Professional Documents
Culture Documents
Mahajan
Mahajan, Virendra N.
Optical imaging and aberrations / Virendra N. Mahajan.
p. cm.
Includes bibliographical references and index.
ISBN 0-8194-4135-X
1. Aberrations. 2. Imaging systems. 3. Geometrical optics.
I. Title.
QC671.M36 1998
621.36—DC21
97-7721
CIP
Published by
SPIE
P.O. Box 10
Bellingham, Washington 98227-0010 USA
Phone: +1 360 676 3290
Fax: +1 360 647 1445
Email: books@spie.org
Web: http://spie.org
The content of this book reflects the work and thought of the author(s).
Every effort has been made to publish reliable and accurate information herein,
but the publisher is not responsible for the validity of the information or for any
outcomes resulting from reliance thereon.
The author was an adjunct professor at the University of Southern California when the book was
first published. He has been an adjunct professor with the University of Arizona since 2004.
In loving memory of my parents:
Three years ago Vini Mahajan published a book on the subject of Geometrical
Images in the presence of aberrations. In this book, Mahajan extends this work to include
the effect of wave optics. He continues his thorough tutorial on image formation with a
detailed look at the approaches to calculating the form of images. Anyone interested in
understanding the methods of predicting the light distribution to be expected in real
imaging situations will find this book of interest.
The book begins with an exhaustive development of the basics of diffraction image
formation. Mahajan covers the issues associated with the calculation of point-spread
functions and discusses the accuracy of such calculations. He introduces the Optical
Transfer Function as the Transform of the Point Spread Function and reviews the
procedures involved in calculating the OTF. Asymptotic and approximate evaluations of
the OTF are included, as are several examples throughout the book.
These approaches are then applied to some real examples of circular and annular
apertures. In this discussion, Mahajan carries out in detail many of the classical
computations for various image descriptors. This is a topic that is generally treated only
lightly in most texts on the subject. Such issues as edge response and line spread function
and encircled power are carefully considered. A good discussion of optimal balancing of
aberrations is also provided. The treatment of aberration balancing and tolerances in
annular pupils is unique in its completeness in this book.
vii
TABLE OF CONTENTS
ix
1.10 Hopkins Ratio ....................................................................................................................... 59
1.11 Line- and Edge-Spread Functions (LSF and ESF) ........................................................... 61
1.11.1 Line-Spread Function ............................................................................................. 61
1.11.2 Edge-Spread Function ............................................................................................ 64
1.11.3 LSF and ESF in Terms of OTF .............................................................................. 64
1.12 Shift-Invariant Imaging of a Coherent Object .................................................................. 67
1.12.1 Coherent Point-Spread Function ............................................................................ 67
1.12.2 Coherent Transfer Function ................................................................................... 69
1.13 Summary of Theorems......................................................................................................... 71
Appendix A: Fourier Transform Definitions ............................................................................... 74
Appendix B: Some Frequently Used Integrals ............................................................................. 75
References ........................................................................................................................................ 76
Problems........................................................................................................................................... 78
x
2.7 Symmetry Properties of an Aberrated PSF ..................................................................... 136
2.7.1 General Theory ..................................................................................................... 137
2.7.2 Symmetry About the Gaussian Image Plane ........................................................ 138
2.7.3 Symmetry of Axial Irradiance.............................................................................. 141
2.8 PSFs for Primary Aberrations .......................................................................................... 142
2.8.1 Defocus ................................................................................................................. 142
2.8.2 Spherical Aberration Combined with Defocus..................................................... 142
2.8.3 Astigmatism Combined with Defocus ................................................................. 144
2.8.4 Coma..................................................................................................................... 148
2.8.5 2-D PSFs .............................................................................................................. 150
2.8.6 Comparison of Diffraction and Geometrical PSFs .............................................. 157
2.9 Line of Sight of an Aberrated System .............................................................................. 159
2.9.1 PSF and its Centroid ............................................................................................. 159
2.9.2 Numerical Results ................................................................................................ 162
2.9.2.1 Wavefront Tilt ...................................................................................... 162
2.9.2.2 Primary Coma ...................................................................................... 162
2.9.2.3 Secondary Coma .................................................................................. 165
2.9.3 Comments ............................................................................................................. 168
2.10 Diffraction OTF for Primary Aberrations ....................................................................... 169
2.10.1 General Relations ................................................................................................. 169
2.10.2 Defocus ................................................................................................................. 172
2.10.3 Spherical Aberration............................................................................................. 173
2.10.4 Astigmatism ......................................................................................................... 173
2.10.5 Coma..................................................................................................................... 175
2.11 Hopkins Ratio ..................................................................................................................... 182
2.11.1 Tolerance for Primary Aberrations ...................................................................... 182
