Introduction to Fourier optics

Textbook: Goodman (chapters 2-4)

Overview:

Linear and invariant systems

The Fourier transform
Scalar diffraction
Fresnel and Fraunhoffer approximations.
1. Linear systems and Fourier transform tutorial (reminder)

A system connects an input g1 to an output g2 such that:

g 2 x2 , y2   Sg1 x1 , y1 

It is said to be linear if it satisfies:

Sg x, y   hx, y   Sg x, y   Shx, y 

The response of a linear system can be expressed by its response to a (basis) set of
elementary functions. In particular, since

g1 x1 , y1    g1  ,  x   , y   dd

and invoking linearity we find that:

g 2 x2 , y 2    g1  , S  x1   , y1   dd

This defines the impulse response (or point spread function) of the linear system:

hx2 , y2 ,  ,   S x1   , y1   

An important subset of linear systems are invariant systems, in which the impulse
response is only dependent on the difference in coordinates:

hx2 , y2 ,  ,   hx2   , y2   

In this case the system response is simply a convolution between the input and the
transfer function:

g 2  g1  h

This implies that in Frequency domain, the output of an invariant linear system is
merely a multiplication of the input by the transfer function H

G2  G1 H

Where


H   hx, y exp  2i f

x 
x  f y y  dx dy

To further discuss this let us remember some of the properties of the Fourier
transform:
Definition of the 2D Fourier transform


F g    g x, y exp  2i f

x 
x  f y y  dx dy

Similarly, the inverse transform is:


F 1 G    G f x  
, f y exp 2i f x x  f y y  df x df y


The conditions for this being definite are that the function is "regular", which applies
for all physical realizations in the optical realm.

The Fourier transform operation can be viewed as a decomposition of a function in a


new basis of elementary functions of the form exp 2i f x x  f y y  , which are tilted 
plane waves with angle   a tan f y f x  to the X-axis, and a spatial period

L  f x2  f y2 
1 2
.

Some important properties of the FT (quite evident from the definition)

Are:

Linearity:

F g  h  F g   F h

Similarity:

1  fx fy 
F g ax, by   G , 
ab  a b 

Shift theorem (linear shift transform to a phase shift):


F g x  a, y  b  exp  2i f x a  f y b F g x, y  

Parseval Theorem ("conservation of energy"):

 

 g x, y  dxdy    G f x , f y  df x df y
2 2

 

Proof in 1D (2D proof follows directly the same procedure)

 

 G f  df x   g x e  2if x x g * x'e 2if x x dxdx df x 

2
x


  g x g * x' e  2if x  x  x  df x dxdx  


  g x g x' x  x dxdx    g x  dx
* 2



Convolution theorem:

 
F    g  , hx   , y   dd   G  f x , f y H  f x , f y 
  

Proof in 1D (2D proof follows directly the same procedure)

 
F   g  hx   d    g  hx   de  2if x x dx 
  
  g  e  2if x dhx   e  2if x  x   d x    
 G  f x H  f x 

Autocorrelation theorem:

 
F    g  , g *   x,  y dd   G  f x , f y 
2

  

Proof in 1D (2D proof follows directly the same procedure)

 
F   g  g *   x d    g  g *   x de  2if x x dx 
  
  g  e  2if x dg *   x e 2if x   x  dx 

 G  f x G *  f x   G  f x 
2

Fourier integral theorem:

F F 1 g  F 1 F g  g

Fourier transform of separable functions:

If

g x, y   g x x g y  y 
then

F g x, y   Fx g x x Fy g y  y 

It is useful to consider some FT

 f2
 
F exp  a 2 x 2  1
exp    x2 
a  a 

f 
F rect ax  
1
sin c x 
a  a 

F  ax  
1
a

 a
F exp iax     f x  
 2

As well as to consider FT in cylindrical coordinates (especially for functions which

are only dependent on r).

