You are on page 1of 36

E.

Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Chapter 1

Fourier Optics

The mathematical representation of the plane wave solutions of Maxwell’s equations as the
components of a Fourier expansion is a powerful approach for analyzing complex optical
wave propagation. The resulting framework, known as Fourier optics, allows the direct
application of the mathematical tools of Fourier analysis to various optical phenomena and
systems. This approach not only greatly simplifies the analysis, but also often yields
additional insight into the physics involved.
After a brief recall of the elements of Fourier analysis, we develop in this chapter the
main tools of Fourier optics. These are employed for analyzing the propagation of optical
beams both in the space and in the time domains. Applications of Fourier optics in optical
imaging and optical pulse propagation are illustrated.

1.1 Fourier Analysis – a Review

We review here the definition of the Fourier transform and recall some useful
properties of this mathematical tool.

1.1.1 Definitions

The Fourier transform is a mathematical operation that connects two conjugate


spaces. In its application in optics, these conjugate spaces are mostly either the spatial
coordinate x and the corresponding wave number k, or the time t and the radian frequency ω.
Thus, the Fourier transform of a function f ( x ) of the spatial coordinate reads


F ( k ) = ∫ f ( x ) e−ikx dx ≡ F [ f ( x )]
€ −∞

and the corresponding inverse Fourier transform is given by


€ ∞
1
f ( x) = ∫ F (k) e +ikx
dk ≡ F −1 [ F ( k )]
2π −∞

Similar expressions hold for the relationship between the function f ( t ) and its Fourier
transform F (ω ) :



1
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics


F (ω ) = ∫ f (t ) e −iωt
dx ≡ F [ f ( t )]
−∞

1
f (t) = ∫ F (ω ) e +iωt
dk ≡ F −1 [ F (ω )]
2π −∞


We will often need to perform a Fourier transform between two two-dimensional (2D)
spaces, e.g., the ( x, y ) plane in real space and its conjugate plane ( k x ,k y ) in the wave number

space. The corresponding 2D Fourier transforms are given by

f ( x, y ) e ( x y ) dxdy
−i k x +k y
€ F ( k x,k y ) = ∫∫ € ≡ F [ f ( x, y )]
−∞

and
1 ∞
F (kx,ky ) e ( x y )dk x dk y ≡ F −1 F (kx,ky )
i k x +k y
€ f (x, y ) =
(2π )
2 ∫∫
−∞
[ ]
In what follows, we will use an equivalent, symmetric definition of the Fourier
transform, written in the form


F (ν x ) = ∫ dx f ( x ) e −i2 πν x x

−∞

f ( x) = ∫ dν F (ν ) e
x x
+i2 πν x x

−∞


where ν x ≡ k /2π denotes the spatial frequency (measured in inverse length). Similarly, in
the time domain we have


€ F (ν ) = ∫ dt f (t ) e −i2 πνt

−∞

f (t) = ∫ dν F (ν ) e +i2 πνt

−∞


where ν ≡ ω /2π indicates the temporal frequency (measured in Hz).

1.1.2 Properties of the Fourier Transform

Some mathematical properties of the Fourier transform that are frequently used are
listed below:

Linearity:
F [ f + g] = F [ f ] + F [ g]
Scaling:

F [ f ( t τ )] = τ F (τν )

2

E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

where F (ν ) = F [ f ( t )] . As an example, note that F [ f (−t )] = F (−ν ) .

Translation:

€ F [ f ( t −€
τ )] = e +i2 πντ F (ν )
F [ f ( t ) e−i2 πν 0 t ] = F (ν + ν 0 )

Symmetry: €

If f ( t ) is a real function,

F (−ν ) = F * (ν )

Convolution theorem:


The convolution of two functions, defined as

f1 ∗ f 2 ≡ ∫ f (τ ) f (t − τ ) dτ
1 2
−∞

is equal to the inverse Fourier transform of the product of their Fourier transforms:

f1 ∗ f 2 = F −1 [ F1 (ν ) F2 (ν )]

where F [ f1 ( t )] = F1 and F [ f 2 ( t )] = F2 .

Autocorrelation theorem:

€ The correlation of
€ two functions, defined as



f (t) ≡ ∫ f (τ ) f (t + τ ) dτ
1 2
−∞

has a Fourier transform related to the transforms of the two functions as



F (ν ) = F1∗ (ν ) F2 (ν )

Parseval’s theorem:

€ its Fourier transform F (ν ) = F [ f ( t )] are equal:


The norms of a function and

∞ ∞
2 2
∫ dt f (t ) = ∫ dν F (ν )
−∞
€ −∞


3
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Uncertainty relations:

We define the mean of a variable as


∞ ∞
2 2
t = ∫ t f ( t ) dt ∫ f (t) dt
−∞ −∞

and its variance as


€ ∞
2

2
2
σ t2 = ∫ (t − t ) f ( t ) dt ∫ f (t) dt
−∞ −∞

We then obtain that the variances in the two conjugate Fourier spaces are related by the
uncertainty relations

1 ' 1*
σ tσ ν ≥ )σ tσ ω ≥ ,
4π ( 2+
and

1 & 1)
€ σ xσ ν x ≥ (σ xσ k ≥ +
4π ' 2*

As an example, note the application of the Fourier transform and these uncertainty relations
to wave packets in quantum mechanics. The momentum of a particle is related to the wave
€ yields directly the Heisenberg relation
number by p = k , which

σ xσ p ≥  2 .

1.2 Spatial Propagation

The Fourier transform allows expanding an optical wave of arbitrary form in a


superposition of plane waves. The principle of superposition (see Optics I) then allows to
analyze the propagation of the optical wave by treating the propagation of each plane wave
component and then using the inverse Fourier transform to reconstruct the waveform. We
will now see how Fourier transforms “naturally” come out when spatial propagation of
optical waves is considered.

1.2.1 Angular Spectrum of Plane Waves

Consider a plane wave in the scalar field approximation, with its complex
representation
 
U ( x, y,z) = Ae i k ⋅ r = Ae ( x y z )
i k x +k y +k z

€ 4
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Figure 1.1: Wave vector of a plane wave and the associated direction angles.

Time dependence of the form e−iωt is assumed. (Note: in our convention, a propagating wave

is written as Ae (
i k ⋅ r − ωt )
). The norm of the wave vector,

1

( 2
k = k x + k y + kz
2 2
) 2


is related to the wavelength by λ = 2π k . The orientation of the k-vector determines the
direction of propagation of the corresponding plane wave in the 3D space. This direction can

be expressed in terms of the angles (θ x , θ y ) between k and its projections on the y-z and x-z
planes, respectively (see
€ Fig. 1.1). These angles are given by

) ; θ y = sin−1 ( ky k )
θ x€ = sin−1 ( k x k€

A plane wave of given amplitude and wavelength can thus be completely characterized by
the direction angles (θ x , θ y ) .