2.11.2 Defocus ................................................................................................................. 182
2.11.3 Hopkins Ratio in Terms of Variance of Aberration Difference Function............ 185
2.11.4 Variance of Aberration Difference Function for Primary Aberrations ................ 186
2.12 Geometrical OTF................................................................................................................ 187
2.12.1 General Relations ................................................................................................. 188
2.12.2 Radially Symmetric Aberrations .......................................................................... 189
2.12.3 Defocus ................................................................................................................. 189
2.12.4 Spherical Aberration Combined with Defocus..................................................... 190
2.12.5 Astigmatism Combined with Defocus ................................................................. 190
2.12.6 Coma..................................................................................................................... 191
2.13 Incoherent Line- and Edge-Spread Functions ................................................................. 191
2.13.1 Theory .................................................................................................................. 192
2.13.1.1 LSF From PSF .................................................................................. 192
2.13.1.2 LSF From Pupil Function ................................................................. 192
2.13.1.3 Struve Ratio and Aberration Tolerances ........................................... 193
2.13.1.4 LSF From OTF ................................................................................. 196
2.13.1.5 ESF From OTF ................................................................................. 198
xi
2.13.2 Numerical Results ................................................................................................ 199
2.14 Miscellaneous Topics ......................................................................................................... 205
2.14.1 Polychromatic PSF ............................................................................................... 205
2.14.2 Polychromatic OTF .............................................................................................. 208
2.14.3 Image of an Incoherent Disc................................................................................. 209
2.14.4 Pinhole Camera .................................................................................................... 218
2.15 Coherent Imaging............................................................................................................... 222
2.15.1 Coherent Spread Function .................................................................................... 222
2.15.2 Coherent Transfer Function ................................................................................. 223
2.15.3 Coherent LSF ....................................................................................................... 224
2.15.4 Coherent ESF ....................................................................................................... 229
2.15.5 Image of a Coherent Disc ..................................................................................... 234
2.15.6 Use of a Lens for Obtaining Fourier Transforms ................................................. 238
2.15.7 Comparison of Coherent and Incoherent Imaging ............................................... 241
References ...................................................................................................................................... 253
Problems......................................................................................................................................... 257
xii
3.9 Line of Sight of an Aberrated System .............................................................................. 322
3.9.1 PSF and its Centroid ............................................................................................. 322
3.9.2 Numerical Results ................................................................................................ 323
3.9.2.1 Wavefront Tilt ...................................................................................... 323
3.9.2.2 Primary Coma ...................................................................................... 324
3.9.2.3 Secondary Coma .................................................................................. 327
References ...................................................................................................................................... 330
Problems......................................................................................................................................... 331
xiii
4.5 Line of Sight of an Aberrated System .............................................................................. 379
4.5.1 PSF and its Centroid ............................................................................................. 380
4.5.2 Numerical Results ................................................................................................ 380
4.5.2.1 Wavefront Tilt ...................................................................................... 380
4.5.2.2 Primary Coma ...................................................................................... 381
4.5.2.3 Secondary Coma .................................................................................. 382
4.6 Summary ............................................................................................................................. 382
References ...................................................................................................................................... 385
Problems......................................................................................................................................... 386
xiv
PREFACE
In Part I of this book, we discussed imaging based on ray geometrical optics. The
aberration-free image of an object according to it is the exact replica of the object, except
for its magnification. The aberration-free image of a point object is also a point. In reality,
however, the image obtained is not a point. Because of diffraction of the object wave at
the aperture stop, or equivalently, the exit pupil of the imaging system, the actual
aberration-free image for a circular exit pupil is a light patch surrounded by dark and light
rings. Geometrical optics is still assumed to hold from the object to the exit pupil in that
rays are traced through the system to determine the shape of the pupil and the aberration
across it. The determination of the characteristics of the diffraction image of an object
formed by an aberrated system is the subject discussed in Part II. The emphasis of this
part is on the numerical results on the effects of aberrations on a diffraction image and not
on the formalism, exposition, or a critique of the variety of diffraction theories proposed
over the years. It is a compilation of the works of masters like Nijboer, Hopkins, Barakat,
and Fried with a sprinkling of my own work.