Derivation of the Fresnel diffraction formula

Lets start again from Maxwell's equations:

1 D
xH  0 D  0
c t
1 B
xE  0 B  0
c t

Since  is generally space invariant, and by using:

    E     E    2 E

We get:
 1 B  1    B      1 E  n2 2 E
    E            2
c t  c  t  c  t c t  c t 2
E  
  E    E  E       E  ln    2 E  ln n 


From which we get, overall:

n2  2 E
 2 E  2E   ln n   0
c 2 t 2

Again, for a homogeneous medium we get the scalar equation:

n2  2 E
2 E  0
c 2 t 2

Which is valid anywhere except the boundaries

If the scalar field is assumed to be of the form (monochromatic):

uP, t   APcos2t   P

And the scalar Maxwell equation holds, we get for U P   APexp i P the
Helmholtz equation:

 2
 k 2 U  0


with k  2n .
c

This equation can be solved by means of the use of Green's theorem, relating the
values of a function inside the volume to a surface integral.

Briefly, this states that for any two "reasonable" functions U,G:

G U 
   dvU G  G U     dsU n  G n 
2 2 

I will not go through the details, but a long description can be found in Goodman.
Intuitively, the Green's function chosen for the Helmholtz equation is a spherical
wave:

e ikR
G
R

For which:
  2
 k 2 G   R 

Where R is the distance from an arbitrary point P1

Performing the integration on the left hand side we are left with the value of U at the
origin (since the laplacian operator elsewhere is just k2 – from Helmholtz equation).
Thus, the integration is done over a surface comprising a plane and a hemisphere
stretching to infinity. Under some conditions, the value of the latter vanishes, leaving
only the integration over the plane.

This results in the Rayleigh-Sommerferld solution, which is really a mathematical

formulation of the Huygens principle.

e ikr01
U P0    
1
 1 r01 cos ds
U
i aperture
P

Where the cosaccounts for the exact choice of the Green's function, and in any case
vanishes for small angles.

Relating this to what we said previously, this can be interpreted as the impulse
response of propagation from the aperture, where:

1 e ikr01
hP0 , P1   cos 
i r01

Now let us consider the small angle approximations for this. We get the integral:

e ik z0  x1  x0   y1  y0  
2 2 2 1/ 2

U P0    U x1 , y1  z 2  x  x 2   y  y 2
1
cos dx1 dy1
i aperture
0  1 0 1 0 1/ 2

The dependence on the integration coordinates appears both in the numerator (phase
term) and in the denominator (amplitude term). For large z, obviously the amplitude
contribution is small. The cosine term is also an amplitude term which has a small
effect for large z. Phase terms can have, of course, a much higher effect, so we will
keep them. We thus get:

ik z02   x1  x0 2   y1  y0 2 
U P0    
1 1/ 2

 1 1
U
iz aperture
x , y e dx1dy1

Fresnel diffraction:

Let us now consider two more restrictive solutions to the problem of diffraction from
an aperture. The first is effectively a small angle solution, retaining only the first order
(in x,y) terms of:
  1  x  2 2 2
  y     1  x     y     
2 2
 
r01  z  x      y     z 1          
2 2
  
2

 2 
 z   z  
 8 
 z   z   
 

Since we want to nerglect the higher order terms, we require that

2
2 z  x     y    
2 2

       1
 8  z   z  

or:

z 3 

4

 x      y   
2 2
 2

 4
4
R
Where R is a typical dimension of observation in both planes.

