case
the bright central spot is simply the light of the incident plane wave, focused by the lens L1 ). This is illustrated by Fig. 3.7.
If no filter is placed in the plane P1 , one observes, in the image plane P2 , a complex amplitude

f2 (x, y) = e4ikF ψ0 t(−x, −y) (3.19)

and the intensity is uniform since |t(−x, −y)|2 = 1. This means that the phase variations of the object are invisible on the
image. To make them visible, several kinds of filters can be placed in the plane P1 . We shall study two of them in the
following: an amplitude filter (strioscopy technique) and a phase filter (phase-contrast technique).

3.5.1 Strioscopy
In the technique of strioscopy, the filter is a circular occulter of very small diameter (like a pinhead), centered on the optical
axis. It is a high-pass filter which blocks only the spatial frequency (0, 0). The optical setup is shown in Fig. 3.8.
The action of the occulter is to cancel the Dirac peak appearing in the complex amplitude in the plane P1 (Eq. 3.18). At
the output of the filter, the complex amplitude becomes

e2ikF  x1 y1 
f1+ (x1 , y1 ) ' ψ0φ̂ , (3.20)
λF λF λF
x1 y1


Note that the occulter cancels also a part of the term φ̂ λF , λF near the origin, but since we made the hypothesis that the
occulter is very small, this effect will be neglected. The complex amplitude in the observation plane P2 is

f2 (x, y) = −i e4ikF ψ0 φ(−x, −y) (3.21)

and the intensity is

I2 (x, y) = |ψ0 |2 |φ(−x, −y)|2 (3.22)
it is proportional to the square of the phase variations of the object. Therefore the object phase is made visible by this
strioscopy technique. Similar results can be obtained by replacing the occulter in the plane P1 by a half plane (Foucault
knife-edge technique). Spectacular videos can be found on the web, see for example
https://www.futura-sciences.com/sciences/videos/voir-invisible-strioscopie-822/.
However this technique has the disavantage that the sign of the phase is lost (i.e. it is impossible to know if a given structure
in the image corresponds to a bump or a hole on the object). Also, since the phase variations φ(x, y) are small, images are
very faint. The alternative technique of “phase contrast”, presented hereafter, provides an answer to these problems.

33
Figure 3.8: Strioscopy technique applied to a phase object placed in the plane P0 . In the Fourier plane (P1 ), there is a
central bright spot (see Eq. 3.18) which is cancelled by a small occulter. In the image plane P2 , the intensity is proportional
to the square of the phase variations of the object.

3.5.2 Phase contrast

The technique of phase contrast was proposed by Zernike in the early 30s (Zernike F., 1934, MNRAS 94, 377). It appeared
to have many applications in microscopy to study quasi-transparent living cells. Zernike was awarded the Nobel prize in 1953
for the phase-constrast micropsopy technique. Phase constrast is similar to strioscopy: the difference is that the occulter is
no more opaque, but is a small parallel plate producing a π2 phase shift to the on-axis light. The complex amplitude in the
plane P1 at the output of the plate becomes

e2ikF h x y1 i
1
f1+ (x1 , y1 ) = ψ0 (λF )2 δ(x1 , y1 ) + φ̂ , (3.23)
λF λF λF
and in the observation plane
f2 (x, y) = −i e4ikF ψ0 [1 + φ(−x, −y)] (3.24)
The corresponding intensity is (under the hypothesis φ(x, y)  1)

I2 (x, y) ' |ψ0 |2 [1 + 2φ(−x, −y)] (3.25)

it is now an affine function of the phase φ(x, y), composed of a constant term (uniform light background) and intensity
variations proportionnal to the phase of the object, with its sign: brightest (resp. darkest) zones in the image correspond to
positive (resp. negative) phase value.

34
Chapter 4

Image formation

Also read : Goodman J.W., ‘Introduction to Fourier Optics”, chap. 6

The aim of this chapter is to derive the object-image relation between intensities of an astronomical object and its image
at the focus of a telescope. The object is supposed to be incoherent and at large distance from the telescope. The optical
scheme is shown hereafter:

We use the following conventions:

Plane :
1 Astronomical object (distance from the telescope is D, can be several parcsecs)
Plane :
2 Converging lens (focal F ) + pupil function P (x, y) (in most cases a uniform disc)
Plane :
3 Image plane = focal plane of the telescope.
Principle of the calculation:
• Place a point-source at position (x0 , y0 ) in plane .
1
• Calculate the corresponding intensity in plane ,
3 using Fourier properties of lenses
• Sum on (x0 , y0 ) to obtain the intensity produced by any object in plane .
1