Chapter 2 on systems with circular pupils starts with the aberration-free PSF and its
encircled and ensquared powers. The effects of primary aberrations on its Strehl ratio are
discussed and aberration tolerances are obtained. Balanced primary aberrations are
xv
considered and identified with Zernike circle polynomials. Focused and collimated beams
are discussed and the concept of near- and far-field distances is introduced. Aberrated
PSFs and their symmetry properties are discussed, and a brief comparison is made with
their counterparts in geometrical optics. The line of sight of an aberrated system,
identified with the centroid of the PSF, is determined for primary aberrations. The OTF
for these aberrations is also discussed, phase contrast reversal is explained, and aberration
tolerances for a certain value of Hopkins ratio are given. Expressions for the geometrical
OTF for primary aberrations are also given. Both incoherent and coherent line- and edge-
spread functions are discussed, and aberration tolerances for a certain value of the Struve
ratio are given. A brief comparison of incoherent and coherent imaging is given with
special reference to the Rayleigh criterion of resolution. The Fourier-transforming
property of wave propagation is illustrated in altering the image of an object by spatial
filtering in the Fourier-transform plane.
Systems with annular pupils are given a cursory look at best in books where imaging
is discussed. Our Chapter 3 is written in a manner similar to Chapter 2 where, for
example, the balanced aberrations are identified with the Zernike annular polynomials.
Although the propagation of Gaussian beams is discussed in books on lasers, their
treatment is generally limited to the weakly truncated aberration-free beams. In Chapter
4, we consider the effect of arbitrary truncation of beams with and without aberrations.
The balanced aberrations in this case are identified with the Zernike-Gauss polynomials.
It is shown, for example, that the pupil radius must be at least three times the beam radius
in order to neglect beam truncation without significant error.
Finally, random aberrations are considered in Chapter 5. The effect of random image
motion is considered first, and expressions and numerical values of time-averaged Strehl
ratio, PSF, and encircled power are given for systems with circular and annular pupils.
The random aberrations introduced by atmospheric turbulence when a wave propagates
through it, as in astronomical observations by a ground-based telescope, are discussed;
and expressions for time-averaged PSF and OTF are obtained. The aberration function for
Kolmogorov turbulence is expanded in terms of the Zernike polynomials, and auto-
correlation and cross-correlation of the expansion coefficients are given. The atmospheric
coherence length is defined, and it is shown that the resolution of a telescope cannot
exceed that of an aberration-free telescope of this diameter. Both the short- and long-
exposure images are considered.
As in Part I, each chapter ends with a set of problems. It is hoped that they will
acquaint the reader with application of the theory in terms of some practical examples.
References for addtional reading are given after the Bibliography. These references
represent the author’s collection as the editor of Milestone Series 74 entitled Effects of
Aberrations in Optical Imaging, published by SPIE Press in 1993.
xvi
ACKNOWLEDGMENTS
It is a great pleasure to acknowledge the generous support I have received over the
years from my employer, The Aerospace Corporation, in preparing this book. My special
thanks go to the senior vice president Mr. John Parsons, for his continuous interest and
encouragement in this endeavor. I thank Mr. John Hoyem for preparing the figures and
Mr. Victor Onouye for some figures as well as the final composition. My thanks also go
to Dr. Rich Boucher for computer generating the 2-D PSFs and Mr. Yunsong Huang for
numerical analysis and computer plotting. The Sanskrit verse on p. was provided by
Professor Sally Sutherland of the University of California at Berkeley.
I thank my sister Pushpa and brother Devinder for encouraging me to study phyiscs. I
can not say enough about the constant support I have received from my wife Shashi over
the many years it has taken me to complete this two-part book. I dedicate this book to my
departed parents who brought and nourished me in this world.
Once again, it has been a pleasure to work with the SPIE Press staff in bringing this
book to its completion. My thanks to them, especially, Sharon Streams, for their
cooperative spirit and quality support.
xvii
SYMBOLS AND NOTATION
xix
Anantaratnaprabhavasya yasya himam
. na saubhagyavilopi jatam
Eko hi doso
. gunasannipate
. ˙ .
nimajjatindoh. kiranesvivankah
.
The snow does not diminish the beauty of the Himalayan mountains
which are the source of countless gems. Indeed, one flaw is lost
among a host of virtues, as the moon’s dark spot is lost among its rays.
IMAGE FORMATION
1
2 IMAGE FORMATION
Geometrical optics is assumed to hold from the point object to the exit pupil in that
rays are traced through the system to determine the shape of the pupil and the aberration
across it for the point object under consideration, as discussed in Section 3.2 of Part I.
The amplitude associated with a ray is obtained by consideration of the transmission and
reflection characteristics of the imaging elements of the system. In many applications, the
amplitude variation across the pupil is negligible. There are instances, however, when the
transmission across the pupil is varied to alter certain characteristics of the diffraction
3
4 IMAGE FORMATION
PSF. The pupil is said to be apodized in these cases. Another important example of
variable amplitude across the pupil is that of laser beams for which the amplitude has a
Gaussian form. The irradiance distribution of the diffraction image of an incoherent
object is obtained by adding the irradiance distribution of the images of its infinitesimal
elements. In particular, we show that the diffraction image of an isoplanatic object, i.e.,
one for which the pupil function of the system is independent of the position of a point on
it, is equal to the convolution of its Gaussian image and the (diffraction) PSF of the
system.