For R=1cm and =1m, this amounts to:

z 3  10 4 cm 3  z  20cm

In this case, free space propagation can be described by:

e ikz  x 2  y  2 
ik
U x, y    U  ,  e 2z
dd
iz aperture

Fraunhoffer diffraction:

This is a more restrictive approximation. Let us first expand the Fresnel formula. We
get:

e ikz  2 2 x  x 2  2 2 y  y 2 
ik
U  x, y    U  ,  e 2z
dd
iz aperture

If z is so large that:

 2
z 

   2 
Than the quadratic phase terms on  and  can be neglected and we are left with:

ik
x  y2
2

ik
ikz 2z   x  y 
U  x, y    U  , e
e e z
dd
iz aperture
Which is just a (scaled) Fourier transform of the aperture field

For R=1cm and =1m, this amounts to:

z  10 4 cm  z  100m
Simple Fraunhoffer diffraction patterns:

Rectangular and circular apertures

For transmission through a rectangular aperture we get:

     
t a  ,   rect  rect  
 2w 
 2 wx   y 

This being a separable function, the Fourier transform is just a multiplication of the
two individual Fourier transforms, giving:

Ae ikz e ik x  y2   2w y y 
2
 2w x 
 
U x, y  sin c x  sin c 
iz  z    z 

and for the intensity I  U

2

A2  2w x   2w y y 
I x, y   sin c 2  x  sin c 2  

 z  z   z 
2 2

z
The width of the central lobe in each direction is: d 
w

For transmission through a circular aperture we get:

q
t a q   circ  
 w

For which the Fourier transform is a Jinc function

Ae ikz e ikr J 1 kwr z 

2

U r   2
iz kwr z

With the corresponding intensity distribution:

A 2 J kwr z 
2

I r   4 2 2 1
 z kwr z 2

z
This is called an Airy pattern. The width of the central lobe in this case is d  1.22
w

Amplitude grating

Consider a rectangular containing an amplitude grating:

       1 m
t a  ,   rect  rect    cos2f 0 
 2w   2 2 

 2 wx   y 

In the y direction this leads to the "standard" sinc dependence. In the x direction, we
can apply the convolution theorem to get the response, using the Fourier transform of
the grating:

1 m  1
F   cos2f 0     f x     f x  f 0     f x  f 0 
m m
2 2  2 4 4

To get:

Ae ikz e ik x  y2   2w y y    2w x  f 0 z   m  2 w  x  f 0 z   
2
 2w x  m
U x, y   sin c sin c x   sin c x   sin c x 
2iz  z   z  2  z  2  z 

If wx f 0  1, corresponding to many grating lines in the aperture, the peaks are well
separated, and the intensity distribution is approximately the sum of the squared
amplitudes:

2  2 w x  x  f 0 z    2 w  x  f 0 z   
A2  2w y y   2  2w x x  m2 m2
I x, y   sin c 2



 sin c    sin c    sin c 2  x 
4 z  z    z  4  z  4  z 
2 2

Thus, the forward transmission is 1/4, and each of the sidelobes (termed +1 and -1
orders) have a maximal efficiency of m2/16, or a maximum of ~6%.

To become more efficient either phase gratings or reflectance gratings have to be

used. These will be discussed more elaborately in the next tutorial.

Angular spectrum of Waves

An alternative solution to the problem of diffraction is presented by considering free-

space propagation as a linear transformation of the solutions of the free-space
Helmholtz equation:

 2
 k 2 U  0

The solutions of which are plane waves.

Let us consider the field distribution at z=0 as a summation of gratings of different

spatial frequencies (a Fourier transform in real space):

Then:

A f x , f y ,0    dxdyU x, y,0e


2i f x x  f y y 

where the inverse transform is:

 Ax, y,0    df x df yU  f x , f y ,0e

2i f x x  f y y 

Physically we can interpret this as a sum of plane waves, since a plane wave of the
form:

px, y, z, t   e i k r t 


with:

 2
k xˆ  yˆ  zˆ 


has an intensity pattern which is periodic in the plane, with the period which increases
as   1   2   2 becomes smaller.

Thus, using:

  f x ;   f y ;   1  f x 2  f y 2

we see that the Fourier transform is just a decomposition to plane waves.

Let us now consider the propagation of these in space.