4.1 Object-image convolution relation

4.1.1 Image created by a point-source
In plane
1

Coordinates the plane 1 are denoted as (x0 , y 0 ). A

point-source is placed at position (x0 , y0 ): its com-
plex amplitude is f1 (x0 , y 0 ) = a0 δ(x0 − x0 , y 0 − y0 )
where a0 is a constant. We change to the following
angular variables
" 0
α0 = xD α0 = xD0

0 (4.1)
β 0 = yD β0 = yD0

35
so that the complex amplitude writes as
a0
f1 (x0 , y 0 ) = a0 δ(x0 − x0 , y 0 − y0 ) = δ(α0 − α0 , β 0 − β0 ) = f0 δ(α0 − α0 , β 0 − β0 ). (4.2)
D2
The corresponding intensity is:
I0 δ(α0 − α0 , β 0 − β0 ) (4.3)
with I0 = |f0 |2

In plane
2
Since D is very large (astronomical distance), Fraunhofer diffraction is assumed between planes
1 and .
2 Therefore the
complex amplitude in plane ,2 just before the telescope pupil, is

eikD ˆ  x y 
f2 (x, y) = f1 , (4.4)
iλD λD λD
with f1 (x0 , y 0 ) = a0 δ(x0 − x0 , y 0 − y0 ) the complex amplitude of the point-source as a function of linear coordinates x0 and y 0
(not angular, to apply the formula for Fraunhofer diffraction). The calculation gives

eikD −2iπ
f2 (x, y) = a0 e λD (xx0 +yy0 )
iλD
↓ change variable α0 = x0 /D, β0 = y0 /D
eikD −2iπ
= a0 e λ (α0 x+β0 y)
iλD
Df0 −2iπ (α0 x+β0 y)
= eikD e λ
iλ
(4.5)

We recognise the complex amplitude of a plane

wave with angles of incidence (−α0 , −β0 ). This
is expected, since a point-source emits a spherical
wave which becomes plane at large distance.

The complex amplitude just after the pupil is

Df0 −2iπ (α0 x+β0 y)
f2+ (x, y) = eikD e λ P (x, y) (4.6)
iλ

In plane
3
The complex amplitude just before the telescope lens is f2+ (x, y). Therefore the complex amplitude at the focal plane is (see
Section 2.2.2, Eq. 2.16):
eikF iπρ2 ˆ  x
 
y 
f3 (x, y) = exp f2+ , (4.7)
iλF λF λF λF
with ρ2 = x2 + y 2 . We get:

eik(F +D) Df0 iπρ2

  h −2iπ i
f3 (x, y) = − exp F e λ (α0 x+β0 y) P (x, y) (4.8)
λ2 F λF
x y
where the Fourier Transform above (F) is to be taken for variables u = λF and v = λF . Here we have
 
h −2iπ i α0 β0
F e λ (α0 x+β0 y) P (x, y) = P̂ u + ,v + (4.9)
λ λ

36
so
eik(F +D) Df0 iπρ2
   
x α0 y β0
f3 (x, y) = − exp P̂ + , +
λ2 F λF λF λ λF λ
↓ change variable α = x/F , β = y/F
eik(F +D) Df0 iπρ2
   
α + α0 β + β0
= − exp P̂ ,
λ2 F λF λ λ
(4.10)

The variables α and β represent angles in the

image plane as seen from the center of the
pupil.

The intensity is |f3 |2 . Using angular variables (α, β), it expresses as

  2
α + α0 β + β0 
|f3 (α, β)|2 = K I0 P̂

, (4.11)
λ λ 
D2
where K = λ4 F 2 is a multiplicative constant, generally neglected. We introduce
  2
 α β 
S(α, β) = P̂ , (4.12)
λ λ 
so that the intensity is simply I0 S(α+α0 , β +β0 ) (neglecting the constant K). As a conclusion, we can see that a point-source
at position (angle) (α0 , β0 ) in plane
1 produces an image whose intensity is the Fourier Transform (square modulus) of the
pupil function, centered at position (−α0 , −β0 ). This is illustrated by the following scheme:

Point-Spread Function: The quantity S(α, β) is called the Point-Spread Function (PSF). It is indeed the image of a
point-source of unit intensity (I0 = 1) located on the optical axis (α0 = β0 = 0). It expresses as the power spectrum (square
modulus of the Fourier Transform) of the pupil function, as defined in Eq. 4.12