We also discuss the optical transfer function of an imaging system, which is equal to
a Fourier transform of its PSF. It describes the amplitude and phase of the sinusoidal
image of a sinusoidal object as a function of its spatial frequency. Thus, it relates the
spatial frequency spectrum of the object to that of the image. We show, for example, that
the slope of its real part at the origin is independent of aberrations. We also compare the
diffraction OTF with the corresponding OTF based on geometrical optics. The asymptotic
behavior and centroid of an aberrated point-spread function are related to the slope of the
real and imaginary parts of the OTF at the origin, respectively. It is shown that the
centroid of the diffraction PSF is identical with that of the geometrical PSF.
Next, image quality based on the Strehl ratio of an aberrated imaging system is
discussed, and approximate expressions in terms of its aberration variance are obtained. A
brief discussion of how the Strehl ratio of a system may be determined is also given. The
image quality based on Hopkins ratio in the spatial frequency domain is also discussed.
Whereas the Strehl ratio represents the ratio of the central value of the irradiance
distribution of the aberrated image of a point object and the corresponding aberration-free
value, the Hopkins ratio represents the ratio of the contrasts of the aberrated and
aberration-free images of a sinusoidal object of a certain spatial frequency. The Strehl and
Hopkins ratios provide simple means of determining the aberration tolerances. The line-
spread function (i.e, the image of a line object) and the edge-spread function (i.e., the
image of an edge or a step object) are discussed in terms of the PSF and related to the
OTF. The Struve ratio of a system, representing the ratio of the aberrated central value of
the LSF and the corresponding aberration-free value, is also considered and aberration
tolerances are derived from it.
coherent spread function, and its incoherent transfer function is equal to the
autocorrelation of its coherent transfer function.
The fundamental relations derived in this chapter are stated as theorems and used in
the succeeding chapters to obtain some practical results for imaging systems with
circular, annular, and Gaussian pupils.
2 2 2
2 = 2 + 2 + . (1-2)
x y z 2
r r
In Figure 1-1, r is used in the z = 0 plane to distinguish it from the r in the z plane, as
the two planes become the source and observation planes, respectively.
6 IMAGE FORMATION
x
x
P (r; 0)
l
P (r; z)
r
l0
O
y r
z = 0 Plane
z
O z
y z - Plane
Figure 1-1. Geometry of wave propagation for determining the complex amplitude
r r
U ( r ; z ) in a z plane from its knowledge U ( r ; 0) in the z = 0 plane. O P = l 0 ,
P P = l and O O = z .
r r
To relate U ( r ; z ) to U ( r ; 0) , we decompose both of them into plane waves. Thus,
for example, we write
r r r r r
U ( r ; 0) = A( v ; 0) exp (
2 ir v ) d v , (1-3)
r
where A( v ; 0) is the amplitude of a plane wave propagating with direction cosines
r
(,
, ) such that the spatial frequency v is given by
r 1
v = ( ,
) (1-4)
and
2 +
2 + 2 = 1 . (1-5)
r r
r(
A v ; 0) is also referred to as the spectral component of U ( r ; 0) with a spatial frequency
v . Written in the form A( ,
; 0) , it is also called the angular spectrum of
r
U ( r ; 0) . [Equation (1-4) may be obtained by comparing
r r the exponentialrin Eq. (1-3) with
[(
the standard form of a plane wave exp i t
k r , where and k are its angular
frequency and propagation vector, and t is time.]