To follow the propagation in space of a given plane wave, we start with its spatial
across the plane z = 0 :
distribution €

U ( x, y,0) ≡ f ( x, y ) = A e (
i kx x +ky y )


We introduce the spatial frequencies related to the two relevant components of the wave
vector,

ν x,y ≡ k x,y 2π

measured in cycles/length. These are related to the direction angles by



θ x,y = sin−1 ( λν x,y )

Using the spatial frequencies, we can rewrite the field at z = 0 as


5€
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

i2 π (ν x x + ν y y )
U ( x, y,0) = Ae

ikz z
To propagate the plane wave to a point z, we just need to multiply it by a factor e .
2 2 2
Since k 2 = kx + k y + k€
z , we have

1
(
kz = ± k 2 − k x − ky
2 2
) 2


Thus, at the point z, the propagated plane wave is written
€ 1
+i( k 2 −kx 2 −ky 2 ) 2 z i2 π (ν x x + ν y y )
U ( x, y,z) = e ikz zU ( x, y,0) = Ae e

Note that for real propagation (as opposed to exponential decay) in the positive z-direction,
2 2
we must have kx + k y < k 2 and the positive sign should be selected. This simple example

shows that real-space propagation looks particularly simple in the Fourier-conjugate (k-
vector) space: it amounts to simple multiplication by a phase factor.
For the more general case, where the field has an arbitrary shape U ( x, y,0) = f ( x, y )

at the starting point z = 0 , we first compute its 2D Fourier transform at that point:

−i2 π (ν x x +ν y y )
F (ν x ,ν y ;0) = ∫∫ dxdy f ( x, y ) e €
€ −∞

Equivalently, this Fourier transform can be written in terms of the corresponding angles
(θ x , θ y ) as €

$ sin θ x sin θ y ' ∞
−i ( x sin θ x +y sin θ y )

F&
% λ
,
λ
;0) =
(
∫∫ dxdy f ( x, y ) e λ

−∞

The inverse Fourier transform then reads


€ ∞
i2 π (ν x x +ν y y )
U ( x, y,0) = ∫∫ dν x dν y F (ν x ,ν y ;0) e
−∞

or,
€ ∞
' sin θ x * ' sin θ y * ' sin θ x sin θ y * i 2λπ ( x sin θ x +y sin θ y )
U ( x, y,0) = ∫∫ d) , d)
( λ + ( λ + ( λ
, F) ,
λ
;0, e
+
−∞

The expressions of the inverse Fourier transform show explicitly that the field at z = 0
can be decomposed into a superposition of plane waves propagating in different directions in

space (see Fig. 1.2). The amplitude of each plane wave is given by F (sin θ x / λ,sin θ y / λ ;0) ,

6
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Figure 1.2: An arbitrary wave at z = 0 is equivalent to a superposition of plane waves of


different amplitudes, propagating at different angles.


which is the Fourier transform of the initial field distribution. This angular spectrum of
plane waves is the starting point for the calculation of the wave propagation in space. (Note:
often we will be interested in propagation of plane wave components whose wave vectors are
almost parallel to the propagation axis. In this paraxial propagation, the paraxial
approximation can be used, for which θ x,y << 1 and hence sin θ x,y ≈ θ x,y ).

1.2.2 Propagation of the Angular Spectrum


€ €
The ease of “propagating” a plane wave in the conjugate Fourier space provides a
powerful strategy for propagating a wave of arbitrary shape. This strategy is illustrated in
Fig. 1.3. The idea is to decompose the given optical wave into a superposition of plane
waves, propagate each plane wave in k-space using a proper phase factor, and then to sum up
all the propagated plane waves in order to get the propagated wave in real space. In practice,
going back and forth between the conjugate Fourier spaces is accomplished via direct and
inverse Fourier transforms, as shown in what follows.
To develop the propagation in free space of each component of the angular spectrum,
we first write the 2D Fourier transform of the field for z > 0 :

−i2 π (ν x x + ν y y )
F (ν x ,ν y ;z) = ∫∫ dxdy U ( x, y,z) e
−∞ €
Equivalently, using the inverse transform, we can write the field in terms of its angular
spectrum components:


i2 π (ν x x +ν y y )
U ( x, y,z) = ∫∫ dν x dν y F (ν x ,ν y ;z) e
−∞

Obviously, the optical field U ( x, y,z) satisfies the Helmholtz equation:



∇ 2U ( x, y,z) + k 2U ( x, y,z) = 0


7
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Figure 1.3: Using Fourier transform for solving optical wave propagation: real-space
and k-space propagations.

We now use this equation to derive an equivalent equation for the angular spectrum.
Inserting the expression of U ( x, y,z) in terms of F (ν x ,ν y ;z) in the Helmholtz equation, we
get

∞ ') d 2 F (ν ,ν ;z) +)
€ + k − (2π ) ν x + ν y F (ν x ,ν y ;z), e ( x y ) = 0
∫∫
−∞
dν x dν y (
)*
x

dz 2
y
[
€2 2 2 2
( )-
)]
i2 π ν x,ν y

which yields the equivalent equation for F (ν x ,ν y ;z) :



d2
dz 2
[ 2
( 2
)]
F (ν x ,ν y ;z) + k 2 1− λ2 ν x + ν y F (ν x ,ν y ;z) = 0

the solution of this equation is easily found:
€ ( )
ik 1− λ2 ν x 2 +ν y 2 z
F (ν x ,ν y ;z) = F (ν x ,ν y ;0)e = F (ν x ,ν y ;0)e iκz

where we have defined the parameter



( )
κ ≡ k 1− λ2 ν x 2 + ν y 2 = k 1− (sin2 θ x + sin 2 θ y )

The propagation in the conjugate Fourier space (of k-vectors or angles) thus amounts to
simple multiplication by the factor e iκz :

F (ν x ,ν y ;z) = F (ν x ,ν y ;0)e iκz

Two particular cases of interest can be distinguished, according to the position of the
spatial frequency of the plane waves in the (ν x ,ν y ) plane (see Fig. 1.4):

8

E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

2 2
1) ν x + ν y < 1 λ2

In this case sin 2 θ x + sin 2 θ y < 1 and κ is real.


€ The propagation in the Fourier space is thus described by a simple phase factor, for
the corresponding angles, with accumulated phase that depends on the angle of
propagation. €

2 2
2) ν x + ν y > 1 λ2 ,
In this case sin 2 θ x + sin 2 θ y > 1 and κ is imaginary.
It is then convenient to write

€ ( ) (
κ ≡ k 1−€λ2 ν x 2 + ν y 2 = ik λ2 ν x 2 + ν y 2 −1 )
and to define a new parameter κ˜ such that κ ≡ iκ˜ (κ˜ ≡ −iκ ) :


(
κ˜ = k λ2 ν x 2 + ν y 2 −1 )
€ € €
such that the propagation in the spatial frequency space reads

€ F (ν x ,ν y ;z) = F (ν x ,ν y ;0) e−κ˜z

One can see that, in this case, each Fourier component decays exponentially along the
propagation direction. In fact, the field of each component reduces to 1/e times its
value at z = 0 €
at a distance

( +
( 2 2
)
z = 1/κ˜ ≡ λ /*2π λ2 ν x + ν y −1- = λ / 2π
) , [ (sin θ
2
x ]
+ sin2 θ y ) −1

which is typically a fraction of a wavelength. These decaying components are
appropriately termed evanescent components of the angular spectrum of plane waves.

Figure 1.4: Circle in the spatial frequencies space, separating the propagating and
evanescent field propagation regions.