4.1.2 Object-image relation

1 we now consider an incoherent object with intensity distribution I0 (α0 , β 0 ) as a function of position angles defined
In plane
by Eq. 4.1. The intensity I0 can be written as
ZZ ∞
0 0 0 0
I0 (α , β ) = I0 ∗ δ(α , β ) = I0 (α0 , β0 ) δ(α0 − α0 , β 0 − β0 ) dα0 dβ0 (4.13)
−∞ | {z } | {z }
weighting factor unit point-source

37
This integral can be interpreted as a continuous sum of point-sources located at positions (α0 , β0 ) and weighted by the
factor I0 (α0 , β0 ). Ech of these point-sources produces in the place
3 a PSF S(α + α0 , β + β0 ), weighted by the same factor,
as demonstrated in previous section. Thus the total intensity observed at plane 3 is the continuous sum of intensities1
produced by each point-source, i.e.

ZZ ∞
I(α, β) = I0 (α0 , β0 )S(α + α0 , β + β0 ) dα0 dβ0
−∞
↓ change variable α0 = −α0 , β 0 = −β0
ZZ ∞
= I0 (−α0 , −β 0 )S(α − α0 , β − β 0 ) dα0 dβ 0
−∞
(4.14)

And we obtain the well-known object-image convolution relation between intensities, expressed with angular variables:

I(α, β) = I0 (−α, −β) ∗ S(α, β) (4.15)

The function I0 (−α, −β) represents the object rotated by 180◦ . This relation is valid only for incoherent sources (a similar
relation exists between amplitudes for coherent light, see for example the chapter 5 of Goodman, Introduction to Fourier
Optics).

Geometric limit, infinite pupil: this is the case where P (x, y) = 1. The lens in plane
2 has no spatial limitation. The
PSF is then a Dirac impulse:
S(α, β) = δ(α, β) (4.16)
so the image of a point-source is a single point (perfect stigmatism): this is the case of geometric optics. Note that the same
result is obtained if λ −→ 0. The object-image relation becomes

I(α, β) = I0 (−α, −β) (4.17)

so the image is the object itself (rotated by 180◦ ).

4.2 Optical transfer function (OTF)

The convolution relationship in the focal plane (Eq. 4.15) corresponds to a linear filtering in the Fourier plane. If we denote
u and v the angular frequencies associated with α and β, the Fourier transform of Eq. 4.15 becomes
ˆ v) = Iˆ0 (−u, −v) . T (u, v)
I(u, (4.18)

with "  2 #
 α β 
T (u, v) = F[S(α, β)] = F P̂ ,  (4.19)
λ λ 

The function T (u, v) is called “Optical transfer function” (OTF). It is a complex funtion; its modulus and phase are denoted
as “modulation transfer function” and “phase transfer function” respectively. u and v are expressed in rad−1 . To calculate
T (u, v), we use the Wiener-Kinchin theorem, which expresses the Fourier transform of the power spectrum of a function:
Z ∞
F[|fˆ(u)|2 ] = Cf (X) = f (x) f (x + X) dx (4.20)
−∞

The quantity Cf (X) is the Autocorrelation function of f (x). It can be interpreted as the integral of superposition of the
function f (x) and the same function conjugated and translated: f (x + X).

Proof of the Wiener-Kinchin theorem:

 2
• Write fˆ(u) = fˆ(u).fˆ(u)
 

1 To calculate the intensity created by the addition of several waves in incoherent light, one adds intensities of individual waves (not amplitudes).

38
Figure 4.1: Illustration of the calculation of an Optical transfer function in the case of a square pupil of side d. Left:
superposition of the overlapping terms: P (x, y) (centered at (0, 0)) and P (x + λu, y + λv) (centered at (−λu, −λv)) which
intervene in the integral of Eq. 4.26. Middle and right: plot of the OTF for v = 0 and in the (u, v) plane. The OTF is a
pyramid in this case.