)]
r r
Multiplying both sides of Eq. (1-3) by exp (2ir v ) and integrating, we obtain
r r r r r r r r r r
U ( r ; 0) exp (2 ir v ) dr = A( v ; 0) d v exp 2 ir ( v
v ) dr [ ]
r r r r
= A( v ; 0) ( v
v ) d v
r
= A( v ; 0) ,
or
r r r r r
A( v ; 0) = U ( r ; 0) exp (2 ir v ) dr , (1-6)
1.2 Rayleigh-Sommerfeld Theory of Diffraction and Huygens-Fresnel Principle 7
r
where () is a Dirac delta function. It is clear from Eqs. (1-3) and (1-6) that U ( r ; 0) and
r r
A( v ; 0) form a 2-D Fourier-transform pair; U ( r ; 0) is the inverse Fourier transform of
r r r
A( v ; 0) and A( v ; 0) is the Fourier transform of U ( r ; 0) . (For a definition of the Fourier
transform, see the Appendix.) Similarly, we may write
r r r r r
U ( r ; z ) = A( v ; z ) exp (
2ir v ) d v , (1-7)
r r
where A( v ; z ) is the amplitude of a plane-wave component of spatial frequency v in the
z plane given by
r r r r r
A( v ; z ) = U ( r ; z ) exp (2ir v ) dr . (1-8)
r
Since U ( r ; z ) satisfies the wave equation, which is linear, each component of the
plane waves that comprise it must also satisfy it. Hence, substituting
r r r
A( v ; z ) exp (
2ir v ) for it into Eq. (1-1), we obtain
r
r 2 A( v ; z )
k A( v ; z ) +
2 2
= 0 . (1-9)
z 2
so that
r r r r
exp (ik z ) = h( r ; z ) exp (2 ir v ) dr . (1-13)
where we have used Eq. (1-3). By change of variables, Eq. (1-14) containing the
convolution integral can also be written
r r r r r
U ( r ; z ) = h( r
r ; z ) U ( r ; 0) dr . (1-15)
r
The integrals in Eqs. (1-14) and (1-15) giving the wave field U ( r ; z ) atthe z plane
8 IMAGE FORMATION
r r
represent a convolution of h( r ; z ) and the wave field U ( r ; 0) in the z = 0 plane.
r
If a point source of unit amplitude is placed at a point ro in the z = 0 plane, i.e., if
r r r
U ( r ; 0) = ( r
ro ) , (1-16)
r
then Eq. (1-15) shows that the complex amplitude at a point r in the z plane is given by
r r r
U ( r ; z ) = h( r
ro ; z ) . (1-17)
r r
Thus, h( r ; z ) represents the complex amplitude at a point r in the z plane due to a point
r
source of unit amplitude located at r = 0 in the z = 0 plane. It is called the complex
amplitude point-spread function (or impulse response) of free space. Carrying out the
integration in Eq. (1-12), it can be shown that4
r 1 1 z exp (ikl0 )
h( r ; z ) =
i , (1-18)
kl0 l0 l0
where
( )
1/ 2
l0 = z 2 + r 2 (1-19)
is the distance OP between the origin in the z = 0 plane and the observation point
r
( r ; z) . Equation (1-18) is a mathematical expression for a Huygens’ spherical wavelet
diverging from the point source, and it describes the complex amplitude point-spread
r r
function of free space. We note from Eq. (1-12) that h( r ; z ) (r ) as z 0 , i.e., it
becomes the point source, as expected. Accordingly, Eq. (1-15) is a mathematical
description of Huygens-Fresnel principle, namely, that the complex amplitude in the z
r r
plane is a linear superposition of Huygens’ secondary spherical wavelets h( r
r ; z )
r
weighted by the amplitudes U ( r ; 0) of the wave where they originate (Theorem 1). The
diffracted wave field described by Eq. (1-15) is shift invariant (or isoplanatic) in that a
r r r r
spherical wavelet at a point r due to a source point at r depends on r
r , i.e., the
form of a spherical wavelet is independent of the location of its origin in the z = 0 plane,
except for a shift in the center of the distribution.
[ r r 2
]
12
l = z2 + r
r (1-21a)
r r 1 r r 4
= z+
1 2
2z
( )
r + r 2
2 r r
3 r
r + ...
8z
(1-21b)
1.2 Rayleigh-Sommerfeld Theory of Diffraction and Huygens-Fresnel Principle 9
r r
is the distance P P between a source point (r ; 0) and the observation point (r ; z ) .
Equation (1-20) is the Rayleigh-Sommerfeld formula describing propagation of a wave
from one plane to another.
r
i r2 ikr 2 r 2 i r r r
U (r ; z) = exp ik z + exp U ( r ; 0) exp
z r r dr ,
z 2 z 2 z
r r 4
for z 3 >> k r
r max 8 ( Fresnel) , (1-22a)
and
r
i r2 r 2 i r r r
U (r ; z) = exp ik z + U ( r ; 0) exp
r r dr ,
z 2 z z
Thus, depending on the value of the distance z relative to the extent of the regions of the
source field and observation, the complex amplitude in a z plane is proportional to the
inverse Fourier transform of the complex amplitude in the z = 0 plane with or without
modification by a quadratic phase factor kr 2 / 2 z (Theorem 2). Equations (1-22a) and
(1-23a) represent diffraction integrals in the Fresnel and Fraunhofer approximations,
respectively. It should be noted that the Fresnel condition of large distance z in Eq. (1-
22a) is a sufficient but not a necessary condition. What is necessary is that the sum of the
neglected terms be small so that their contribution to the diffraction integral is negligible.