9
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Once the propagation scheme of any component of the angular spectrum has been
found, the propagation in real space can be determined by implementing the corresponding
Fourier transform. We get:

i2 π (ν x x +ν y y )
U (x, y,z) = ∫∫ dν x dν y F (ν x ,ν y ;z)e
−∞

i2 π (ν x x +ν y y )
= ∫∫
−∞
[
dν x dν y F (ν x ,ν y ;0)e iκz e ]
Thus, the strategy for evaluating the propagation in real free space is as follows:

(i) €Calculate the Fourier transform, or angular spectrum, of the starting field;
(ii) Propagate the angular spectrum by multiplication of the exponential factor;
(iii) Inverse Fourier transform to get back to real space.

The simplicity of the “propagator” in the Fourier space, e iκz combined with the well-
known techniques of the Fourier transform are at the basis of this powerful approach for
calculating the propagation of complex optical fields.

1.2.3 Spatial Filtering

So far we have considered propagation in free space, free of obstacles or absorbing


media. The propagation past obstacles (e.g., apertures) can be evaluated also in Fourier space
by considering the impact of the obstacle on each angular spectrum component and then
inverse Fourier transforming to real space. In this approach, the details of the propagation in
real space are manifested in modifications of the angular spectrum components by different
elements of the propagation medium. In particular, this includes the possible elimination of
some angular spectrum components, which is referred to as spatial filtering.
We consider first propagation through an aperture Σ placed in a screen at z = 0 in the
x-y plane. The aperture is defined by its transmittance function:

#1 in Σ
t ( x, y ) = $ €
%0 otherwise

Using Kirchhoff’s boundary conditions (see Optics I), the transmitted wave just past the
screen is given by

U t ( x, y,0) = U i ( x, y,0) ⋅ t ( x, y )

To find the effect of the aperture on the angular spectrum components, we use the
convolution theorem of the Fourier transform, which directly yields

Ft (ν x ,ν y ;0) = Fi (ν x ,ν y ;0) ∗ T (ν x ,ν y )

In this expression,

10
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics


−i2 π (ν x x +ν y y )
T (ν x ,ν y ) = ∫∫ t ( x, y ) e dxdy
−∞

is the Fourier transform of the aperture function.


€ example, consider a plane wave propagating along the z-axis. We then
As a simple
have an incident angular spectrum of the form

Fi (ν x ,ν y ;0) = δ (ν x ,ν y )

The transmitted function is found by the convolution



Ft (ν x ,ν y ;0) = δ (ν x ,ν y ) ∗ T (ν x ,ν y )

which yields
€ −i2 π (ν x ' x +ν y 'y )
Ft (ν x ,ν y ;0)=∫ ∫ ∫ ∫ δ (ν '−ν , ν '−ν ) dν ' dν ' t (x, y) e
x x y y x y dxdy

= ∫∫ dxdy t (x, y ) e (
π ν ν ) −i2 xx+ yy
= T (ν ,ν ) x y

Hence, T (ν x ,ν y ) , the Fourier transform of the transmission function of the aperture, yields
€the transmitted angular spectrum. This is exactly the result we have obtained using
diffraction theory (see Optics I).
€ Consider now again the propagation in Fourier space. As we have shown, the angular
spectrum propagates from z = 0 to z according to

F (ν x ,ν y ;z) = F (ν x ,ν y ;0) e iκz



2
( 2
)
with κ = k 1− λ2 ν x + ν y . In real space, the propagated field is then given by


i2 π (ν x x +ν y y )
U ( x, y,z) = ∫∫ dν x dν y F (ν x ,ν y ;0) e iκz e
€ −∞
It is convenient to describe the propagation in terms of a transfer function

F (ν x ,ν y ;z)
€ H (ν x ,ν y ;z) ≡ = e iκz
F (ν x ,ν y ;0)

that implements the propagation in the Fourier space through simple multiplication. For
sufficiently long propagation distance, the evanescent terms (i.e., the terms with spatial
frequencies yielding€ imaginary parameter κ ) can be neglected and the transfer function
becomes

11
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

⎧
⎪e ik 1− λ2 ν x 2 +ν y 2 z
( ) 2 2
H (ν x ,ν y ;z) → ⎨ ; ν x + ν y < 1/ λ2
z →∞ ⎪ 0 ; otherwise
⎩

In this sense, propagation through free space “removes” the evanescent terms from the
propagated field such that only terms associated with small enough spatial frequencies (those
obeying €ν x 2 + ν y 2 < 1 λ2 ) survive. This process is referred to as spatial filtering.
We will implement the concepts presented above in a simple yet interesting example
involving propagation through a transparent film of nonuniform refractive index distribution.

Phase gratings

Consider first a plane wave propagating in the positive z-direction and illuminating a
phase grating placed in the x-y plane (see Fig. 1.5). The field just before the grating is given
by

U ( x, y,0− ) = A

The phase grating is characterized by a transmittance function given by


€ i2 π (ν x x +ν y y )
t ( x, y ) = e

This result is just expressing the fact that the phase of an incident wave is modulated
−1
according to the argument ν x x + ν y y . For example, the special case ν x = Λ x ; ν y = 0

corresponds to a phase modulation along the x-direction with a period Λ x . The field just past
the phase grating then becomes
€ i2 π (ν x x +ν y y ) €
U ( x, y,0 + ) = At ( x, y ) = A e €

Thus, the incident plane wave, which has no phase variation in the x-y plane before
impinging on the phase, now propagates with direction angles given by

Figure 1.5: Propagation of a plane wave impinging on a phase grating.

12
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

(θ ,θ ) = (sin
x y λν x ,sin−1 λν y ) . That is, the phase grating represents a diffraction grating that
−1

diffracts the incident plane wave into the appropriate direction in space.

In a more general case, the phase grating is made of a superposition of many “simple”
€ phase gratings, and the diffracted field becomes

U ( x, y,0 + ) = At ( x, y ) = A ∫∫ F (ν x ,ν y ;0) e ( x y ) dν x dν y
i2 π ν x +ν y

where F (ν x ,ν y ;0) are now the Fourier components of the corresponding transform of the
grating€transmittance function t ( x, y ) . (From another viewpoint, this also corresponds to a
transparent film of arbitrary variation in the refractive index, or thickness, in its plane; the
Fourier transform of the phase then represents the variation in terms of many periodic

components superposed on each other.) Thus, the optical field right past the gratings

contains many diffraction orders, each with relative weight (or diffraction efficiency) given
by the corresponding Fourier component F (ν x ,ν y ;0) . However, not all of the corresponding
angular components propagate a long distance past the grating, as all components with
ν x 2 + ν y 2 > 1 λ2 are evanescent and therefore do not transfer field energy. This is
reminiscent of the phenomenon€ of total internal reflection, where angles greater than the
critical angle do not contribute to energy flow into the low refractive index medium.
€ For a slowly varying phase in the x-y plane we can rewrite the transmittance,

t ( x, y ) = e i2 πφ ( x,y)

using the approximation

φ ( x, y )€ ≈ φ ( x 0 , y 0 ) + ( x − x 0 ) νx + ( y − y 0 ) ν y

∂φ ∂φ
= =
∂x x0 ∂y y0

Based on the previous discussion, we can see that this yields a position dependent diffraction
angle. As an example, consider the case

φ ( x ) = −x 2 2λf

where f is a given parameter. This particular phase variation yields a spatial frequency

€ x sin θ x
νx = − =
λf λ

and hence a diffraction angle

€ sin θ x = − x f

With such quadratic phase variation, one can distinguish two different cases,
depending on the relative size of the beam with respect to the phase variation:

13
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

• Narrow beam: yielding steering of the diffracted beam when the incident beam is
scanned across the transparency (see Fig. 1.6).
• Wide beam: resulting in focusing (imaging), acting as a Fresnel lens (see Fig. 1.7).