• Use the following Fourier properties :

F[fˆ(u)] = f (−X) (4.21)
and h i
F f (x) = fˆ(−u) (4.22)

• The Fourier transform of |fˆ(u)|2 is obtained as the convolution

Cf (X) = f (−X) ∗ f (X) (4.23)

which develops as Z ∞ Z ∞
Cf (X) = f (−x0 ) f (X − x0 ) dx0 = f (x) f (x + X) dx (4.24)
−∞ −∞

To compute the optical transfer function, we make use of the Wiener-Kinchin theorem and the Fourier property
h  x i
F f = |a|fˆ(au)
a
so that "  2 #
α β 
= λ2 CP (λu, λv)

T (u, v) = F P̂
 , (4.25)
λ λ 

and in general we introduce a multiplicative constant to normalise the transfer function so that T (0, 0) = 1. We shall then
define the OTF as ZZ ∞
1 1
T (u, v) = CP (λu, λv) = P (x, y) P (x + λu, y + λv) dx dy (4.26)
S S −∞

where S is the surface of the pupil. This is the integral of superposition of two identical pupils, one being conjugated and
shifted by a quantity (−λu, −λv) as illustrated by Fig. 4.1

Properties of the OTF:

• It is a normalised fonction T (0, 0) = 1

• If the pupil P (x, y) is real, then the OTF T (u, v) is real and even, T (−u, −v) = T (u, v)
• If the pupil is a separable function P (x, y) = P1 (x).P2 (y) then the OTF is also a separable function T (u, v) =
1
CP1 (λu).CP2 (λv)
S
• If the pupil is isotropic (i.e. P (x, y) = P (ρ) with ρ2 = x2 + y 2 ) then the OTF is also isotropic ( T (u, v) = T (q) with
q 2 = u2 + v 2 )

39
Figure 4.2: Illustration of the physical significance of the OTF in the case of an image of a sinusoı̈dal grid. Left: object of
ˆ 0) together with the OTF T (u, 0) (taken as a triangle for this example).
intensity I0 (α, β). Middle: Fourier transform I(u,
Right: Observed image, with a contrast |T (f, 0)|.

Physical significance of the transfer function: To understand the physical significance of the OTF, let’s consider an
incoherent object which is a pure sinusoı̈dal grid of angular frequency f in the α direction, as in Fig. 4.2. The object has an
intensity distribution
1
I0 (α, β) = [1 + cos(2πf α)] (4.27)
2
This object has indeed 3 vector angular frequencies in the (u, v) plane: f~1 = (0, 0), f~2 = (f, 0) and f~3 = (−f, 0). Its Fourier
transform is
1 1 1
Iˆ0 (u, v) = δ(u, v) + δ(u − f, v) + δ(u + f, v) (4.28)
2 4 4
Applying the object-image relation in the Fourier plane (Eq. 4.18), we can derive the F.T. of the intensity distribution of
the image:
ˆ v) = T (0, 0) δ(u, v) + T (f, 0) δ(u − f, v) + T (−f, 0) δ(u + f, v)
I(u, (4.29)
2 4 4
With T (0, 0) = 1 by definition. The lateral terms in I(u, ˆ v) are multiplied by the value of the OTF at (u, v) equals to the
grid frequencies. In the case of an even OTF (corresponding to square or a circular pupil), we have T (f, 0) = T (−f, 0) and
the expression simplifies. It is then easy to obtain the image intensity distribution by inverse F.T.:
1
I(α, β) = [1 + T (f, 0) cos(2πf α)] (4.30)
2
Which is also a sinusoı̈dal grid of same frequency as the object. The difference is that the cosine term has been attenuated.
The contrast of this figure is
Imax − Imin
C= = |T (f, 0)| (4.31)
Imax + Imin
which gives the physical significance of the OTF, in fact of its modulus (the MTF): it is indeed the contrast (also known as
“visibility”) of the image of the grid. Changing the grid frequency would result into another value of the contrast. It is then
possible to estimate the MTF by making images of grids of various frequencies and measure their contrast.

4.3 Case of a circular pupil

4.3.1 PSF and resolving power

The pupil is a uniform disc of diameter d. It expresses as 

ρ
P (x, y) = Π with ρ2 = x2 + y 2 (4.32)  d-
d 
the PSF is an Airy disc of intensity
2 2
πd2
 
πdθ
S(α, β) = 4 jinc (4.33)
4 λ
it is an isotropic function of the angle θ2 = α2 + β 2 . It is independent of the telescope focal. The multiplicative constant
 2 2
πd
4 may be ommited. The PSF is displayed in Fig. 4.3. It exhibits a central peak of angular radius 1.22 λd surrounded
by dark and bright faint rings (the brightest ring has a maximum intensity of 1.7% of the peak intensity of the PSF).