The integrals in Eqs. (1-22a) and (1-23a) are referred to as representing the Fresnel
r
and Fraunhofer diffraction patterns of the distribution U ( r ; 0) . The region of space
satisfying the condition z >> krmax
2 / 2 is called the Fraunhofer or the far-field region of
diffraction. The condition itself is called the Fraunhofer or the far-field condition. The
r r 4
region of space satisfying the Fresnel condition z 3 >> k r
r max / 8 but not the
Fraunhofer condition is often referred to as the region of Fresnel or near-field diffraction.
The region of very small z values is referred to as the Rayleigh-Sommerfeld region of
diffraction. It should be evident, though, that the Rayleigh-Sommerfeld integral in Eq. (1-
20) will yield accurate results (within the range of its validity) regardless of the value of
z. Similarly, the Fresnel integral will yield accurate results in the Fraunhofer region as
well. However, calculations in the Fraunhofer region are simpler and more common in
imaging applications (due to cancellation of the quadratic phase factor of free-space
propagation by the focusing quadratic phase factor provided by the imaging system).
10 IMAGE FORMATION
Once z is large enough to satisfy the Fraunhofer condition, a larger z value changes only
the scale of the irradiance distribution without changing the distribution.
r r
r r
( )
2
r 2 r r r 2
r r
l = l0 +
+ ... . (1-21c)
2l0 l0 8l0 3
Accordingly, Eqs. (1-22a) and (1-23a) representing the Fresnel and Fraunhofer
approximations are replaced by
r
i z exp(ikl0 ) ikr 2 r 2 i r r r
U (r ; z) = exp U ( r ; 0) exp
r r dr (1-22b)
l0 l0 2l0 l0
and
r
i z exp(ikl0 ) r 2 i r r r
U (r ; z) = U ( r ; 0) exp
l r r dr . (1-23b)
l0 l0 0
It should be evident that exp(ikl0 ) l0 represents a spherical wave originating at the origin
r
of the z = 0 plane, and ( r ; z ) l0 represents the direction cosines of the position vector of
the observation point. Since l0 varies with the point of observation, a uniform array of
points in the observation plane becomes nonuniform when divided by l0 , thereby making
the numerical calculations of the Fourier transform cumbersome. In Eq. (1-22b), there is
an additional difficulty due to the quadratic phase factor under the integral. These
difficulties may be overcome by making observations on a hemisphere of radius l0 , since
the points in that case are uniformly distributed in the direction cosine space. The
obliquity factor z l0 reduces to unity in that case. In imaging, however, the observations
are made in a plane. Also, the fields of view of high-quality imaging systems are
generally small, and Eqs. (1-22a) and (1-23a) representing diffraction at small angles
suffice. Moreover, imaging systems generally have aberrations that increase rapidly with
the field of view and impact the diffracted field a lot more significantly than any error
made in assuming small angles of diffraction.
To get an idea of the distances of the Fresnel and Fraunhofer regions from a
diffracting aperture, we consider a circular aperture with a diameter of 1 cm (so that
rmax = 0.5 cm ) illuminated by a collimated beam of light with a wavelength of 0.5 mm.
r4
The Fresnel condition z 3 > k rmax
/ 8 for on-axis diffraction is satisfied for z > 10 cm. Of
course, the larger the distance, the better the approximation. However, for the Fraunhofer
r2
condition z > k rmax
/ 2 to be satisfied, observations must be made at z > 157 m. In the
nomenclature of optical aberrations, a phase factor varying as r 4 is referred to as a
primary spherical aberration. Similarly, a phase factor varying as r 2 is referred to as a
defocus aberration. They are discussed in detail in Section 1.4.2. At z = 10 m, for
1.2 Rayleigh-Sommerfeld Theory of Diffraction and Huygens-Fresnel Principle 11
example, the amount of defocus wave aberration rmax 2 2 z = 2.5 , but the spherical
4 8z 3 is only onthe order of 10-7 l. Even at z = 1 m, the defocus aberration
aberration rmax
is 25 l but spherical aberration is only on the order of 10-4 l and, therefore, negligibly
small. At shorter distances, it increases rapidly and the Fresnel approximation does not
hold. Since the only difference between the diffraction integrals represented by Eqs. (1-
22a) and (1-23a) is the presence of the quadratic phase factor in the former, Fresnel
diffraction may be referred to as the defocused Fraunhofer diffraction.