Since:

  
      2i  x  y 
U x, y, z     d d A , , z e    
    

Taking the Laplacian of U, we get:

  
 2 
  d  d  A  ,  , z e     
2


2i  x  y 
 U  x, y , z   
2
   
2 2

    
  
  d2     2i  x  y 
d d
  dz 2
A , , z e    
  

which, using the Helmholtz equation becomes:

  
 2 
  d  d  A  ,  , z e     
2


2i  x  y 
  1  
2 2

    
  
  d 2     2i   x   y 
   d  dz 2 A  ,  , z e
d 0

Since this integral equation is fulfilled for any distribution over the integration
variables, the equation must also hold for the integrand, not just the integral, leaving:
 d 2      2    
2

A , , z    2 2

 1     A , , z   0 
dz       
2
  

Whose solution is:

2i
      1 2   2 z
A , , z   A , ,0 e 
     

Physically this means that free space propagation just introduces a relative phase shift
between interferences of various plane wave components.

The Fresnel approximation corresponds to the approximation

f x 2 f y 2
1  f x   f y   1  
2 2

2 2
or to lowest order small angle diffraction.

Evanescent waves

Clearly, something is wrong with this picture when the spatial frequency is greater
than 2/. In this case  2   2  1 , which means that  is imaginary. The solution in
that case is not of a propagating wave, but rather of an exponentially decaying wave:

2i
       z  2   2 1
A , , z   A , ,0 e 
     

These waves do not propagate, and decay over a length scale of the inverse spatial
frequency. All the information on these, which represents the fine detail of the
aperture, is lost in the far field.

In fact, free space propagation in the far-field can be considered as a circular filter in
spatial frequency space.

 
2i
      1 2   2
A , , z   A , ,0 circ  2   2 e 
     

The effect of a diffracting aperture on the ASPW

For a diffracting aperture, the transmitted signal is:

U t x, y   U 0 x, y t a x, y 

Since multiplication is real space is convolution in Fourier space, then:

         
At  ,   Ai  ,   T  , 
        

  
    2i  x  y 
T  ,     t a x, y e     dxdy
  

For an incoming plane wave, the outgoing angular spectrum is just the Fourier
transform of the aperture. Thus, the smaller the aperture, the broader the angular
distribution (wider angle diffraction).

Fresnel diffraction of an amplitude grating (Talbot images).

Let us consider again an amplitude grating, but now only in the Fresnel
approximation, and using the ASPW approach. For simplicity lets consider an infinite
grating in 1D.

We start with
1 m 
t a      cos2f 0 
2 2 

The transfer function for any given component of the ASPW is:

 
2i
1 2   2
H  ,    circ    e
2 2 

Such that for L=1/2f0 we get:

iz
 1  
H   ,0   e L2

 L 

Since at z=0 the Fourier transform is:

1 m  1
F   cos2f 0     f x     f x  f 0     f x  f 0 
m m
2 2  2 4 4

The at any z we get:

 z  z
j 2 j 2
F U x, y, z     f x     f x  f 0 e L    f x  f 0 e L
1 m m
2 4 4

Which is simplified to:

 z
 2x   j L2
U x, y, z  
1 m
 cos e
2 2  L 

To yield the intensity distribution:

 1  2x   z  2  2x 
I x, y, z   1  2m cos  cos 2   m cos 
2

4  L   L   L 

Now there are three special cases, where this expression get further simplified.

 z
For  2n :
L2

2
1  2x 
I x, y, z   1  m cos 
4  L 

Which is an exact image of the one at z=0.

 z
For  2n   :
L2

2
1  2x 
I x, y, z   1  m cos 
4  L 

Which is an inverted image of the one at z=0.

z
For  n   2 :
L2
2
1  m2 m2  4x 
I x, y, z   1   cos 
4 2 2  L 

Which is an amplitude grating at the doubled frequency.

This phenomenon is related to any revival in a periodic system. For the case of a
general periodic system it holds only in the Fresnel approximation, since then the
various frequency components propagate in integer multiples of the same phase
factor.