40
 (a) (b) (c)
0
10
1.0

−1
10
Normalised intensity
0.8
0.017
−2
10

0.6 0.004
−3
10

0.4

1.63

2.23

2.68

3.24
1.22
−4
10

0.2

−5
10

0.0
−6
10
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5

θ (units of λ/d) θ (units of λ/d) 2.44 λ/ d

Figure 4.3: Airy disc: intensity PSF for a circular pupil of diameter d (Eq. 4.33). (a) plot of the intensity profile as a function
of position angle θ in the focal plane. Positions of the first zeros and the first secondary maxima are written on the graph (in
units of λ/d). (b) Same plot in semi-log scale. Relative intensities of the first secondary maxima are written on the graph.
(c) Gray-level plot of the Airy disc.

(a) (b) (c)

Intensity produced by two point-sources at Rayleigh limit

1.0
∆ 1st point

2nd point

sum

0.8

0.6 β

0.4

0.2

β α
α (units of λ/d)
0.0 α
−1 0 1 2 3
1.22

Figure 4.4: Image of two point-sources by a circular pupil at Rayleigh limit: star separation ∆ is equal to the PSF radius
= 1.22λ/d. (a) Plot of the intensity as a function of the angle α in the focal plane. (b) Surface plot of the intensity I(α, β).
(c) Gray level plot of the intensity showing the aspect of the image.

Resolving power: it is defined from the image of two point-sources, composed by the sum of two shifted PSF. Images of
each point-source are distinct if their separation is greater than the size of the PSF. An empiric definition was proposed by
Lord Rayleigh in 1879 (Phil. Mag. S 5. Vol. 8, Oct. 1879). It corresponds to the situation where the angular separation ∆
of the two points is exactly equal to the PSF radius, i.e. the radius of the first dark ring of the Airy disk (see Fig 4.4):

λ
∆ = 1.22 [unit: radian] (4.34)
d
Typical values for λ = 500 nm:
• d = 12 cm =⇒ ∆ = 1 arcsec

• d = 2 mm (human eye in daylight) =⇒ ∆ = 1 arcmin

• d = 2.40 m (Hubble Space Telescope) =⇒ ∆ = 0.05 arcsec

4.3.2 Optical transfer function

The OTF T (u, v) is the surface of the intersection of two discs of diameter d, shifted by a vector (λu, λv) as defined by
Eq. 4.26. It is a real, isotropic function. The calculus is made in a geometrical way (see Goodman, “Introduction to Fourier
Optics” for details) and gives a chinese hat function
 
λq
T (u, v) = with q 2 = u2 + v 2 (4.35)
d

41
 λu

d x

T(u,v)

Figure 4.5: Transfer function of a circular pupil. Left: geometric interpretation of T (u, v) as the area of the intersection of
two identical shifted pupils (Eq. 4.26). Middle: 3D plot of the OTF showing the typical “chinese hat” shape. Right: Plot of
T (u, 0), showing the angular cutoff frequency fc = d/λ.

1
chinese hat
triangle
0.8

where
2h p i 0.6
(x) = acos(|x|) − |x| 1 − x2 (4.36)
π
0.4

The function (x) is displayed on the right: it looks like a

0.2
triangle with curved sides. It is an even function, vanishing for
|x| ≥ 1. 0
-1 -0.5 0 0.5 1

The graph of the transfer T (u, v) function is shown in Fig. 4.5. It is an isotropic chinese hat which has non-zero values for
q ≤ λd . It corresponds to a low-pass filtering of angular frequencies of the object.

Cutoff frequency: it is defined as the highest frequency for which the transfer function is not zero. For a circular pupil
of diameter d, the angular cutoff frequency is
d
fc = unit: rad−1 (4.37)
λ
The cutoff frequency is the frequency of the finest sinusoı̈dal grid that the telescope is capable to image with non-zero
contrast. In other words, if one considers an pure sinusoidal object having an intensity distribution I0 (α, β) = A cos(πu0 α)2
of angular frequency u0 , then its image becomes uniform when u0 ≥ d/λ.
It is possible to use fc to define a resolving power in a sense less empirical than the Rayleigh criterion presented before. Let’s
define ∆c = f1c the cutoff period, i.e. the angular period of the finest imageable grid. Indeed, ∆c represents the angular size
of the finest details in the image: this gives another definition of the resolving power for a pupil of diameter d:
λ
∆c = (4.38)
d
It is sometimes called “resolution element” or resel. We see that ∆c is close to the Rayleigh definition ∆ = 1.22 λd . Note
that
• When d % then ∆c & : better resolution for a large telescope

• When λ & then ∆c & : better resolution at short wavelengths

42