In the Fresnel approximation, Eq. (1-18) for the point-spread function reduces to
r i r2
h( r ; z ) =
exp ikz 1 + 2 . (1-24)
z 2 z
Here, kz represents the phase delay of all components in propagating from one plane to
another a distance z apart. The factor of
i indicates that the diffracted wave is out of
phase with the incident wave by / 2 . The inverse dependence on z represents the
inverse-square law of wave intensities. Equation (1-24) shows that a Huygens’ spherical
wavelet is replaced by a parabolic one in the Fresnel approximation. The parabolic
surface shifts laterally as the source of the secondary wavelet shifts in the z = 0 plane,
and the diffracted field is shift-invariant. The shift-invariant property may also be seen by
r r r
letting U ( r ; 0) equal a Dirac delta function ( r
r0 ) in Eq. (1-22).
r
If we consider a Huygens’ wavelet centered at r , Eq. (1-24) is replaced by
r r 2
r r i " ( r
r ) $"
h( r
r ; z ) =
exp !ikz 1 + % . (1-25)
z 2 z 2 "
#" &
In the Fraunhofer approximation, the term in r 2 in the exponent on the right-hand side
of Eq. (1-25) is neglected so that it reduces to
rr r r
i r 2
2r r
h( r ; r ; z ) =
exp ikz 1 + . (1-26)
z 2z 2
If we ignore the quadratic phase factor varying as r 2 in Eq. (1-26) (because it is small for
large z and it does not impact the irradiance distribution of the Fraunhofer pattern), Eq.
(1-26) shows that in effect the Huygens’ spherical wavelet has been replaced by a planar
one. The direction cosines of its surface normal are given by (,
, ) , where
r
(,
) = (r / z) . Physically, this is understandable since a small portion of a large
spherical surface can be approximated by a plane. As the source of the secondary wavelet
r
shifts in the z = 0 plane, thereby changing r , the plane surface representing it is tilted.
Hence, the shift-invariant property of the diffracted field is lost. This may also be seen by
r r r
letting U ( r ; 0) equal a Dirac delta function ( r
r0 ) in Eq. (1-23). However, this loss
is only in a mathematical sense. In reality, the quadratic phase factor kr 2 / 2 z (which is
12 IMAGE FORMATION
negligibly small in the Fraunhofer region) is nonzero, and the diffracted wave field
remains shift invariant.
[ ( )]
1/ 2 $
H (,
; z ) = exp !ik z 1
2 +
2 % ( Rayleigh
Sommerfeld) . (1-27)
# &
For small values of the direction cosines a and b, i.e., for small angles of diffraction with
z axis, Eq. (1-27) reduces to
# 2
(
H (,
; z ) = exp !ik z 1
2 +
2
1
)$%& ( Fresnel and Fraunhofer ) . (1-28)
It is easy to show that it represents the Fourier transform of the point-spread function in
the Fresnel approximation given by Eq. (1-24)[see Problem 1]. Hence, the Fresnel
approximation is synonymous with diffraction in the paraxial approximation. It should be
noted that, mathematically, there is no transfer function associated with Fraunhofer
diffraction since the shift invariant property of the diffracted field is destroyed in the
Fraunhofer approximation. However, as explained above following Eq. (1-26), the
Fresnel transfer function can be used in this approximation as well.
Theorem 3. The Gaussian image of a small object lying at a large distance from an
imaging system is an exact replica of the object, except for its magnification.
The object and the image lie in mutually parallel planes that are perpendicular to the
optical axis of the system. The aperture stop of the system and its images in the object
and image spaces, namely, the entrance and exit pupils, respectively, also lie in planes
that are parallel to the object and image planes.
1.3 Gaussian Image 13
r
Consider, as illustrated in Figure 1-2, an object of radiance B ( ro ) (in W m 2 sr ),
r
where ro is the position vector of a point in the object plane lying at a distance zo from
the plane of the entrance pupil. The position of the Gaussian image of a point object
r
located at ro is given by
r r
rg = M ro (1-29)
in the Gaussian image plane, where M is its transverse magnification. An object area
r r
element * ro centered at ro is imaged as an area element
r r
* rg = M 2 * ro (1-30)
r
centered at rg . We assume that the object lies at a very small angle from the optical axis
of the system so that its cosine is approximately equal to unity. We also assume that the
size of the entrance pupil is small compared to zo , so that all points on it lie at
approximately the same distance zo from the object element. Accordingly, the solid angle
subtended by an entrance pupil of an area Sen at the object element is Sen zo2 . The power
entering the pupil from the object element is given by
r r
( )
Pen = Sen zo2 B ( ro ) * ro . (1-31)
If , is the transmission factor of the system representing the fraction of the incident
power transmitted by it, the corresponding power exiting from the exit pupil is given by
Pex = , Pen
r r
( )
= , Sen zo2 B ( ro ) * ro . (1-32)
EnP ExP
Po( ro) Pg( rg)
Oen OA Oex
Figure 1-2. Schematic of Gaussian imaging by an optical system with its optical axis
OA, entrance pupil EnP, and exit pupil ExP. A point object lies at Po with a position
r
vector ro in the object plane at a (numerically negative) distance zo from the
r
entrance pupil. Its Gaussian image lies at Pg with a position vector rg in the
Gaussian image plane at a distance z g from the exit pupil.