Figure 1.6: Beam scanning with a phase grating.

Figure 1.7: Beam focusing with a phase grating.

1.2.4 Fresnel and Fraunhofer Diffraction

The Sommerfeld-Kirchhoff diffraction formula, derived in chapter 6 of Optics I,


yields the (scalar) optical field U ( P0 ) diffracted by an aperture Σ :

A e ik( r21 +r01 )


U ( P0 ) =

∫∫ r21r01 €
cos(nˆ ,r01 ) ds
€ Σ

We will show now that, under certain useful approximations, this diffraction formula can be
cast in the form of a 2D (spatial) Fourier transform.

14
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Figure 1.8: Diffraction through an aperture.

For the geometry shown in Fig. 1.8, the diffracted optical field can be rewritten as

1 e ikr01 
U ( x 0, y 0 ) =

∫∫ dx1dy1 r01
U ( x1, y1 ) cos( nˆ , r01 )
Σ

where U ( x1, y1 ) is the optical field across the aperture. In the spirit of the Kirchhoff boundary
conditions, we demand that the field vanishes on the opaque screen, which permits us to

extend the integration limit to infinity:

1 ∞ ∞ e ikr01 
U ( x0 , y0 ) = ∫ ∫ dx1 dy1 U ( x1 , y1 )cos( nˆ,r01 )
iλ −∞ −∞ r01

A common situation is when the distance between the plane of the aperture and the
observation plane (screen) is much bigger than the size of the aperture. In this case, we can

make the approximations

#cos( nˆ , r01 ) ≈ 1
$ ikr ikr
%e 01 /r01 ≈ e 01 /z

We also expand the distance

€ 1/ 2
2 1/ 2
) # x − x &2 # y − y &2,
[ 2 2
r01 = z + ( x 0 − x1 ) + ( y 0 − y1 ) ] = z +1+ % 0 1 ( + % 0 1 ( .
* $ z ' $ z '-

using the Taylor series


15
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

1 1
1+ b = 1+ b − b 2 + ... ( b < 1)
2 8

and obtain
€ * 1 $ x − x '2 1 $ y − y '2-
r01 ≈ z ,1+ & 0 1 ) + & 0 1 ) /
+ 2% z ( 2% z ( .

This approximation is valid for points sufficiently close to the z-axis, and hence is known as
the paraxial approximation. Inserting the latter expression in the Sommerfeld-Kirchhoff
€ we get
diffraction formula,

$e ikz i k ( x 0 2 +y 0 2 ) ' ∞ ∞ k
i ( x1 2 +y1 2 ) −i

( x 0 x1 +y 0 y1 )
U ( x 0 , y 0 ) ≅ & e 2z ) ∫ ∫ dx1dy1U ( x1, y1 ) e 2z e λz
% iλz ( −∞ −∞

This formulation of the optical field involves the integration of the field amplitude, multiplied
by a certain phase, across the aperture. Notice that the phase involves only terms that are
€ or quadratic in the coordinates of the aperture ( x , y ) , as a result of the paraxial
linear 1 1
approximation. This expression is referred to as the Fresnel approximation to the diffraction
formula.
€ rewritten as
The Fresnel diffraction expression can be

$e ikz i k ( x 0 2 +y 0 2 ) ' ∞ ∞ * i ( x1 2 +y1 2 ) - −i


k 2π
( x 0 x1 +y 0 y1 )
U ( x 0 , y 0 ) ≅ & e 2z ) ∫ ∫ 1 1 ( 1 1)
dx dy + U x , y e 2z
.e λz

% iλz ( −∞ −∞ , /

which is nothing but the 2D Fourier transform

€ #e ikz i k ( x 0 2 +y 0 2 ) & ) i ( x1 2 +y1 2 ) ,


k

U ( x 0 , y 0 ) = % e 2z ( F *U ( x1, y1 ) e 2z -
$ iλz ' + .

implemented with the spatial frequencies


€ x0 y
νx = ; νy = 0
λz λz

A closer look at the quadratic term will permit us to further refine the approximation
of the diffraction formula. In fact, the quadratic phase term multiplying the near field
distribution U ( x1, y1 ) can €
be approximated as unity provided that the aperture size is
sufficiently small or, equivalently, that the distance is sufficiently large:
k
i (x1 2 +y1 2 ) k
€ e 2z
≈1 if
2
z >> max x1 + y1
2
(
2
)



16
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

This condition, which is stricter than the Fresnel approximation case, is known as the
Fraunhofer approximation. In this case, the diffraction pattern U ( x 0 , y 0 ) becomes

$ ikz i k ( x 0 2 +y 0 2 ) ( ∞ ∞
& e e 2z & −i2 π (ν x x1 +ν y y1 )
U ( x 0, y 0 ) ≅ % ) ∫ ∫ dx1dy
€ 1U ( x1, y1 ) e
& iλz & −∞ −∞
' *

That is, the Fraunhofer diffraction pattern is proportional to the 2D Fourier transform of the
near field pattern U ( x1, y1 ) :

$ ikz i k ( x 0 2 +y 0 2 ) (
€ & e e 2z &
U ( x 0, y 0 ) ≅ % ) F {U ( x1, y1 ) } Fraunhofer diffraction
& iλz &
' *

Denoting

( 2
max x1 + y1 ≡ R 2
2
)
we can rewrite the conditions for the Fraunhofer approximation as

kR 2 πR 2
<< z ⇔ << z
2 λ

or

€ zλ >> πR 2

It is also customary to define the Fresnel number



R0 2
NF ≡
λz

with which the conditions of the two approximations can be expressed as:

Fresnel diffraction : NF > 1
Fraunhofer diffraction : NF < 1

The Fresnel and Fraunhofer approximations can also be interpreted in the framework
of the transfer functions introduced earlier. For the general case, the free space transfer
function is given by€
1/2

H (ν x ,ν y ) = e ikz = e
[ (
ik 1− λ2 ν x 2 +ν y 2 )] z

€ 17
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

The paraxial approximation corresponds to propagation at small angles with respect to the z-
axis, i.e.,

x 0 − x1 y 0 − y1
; << 1
z z

which is equivalent to
€ 1
ϑ x,y ≅ λν x,y << 1 ⇒ ν x 2 + ν y 2 << 2
λ

In terms of the angles subtended by the direction of propagation, the paraxial approximation
can be rewritten

ϑ 2 = ϑ x 2 + ϑ y 2 = λ2 ν x 2 + ν y 2 << 1( )
In this limit, we can write
€ % ϑ2(
1/ 2 ik '1− *z
(
ik 1−ϑ 2 ) z & 2 )
e ikz = e →e
ϑ <<1

and the transfer function in the paraxial approximation becomes

€ H (ν x ,ν y ) ≅ e ikze−ikϑ
2
z/2

= e ikze
( )
−iλπ ν x 2 +ν y 2 z

Note that the second factor represents the quadratic phase variation (in the x-y plane) in the
1/ 2 1ϑ2 ϑ4
Fresnel approximation. € [In fact, since (1− ϑ 2 ) = 1− + ..., the Fresnel
2 2 8
2π ϑ 4 ϑ 4z
( 2
( 2
approximation is valid for k 1− λ2 ϑ x + ϑ y z →
λ
⋅ z⋅ ))
8
=

<< 1.]


Using the definition of the Fresnel number, we can re-express the paraxial condition
as

2
N Fϑ max
<< 1
4

where the angle of view of the aperture from the observation point is given by

€ ϑ max ≅ R /z

Impulse Response Function



Further insight into the propagation in free space can be gained by introducing the
notion of the impulse response function. Consider again the propagation of a field from the

18
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

point z=0, where it is given by U ( x, y,0) , to the point z, where it is given by U ( x, y,z) . The
impulse response function, denoted h ( x, y ) , is defined as the result of the propagation of a
particular field, given by
€ €
f ( x, y ) ≡ U ( x, y,0) = δ ( x, y )

To calculate the impulse response function, we first calculate the Fourier transform of the
launched field,

F (ν x ,ν y ;0) = F [δ ( x, y )] = 1

We then propagate the field in Fourier space to the point z, which gives

F (ν x ,ν y ;z) = H (ν x ,ν y ) ×1 = H (ν x ,ν y )

and finally inverse Fourier transform to obtain


€ ∞ ∞
i2 π (ν x x +ν y y )
U ( x, y,z) ≡ h ( x, y ) = ∫ ∫ dν dν H (ν ,ν ) e
x y x y
−∞ −∞

That is, the impulse response function is



[
h ( x, y ) = F −1 H (ν x ,ν y ) ]
and is therefore an inherent characteristic of the propagation medium (in this case, free
space).

For cases when Fresnel’s approximation can be used, we obtain the impulse response
function

x 2 +y 2
i +ik
h ( x, y ) = e ikze 2z
λz

which assumes the form of a paraboloidal wave. Thus, every point source at the input plane
generates a paraboloidal wave upon propagation in free space. This is reminiscent of the
Huygens wavelet concept.€
The notion of the impulse response function makes it possible to represent the
propagation of a more general, initial waveform as a convolution integral. A general input
field of the form f ( x, y ) can be written as the convolution

∞ ∞
f ( x, y) = ∫ ∫ dx dy f ( x, y ) δ ( x − x, y − y)
−∞ −∞

Using the convolution definition, we rewrite this as f ∗ δ . Passing to Fourier space using the
convolution theorem then yields

f ∗ δ → F ×1 = F

19
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Propagation through distance z is accomplished by simple multiplication by the transfer


function, giving

H ×F

and using the convolution theorem backwards, going back to real space, gives

h∗ f

Explicitly, this yields the desired propagated wave in real space as


ik
[(x−x') 2
+( y−y')
2
]
i ikz
U ( x, y,z ) =
λz
e ∫ dx' dy' f ( x' y') e 2z

−∞

This result expresses the propagated field explicitly as a Huygens-Fresnel construction of


€ originating from each point at the near field pattern f (x' , y' ).
paraboloidal waves

1.3 Pulse Propagation

In this section, we apply the tools of Fourier optics to study optical pulse propagation
in dispersive media. Short optical pulses, with widths down to the single optical cycle regime
(attosecond pulses) are nowadays generated with a wide range of techniques. They are
applied in many practical systems and fundamental physics and chemistry studied both
because of their short duration as well as because of the high peak power they allow to
achieve. Examples of their applications include optical data processing communication,
medical diagnostics, chemical reaction probe and control, and material cutting and welding.
Their generation and use necessitate the understanding of their propagation in various optical
media. One important example of such media is optical fibers, in which the characteristics of
short pulse propagation are essential for understanding the limits of transmission at high data
rates, particularly along long (>km) distances.
By proper presentation of the evolution of a short pulse in time and space, we will
show that it can be formulated using Fourier optics concepts. The resulting propagation in
time has similarities to propagation in space that we already formulated in terms of Fourier
transforms. In particular, we will show that in analogy to beam spreading in space due to
diffraction, short pulses tend to broaden in time due to dispersion.

A dispersive medium is characterized by its refractive index and absorption


coefficient, both of which are frequency dependent: {n (ν ); α (ν )} . This implies that the
(real) propagation constant of a plane wave takes the form

2πν
β (ν ) = k 0 n (ν€) = n (ν )
c

20
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Figure 1.9: Evolution of the envelope function of a wave packet in time and space during
its propagation in a dispersive medium.

Figure 1.10: Propagation and attenuation constants for a dispersive optical medium.

(see Fig. 1.10). The optical field of a pulse (or a wave packet) can be written as

U ( z,t ) = A ( z,t ) e i( β 0 z−2 πν 0 t )

where A ( z,t ) is the envelope function describing the shape of the pulse and
β 0 = β (ν 0 ) = 2πν 0 /cn (ν 0 ) . We assume that the pulse is launched at z = 0 with a shape A ( 0,t )

and would like to calculate its form A ( z, t ) when it arrives at z (see Fig. 1.9).
€ For an arbitrary pulse (envelope) shape, one can Fourier-analyze the pulse profile so
that its shape is characterized by all the corresponding Fourier € harmonics. This yields
€ € €
€ ∞
A ( z,t ) = ∫ df A( z, f )e−i2 πft
−∞

The propagation problem can thus be treated in the Fourier space by following the evolution
of the Fourier components A( z, f ) along the z-axis. We concentrate here on the propagation

of one such harmonic, and show later how the general case is treated by an inverse Fourier
transformation. For such simple case of a harmonic pulse, the pulse shape at the launch point
is described by a single Fourier component, of the form

−i 2 πft
A ( z,t ) = A( z, f ) e

21
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

The pulse propagation can be evaluated in either Fourier space or in real space. We
analyze now each case in order to derive the envelope of the pulse at the end of the
propagation in terms of the initial envelope.

Propagation in Fourier Space:

At the onset, we have the envelope

−i 2 πft
A ( 0,t ) = A( 0, f ) e

The full pulse is thus


U (0,t ) = A (0,t ) e−i2 πν 0 t = A( 0, f ) e−i2 π (ν 0 + f ) t

which shows explicitly that the corresponding wave is a monochromatic wave of effective
frequency (i.e., frequency that accounts not only for the carrier frequency but also for the
€ we can already write directly the result for the propagated wave:
pulse shape. Hence,
1
− α (ν 0 + f ) z
U ( z,t ) = A( 0, f ) e−i2 π (ν 0 + f ) t e iβ (ν 0 + f ) z e 2

Propagation in Real Space:



In real space, the propagated pulse is written

U ( z,t ) = A ( z,t ) e i( β 0 z−2 πν 0 t )

and Fourier transforming the pulse envelope (or, in our case, taking one frequency
component of the transform), we get

U ( z,t ) = A( z, f ) e−i 2 πft e i( β 0 z−2 πν 0 t )

By comparing the propagated pulses obtained from the analysis in the Fourier space
and in real space, we obtain

1
− α (ν 0 + f ) z
A( z, f ) = A(0, f ) e [
i β ( ν 0 + f ) − β (ν 0 )] z
e 2

Equivalently, by introducing the transfer function for the pulse shape,

€ A( z, f ) ≡ A(0, f ) ⋅ H ( f )

we have
€ 1
− α (ν 0 + f ) z
i[ β (ν 0 + f ) − β ( ν 0 ) ] z
H( f ) = e e 2

€ 22
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Figure 1.11: Propagation of an optical pulse in an optical fiber.

The general case of an arbitrary pulse form is treated using a full Fourier expansion of
the pulse envelope function:

& ∞

( A ( z,t ) = ∫ df A( z, f )e−i2 πft


( −∞
' ∞
(
() ( ) ∫ dt A ( z,t )e
+i2 πft
A z, f =
−∞

With this approach, the initial pulse shape A (0,t ) is first Fourier transformed to obtain
A(0, f ) . This component is, in turn, propagated using the transfer function H ( f ) to obtain

A( z, f ) , which is finally inverse Fourier transformed to get the final pulse shape A ( z,t ) .
Symbolically, the analysis procedure€is:
€ €
A (0,t ) → A(0, f ) → A( z, f ) →−1 A ( z,t )
€ F H F €

For sufficiently narrow pulse envelopes with a spectral width Δν << ν 0 , a useful
approximation can be derived. In that case, we can expand the propagation constant in a
Taylor series: €

& €
dβ 1 2 d 2β
(β (ν 0 + f ) ≈ β

(ν
 0) + f ν0 + f ν0
' dν 2 dν 2
=β0
(
)α (ν ) ≈ α

and the transfer function becomes


€ dβ i 2 d 2β
−α f z
z i f dν z
2 dν 2
H( f ) ≈ e 2
e ν0
e ν0

We now define the group velocity


€ % dβ (−1 % dβ (−1
vg ≡ ' * = 2π ' * ν0
& dω ) & dν )

with which the second term in the transfer function is rewritten


23
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

dβ 2π
if z if z
vg
e dν
=e ≡ e i2 πfτ d

where in the last step we defined the time delay suffered by the pulse:
€ z
τd ≡
vg

Next, we introduce the dispersion coefficient


d % 1 ( 1 d 2β
Dν ≡ ' *=
dν '& υ g *) 2π dν 2
ν0

which lets us rewrite

€ i 2 d 2β
f z i 2
2 dν 2 f 2 πDν z 2
e ν0
=e 2
= e iπDν f z

With these two definitions, the transfer function becomes

€ 2
H ( f ) ≅ e−αz / 2e i2 πfτ d e iπDv f z

This simplified form of the transfer function allows the physical interpretation of the
different terms. In the absence of dispersion Dν = 0 , and the propagation results only in a

time-delayed pulse, because
α
€ − z
A( z, f ) = A(0, f ) ⋅ e 2 i2 πfτ d
e

and hence

€ A( z,t ) = e−αz / 2 A (0,t − τ d )

In the absence of dispersion, the pulse shape does not change.


If Dν ≠ 0 , the time delay is different for different frequencies, and can be written as

τ d (ν ) = z /v g (ν ) . The difference in time delays for the different frequency components is
then

d %z(
€ δτ d = ' * δν = Dν zδν
dν '& v g *)

which results in pulse spreading (or diffusion).


1 2
The dispersion coefficient
€ is measured in [ Dν ] = s = s /m ⋅ Hz . Using the relation
m

Dν dν = Dλ dλ


24
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

we find

dν d $c' c
Dλ = Dν = Dν & ) = − 2 Dν
dλ dλ % λ ( λ

where [ Dλ ] = s /m ⋅ nm .

We distinguish two cases of dispersion that yield different pulse spreading
characteristics. For normal dispersion of the group index n g = c /v g we have

d %1(
Dν > 0 ⇒ ' *>0
dν€'& υ g *)

and using

€ n % ν dn (
ng = ≅ n '1+ *
ν dn & n d ν )
1−
n dν

dn g
we find that > 0 . That is, δτ d > 0 for δν > 0 . This implies that, after the propagation,
dν €
the blue components of the pulse suffer a greater time delay than the red ones.
dn g
In the case of negative dispersion, Dν < 0 , < 0 and hence δτ d > 0 for δν < 0 :
€ € dν

the red components suffer a greater time delay.

€ € €

1.4 Image Formation and Processing

Fourier optics is particularly useful for analyzing optical imaging systems, as it


provides powerful tools for dealing with image formation and processing. We start with the
formulation of simple imaging elements and then proceed to elaborate more complex image
treatments using Fourier analysis.

1.4.1 Thin Optical Lens

Consider a thin lens made of transparent material of refractive index n and positioned
with its axis oriented in the z direction (see Fig. 1.12). A plane wave of free space wave
number k0 ≡ ω /c = 2π / λ0 , propagating along z and impinging on the lens, will suffer a phase
delay after passing through the lens, depending on the position ( x, y ) in the lens plane. This
phase delay can be written as

φ ( x, y ) = k 0 nΔ ( x, y ) + k 0 [Δ 0 − Δ ( x, y )]
  €  
lens material free space

€ 25
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

where Δ ( x, y ) is the lens thickness and Δ 0 = Δ ( 0, 0 ) , or as

φ ( x, y ) = k 0Δ 0 + k 0Δ ( x, y ) ( n −1)

An incident optical field U in ( x, y ) then yields a transmitted optical field
€ U out ( x, y ) = t ( x, y ) U in ( x, y )

Figure 1.12: Parameters of a thin optical lens.

where the transmission function of the thin lens is given by

t ( x, y ) = e iφ ( x,y) = e ik0 Δ 0 e ik0 Δ ( x,y)( n−1)

(Note that the term “thin lens” refers to a lens in which the input and output planes ( x, y )
practically coincide.)
The thickness€ of the thin lens can be calculated in terms of the radius of curvature R
of each of its facets (see Fig. 1.13). We will adopt the convention according to which R > 0 (

R < 0 ) for a convex (concave) surface with respect to the direction of incidence. For a lens
with one flat surface and one convex surface, we find

€ {
Δ ( x, y ) = Δ 0 − R − R 2 − ( x 2 + y 2 ) }

26
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

Figure 1.13: Geometry for calculating the thickness of a thin lens.

Figure 1.14: A composite thin lens.

whereas for a lens with one flat surface and one concave surface we have

{
Δ ( x, y ) = Δ 0 − −R − R 2 − ( x 2 + y 2 ) }
For a composite lens, consisting of one convex and one concave surfaces of radii R1 and R2 ,
(see Fig. 1.14), we obtain

* $ 2 2 '1/ 2 - * $ 2 2 '1/ 2 -
x + y € x + y
Δ ( x, y ) = Δ1 ( x, y ) + Δ 2 ( x, y ) = Δ , / , /
0 − R1,1− &%1− R 2 )( / + R2 ,1− &%1− R 2 )( /
Δ 01 +Δ 02 + 1 . + 2 .

For paraxial rays, we may expand the terms in parentheses:



# x 2 + y 2 &1/ 2 x2 + y2
%1− ( ≈ 1−
$ R2 ' 2R 2

27

E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

which yields
% x2 + y2 ( % x2 + y2 ( x2 + y2 % 1 1(
Δ ( x, y ) ≅ Δ 0 − R1' 2 * + R2 ' 2 * = Δ 0 − ' − *
& 2R1 ) & 2R2 ) 2 & R1 R2 )

We thus obtain the transmission function for a thin lens in the paraxial approximation:
€ $ x 2 +y 2 ' $ 1 1 '
−ik0 ( n−1)& )& − )
% 2 ( % R1 R 2 (
t ( x, y ) = e ik0 Δ 0 e ik0 Δ 0 ( n−1)e

The focal length f of a thin lens, well know in geometrical optics, is written in terms
of its shape and refractive index as

$1 1'
f −1 ≡ ( n −1)& − )
% R1 R2 (

Figure 1.15: Converging (left) and diverging (right) lens.

(Note that f can be positive or negative, depending on the type of the surfaces.) The
transmission function is thus expressed in terms of the focal length as
−ik0 2 2
ik0 Δ 0 n 2f
(
x +y )
t ( x, y ) = e e

For an incident plane wave with unity amplitude U m ( x, y ) = 1, the wave just after the
lens is given by

k 2 2
ik0 Δ 0 n
−i
2f
(
x +y )
U out ( x, y ) = e € e

This is just the paraxial approximation for a converging ( f > 0) or a diverging ( f < 0)
spherical wave whose origin is at a distance z = + f or z = − f from the lens (see Fig. 1.15).

€ €
€ €

28
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

1.4.2 Thin Lens as a Fourier Transformer

We consider now the effect of placing a lens just behind a “transparency” t o ( x, y ) ,


observing the resulting image on a screen placed a distance f beyond the plane of the lens (see
Fig. 1.16). The transparency can be considered as an object that emanates light, and under
proper conditions the lens can form an image of such object. For an incident plane wave of
amplitude A, the optical field immediately after the transparency is given by€

U l ( x, y ) = At o ( x, y )

In the paraxial approximation, we can write the field just after the lens (neglecting the
constant phase factor) as

k0 2 2
−i
2f
(x +y )
U'l ( x, y ) = U l ( x, y )e

To propagate the field along the z-axis starting from the plane of the lens (and
transparency), we use the paraxial transfer function and obtain

Figure 1.16: A transparency placed in front of a thin lens.

k k i2 π
e f i 2z f ( x f ) ( )
ikz 2
+y f 2 ∞ i x 2 +y 2 − ( xx f +yy f )
2z f λz f
U( x f , y f ) ≅ e ∫∫ dxdy U'l ( x, y )e e
iλz f −∞

Substituting the expression for U'l we get (with z f = f )


€ k ∞ i2 π
e ikf i 2 f ( x f )
2
+y f 2 − ( xx f +yy f )
(
U xf ,yf

) ≅
iλf
e ∫∫ dxdy U l ( x, y ) e λf

€ −∞

This expression is nothing but a 2D Fourier transform of the function, except for the pre-
phase factor.€Putting explicitly the spatial frequencies

29
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

xf yf & 1 )
νx = ;ν y = (ν x,y ≅ θ x,y +
λf λf ' λ *

We can rewrite the result as



ikf k 2 2
e i
2f
(
x f +y f ) ∞
−i2 π (ν x x +ν y y )
U (x f , y f )≅ A e ∫∫ dxdyto (x, y ) e
iλf −∞
ikf k 2 2
e i
2f
(
x f +y f )
=A e F {t o (x, y )}
iλf

It is worth noting that the measured intensity distribution at the focal plane, which is
2

€ to U ( x f , y f ) , is indistinguishable from the intensity distribution of the exact


proportional
Fourier transform of t o ( x, y ) .

Consider next the effect of placing the transparency (or an object) at a distance do in

front of the lens, (see Fig. 1.17). In this case we need first to “propagate” the optical field
emanating € from the object to the plane of the lens. We will indicate the Fourier transform of
the field at the object point by

Figure 1.17: An object placed in front of a thin lens.

Fo (ν x ,ν y ) = F { At o ( x, y )}

where t o ( x, y ) represents the field at the object (or the field transmitted by the transparency).
The field arriving at the lens (before transmission), U l ( x, y ), can be found by first calculating

its Fourier transform

Fl (ν x ,ν y ) = F {U l ( x, y )}

using the free space transfer function. In the paraxial approximation, we have
€ ( )
−iπλ ν x 2 +ν y 2 z
H (ν x ,ν y ) ≈ e ikz = e ikz e


30
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

and hence, ignoring constant phase, we obtain

(
−iπλ ν x 2 +ν y 2 d 0 )
( )
Fl ν x , ν y = F0 ν x , ν y e ( )
After propagation from the lens to the focal plane z = f , we now get:

k ∞
A i 2 f (x f )
2
+y f 2
U f (x f , y f ) =
(
−iπλ ν x 2 +ν y 2 d o) −i2 π ( x oν x +y oν y )

iλf
e e ∫∫ t ( x , y ) dx dy
o o o o o e
€ −∞

The phase of the term preceding the integral can be rewritten


€ k k & )
i
2f
( 2 2
) 2 2
x f + y f − iπλ ν x + ν y do = i
2f
(
2
xf + yf
2
) ( )('1− df +* o

and we finally obtain


€ k $ d '
A i 2 f (x f )
2
+y f 2 &1− o ) ∞
% f ( −i2 π ( x oν x +y oν y )
U f (x f , y f ) = e ∫∫ t ( x , y ) dx dy
o o o o o e
iλf −∞

A particular case of great interest is when the object is placed at the focal plane of the
lens, i.e., when

do = f

We then have

A ∞
−i2 π (x oν x +y oν y )
U f (x f , y f )= ∫∫ t (x , y ) dx dy
o o o o o e
iλf −∞

A
= F {t o (x o , y o )}
iλf

That is, the field at the back focal plane of a lens is proportional to the Fourier transform of
the field at the front focal length of the same lens. This shows how a thin lens can be used to
€ 2D Fourier transform of a given function defined in a plane.
“calculate” the

1.4.3 Image Formation

Consider an object placed at a distance do in front of a positive lens (see Fig. 1.18).
We wish to find the conditions for forming an image at a distance di from the lens. We
assume that the object is illuminated with monochromatic light.
For unit amplitude plane wave emanating from the object, the optical field at the

image plane can be written in terms of the impulse function h ( x i , y i ;x 0 , y 0 ) :


31
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics


U i ( x i, y i ) = ∫∫ dx 0 dy 0 h ( x i , y i ;x 0 , y 0 ) U 0 ( x 0 , y 0 )
−∞

The condition for imaging can be defined as the case where the impulse function takes the
form

h ( x i , y i ;x 0 , y 0 ) ≅ Kδ ( x i ± Mx 0 , y i ± My 0 )

where M is the magnification of the system.


€ a single “pixel” at ( x , y ) in the object plane, represented by a δ -
Starting with 0 0
function centered at ( x 0 , y 0 ) , we can write the field at the plane of the lens, in front of it, as

k0 €
1 i 2d 0 [ ( x−x 0 ) ]
2 2
+( y−y 0 )
U l€( x, y ) = e
€ iλd0

After propagation through the thin lens, we have


€ k 2 2
−i
2f
(x +y )
U l ' ( x, y ) = U l ( x, y ) P ( x, y ) e

Figure 1.18: Imaging of an object placed in front of a thin lens.

where P is a pupil function. By propagating from the lens plane to the image plane, we
obtain (omitting a phase factor)
k
1 ∞ ∞ i [ ( xi −x) 2
+( y i −y )
2
]
U ( x i, y i ) = ∫ ∫
iλdi −∞ −∞
U l ' ( x, y ) e 2d i dxdy

and thus the full impulse response is given by



32
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

k
1 i (x i 2
+y i 2 ) i dk (x 0
2
+y 0 2 )
h(x i , y i ;x 0, y 0 ) = 2
e di e 0

λ d0 d i
k ⎛ 1 1 1 ⎞ ⎡⎛ x x ⎞ ⎛ y y ⎞ ⎤
∞ ∞ i ⎜ + − ⎟ x 2 +y 2
2 ⎝ d 0 d i f ⎠
( ) −ik⎢⎣⎜⎝ d 00 + d ii ⎟⎠ x +⎜⎝ d00 + dii ⎟⎠y ⎥⎦
×∫ ∫ dxdy P (x, y ) e e
−∞ −∞

For incoherent illumination of the object, and for image construction by intensity, the
two phase factors can be dropped, yielding

k' 1 1 1* -' x x * 'y y * 0
i ) + − , (x 2
+y 2 ) −ik /) 0 + i , x +) 0 + i , y 2
1 ∞ ∞ 2 d d f .( d 0 d i + ( d 0 d i + 1
h ( x i , y i ;x 0 , y 0 ) ≅ 2 ∫ ∫
λ d0 di −∞ −∞
dxdy P ( x, y ) e ( 0 i + e

If we restrict ourselves to the image plane, defined such that


€ 1 1 1
+ =
d0 d i f

(recall the expression from geometrical optics!), we find


€ -' x
0 xi * ' y0 yi * 0
1 ∞ ∞ −ik /) + , x +) + ,y 2
.( d 0 d i + ( d0 di + 1
h ( x i , y i ;x 0 , y 0 ) ≅ 2 ∫ ∫
λ d0 di −∞ −∞
dxdy P ( x, y ) e

Moreover, by defining the magnification factor as M ≡ di /d0 , we can rewrite the impulse
function as

ik
1 ∞ ∞€ − [ ( x i +Mx 0 ) x +( y i +My 0 ) y ]
h ( x i , y i ;x 0 , y 0 ) ≅ 2 ∫ ∫
λ d0 di −∞ −∞
dxdy P ( x, y ) e d i

With this impulse function, the optical field in the image plane is reconstructed as
€ ∞
U i ( x i, y i ) = ∫∫ dx 0 dy 0 h ( x i , y i ;x 0 , y 0 ) U 0 ( x 0 , y 0 )
−∞

To estimate the physical limits of the optical imaging process, it is worth considering
separately the geometrical optics case and the effects of diffraction.

Geometrical optics limit

The special limiting case of geometrical optics is attained when the relevant system
dimensions are large compared to the optical wavelength, i.e., λ → 0 . We define the scaled
spatial variables

x y
x˜ = ; y˜ = €
λdi λdi

33

E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

with which the impulse function becomes



−i2 π [ ( x i +Mx 0 ) x˜ +( y i +My 0 ) y˜ ]
h ( x i , y i ;x 0 , y 0 ) ≅ M ∫∫ d˜xd˜y P ( λdi x˜ , λdi y˜ ) e
−∞

When λ → 0 , the range of ( x˜ , y˜ ) over which P = 1 grows without bound, and thus P →1.
We then have

€ 1 $ xi y '
x i , y i ;x 0 , y 0 ) → Mδ€( x i + Mx 0 ;y i + My 0 ) =
h (€ δ & + x 0 , €i + y 0 )
M %M M (

and the optical field at the image is written


€ ∞
% x
1 y (
U i ( x i, y i ) = ∫∫ dx 0 dy 0 h ( x i , y i ;x 0 , y 0 ) U 0 ( x 0 , y 0 ) = U 0 ' − i ,− i *
M & M M)
−∞

This represents the simple correspondence of points in the object and the image plane, taking
in consideration the magnification, as well known from geometrical optics.

Including diffraction effects

To better express the effects of the wave nature of light, we define the variables

x˜ 0 ≡ −Mx 0 ; y˜ 0 ≡ −My 0
with which the impulse function becomes

−i2 π [ ( x i − x˜ 0 ) x˜ +( y i − y˜ 0 ) y˜ ]
h ( x i , y i ;x 0€, y 0 ) = M ∫∫ d˜xd˜y P ( λdi x˜ , λdi y˜ ) e d˜xd˜y
−∞

Notice that this expression is space invariant, i.e., it depends only on the relative distances
1
x i −€x˜ 0 and y i − y˜ 0. Introducing now the function h˜ = h , we get
M

+ 1 % x˜ y˜ (.
U i ( x i, y i ) = ∫∫ d˜x 0 d˜y 0 h˜ ( x i − x˜ 0 , y i − y˜ 0 ) - U 0 ' − 0 ,− 0 *0
€ € , M & M M )/
−∞ €

That is, the field at the image plane is in fact the convolution of h˜ with the field obtained in
the geometrical optics limit.
€To express this more explicitly, it is useful to define the “geometrical optics image”:


1 # xi yi &
U g ( x i , y i ) = U 0 % − ,− (
M $ M M'

and obtain

34
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

U i ( x i , y i ) = h˜ ( x i , y i ) ∗U g ( x i , y i )

where

€ ∞
h˜ ( x i , y i ) = ∫∫ P (λd x˜, λd y˜ ) e
i i
−i2 π ( x i x˜ +y i y˜ )
d˜xd˜y
−∞

This latter function contains the effects of diffraction through the aperture P defining the
effective size of the lens. Increasing this size reduces the diffracted beam width and hence
€ resolution of the imaging system.
increases the spatial

35
E. Kapon/Optics III (2015/2016)/Chapter 1: Fourier Optics

1.5 Summary

- Spatial frequencies: θ x,y = sin−1 ( λν x,y )

i2 π (ν x x + ν y y )
- Angular spectrum of plane waves: U ( x, y,0) = Ae

F (ν x ,ν y ;z)
- Transfer function: H (ν x ,ν y ;z) ≡ = e iκz
€ F (ν x ,ν y ;0)

(
κ ≡ k 1− λ2 ν x 2 + ν y 2 )
€ $ ikz i k ( x 0 2 +y 0 2 ) (
& e e 2z &
- Fraunhofer diffraction:

U ( x 0, y 0 ) ≅ % ) F {U ( x1, y1 ) }
& iλz &
' *

- Impulse response function: [


h ( x, y ) = F −1 H (ν x ,ν y ) ]

−α dβ i 2 d 2β
z if z f z
2 dν 2
-Transfer function for pulse propagation: H( f ) ≈ e 2
e dν
e

k 2 2
ik0 Δ 0 n
−i
2f
(x +y )
- Thin lens: U out ( x, y ) = e e

A
- Fourier transform by a thin lens U f (x f , y f ) = F {t o ( x o , y o )}
iλf

1 # xi yi &
-Geometrical optics imaging: U g ( x i, y i ) = U 0 % − ,− (
M $ M M'

- Diffraction limited imaging: U i ( x i , y i ) = h˜ ( x i , y i ) ∗U g ( x i , y i )
€ ∞
h˜ ( x i , y i ) = ∫∫ P (λd x˜, λd y˜ ) e
i i
−i2 π ( x i x˜ +y i y˜ )
d˜xd˜y
€ −∞

36

You might also like