14 IMAGE FORMATION
r
(r )
This power is contained in the image element * rg . If Ig rg is the irradiance (in W m 2 )
of this image element, then
(r )
Pex = Ig rg * rg
r
. (1-33)
(r ) r
( ) r
Ig rg * rg = , Sen zo2 B ( ro ) * ro
r
. (1-34)
(r ) (
Ig rg = , Sen zo2 M 2 B rg M) (r ) . (1-35)
Equation (1-35) describes the irradiance distribution of the Gaussian image, i.e., the
aberration-free image according to geometrical optics. It shows that, except for its
magnification, the Gaussian image of a small object lying at a large distance from the
imaging system is an exact replica of the object (Theorem 3). For systems with large
fields of view, the irradiance actually decreases according to the fourth power of the
cosine of the chief ray angle (i.e., the angle of the ray passing through the center of the
aperture stop, and therefore through the centers of the entrance and exit pupils), as
discussed in Section 2.6 of Part I. There is an additional decrease beyond a certain angle
due to vignetting of the rays by one or more elements of the system, resulting in an
effective aperture with a reduced area compared to that for an on-axis point object (see
Section 2.2.4 of Part I). Although the aberrated image according to geometrical optics can
be calculated, we will not do so, since what is observed in practice is determined by
diffraction, which we discuss next.
surface S passing through the center of the exit pupil with its center of curvature at the
Gaussian image point Pg is called the Gaussian reference sphere. A spherical wave
emerging from the exit pupil implies that the object rays transmitted by the system travel
equal optical path lengths when propagating from the object point to the reference sphere.
It should be evident that the reason for considering diffraction of the object radiation
at the exit pupil is that it represents the limiting aperture of the ray bundle converging to
the image. However, for point objects with sufficiently large field angles, different
portions of a ray bundle may be limited by different apertures (including imaging
elements) of the system. In such cases, the complex amplitude in the image plane may be
determined by considering diffraction at each of the limiting apertures. In practice,
however, the shape (perimeter) of the exit pupil is determined in the plane of the axial
exit pupil and the image is determined by considering diffraction at it.
If the optical path lengths of the rays are not equal to each other, a distorted spherical
wave emerges from the exit pupil and the system is said to be aberrated. If rays from a
point object are traced through the system so that they travel an optical path length equal
to that of the chief ray up to the exit pupil, the surface passing through their end points is
called the system wavefront for that point object. The optical deviation (i.e., the
ExP
Pi (ri; zi)
Pig
Pg(rg; zg)
R
O l s
Pp (rp; 0) W
S
Defocused Gaussian
image plane image plane
zi
zg
geometrical deviation times the refractive index ni of the image space) of the wavefront
from the Gaussian reference sphere along a ray is called the wave aberration. It
represents the difference between the optical path lengths of the ray under consideration
and the chief ray (passing through the center of the exit pupil) in propagating from the
point object to the reference sphere. Thus, the wave aberration of a ray is numerically
positive if it travels a longer optical path length compared to the chief ray in reaching the
reference sphere. Since the optical path lengths of the rays from the reference sphere to
the Gaussian image point are equal, the wave aberration of a ray is also equal to the
difference between its optical path length and that of the chief ray in propagating from the
point object Po to the Gaussian image point Pg .
r (r r )
For an aberrated system, let W rp ; ro be the wave aberration of a ray passing
r
through a point rp in the plane of the exit pupil. The complex amplitude at rp due to an
r r
object element * ro centered at ro may be written
(r r ) (r r )
*Uex rp ; ro = P rp ; ro exp (
iks) , (1-36)
where
xp
Pp(rp)
p xi
rp
Pi (ri)
Op
Pig ri i xg
an il
pl up
e
P
zi
R Pg(rg)
yp
Oi g
rg
e ion
an at
pl erv
zg
bs
O
Og
yi
z
pl n
e
e sia
an
ag us
im Ga
yg
Figure 1-4. Right-hand coordinate system in the pupil plane, and defocused and
Gaussian image planes. The optical axis of the system is along the z axis.
pg = p
g and pi = p
i .
1.4 Diffraction Image 17
r r2 12
s = zg2 + rp
rg (1-37a)
1 r r 2 1 r r 4
= zg + rp – rg
rp – rg + .... (1-37b)
2 zg 8zg3
1
( 1
) (r )
2
= zg + rp2 + rg2
2 rp rg cos pg
3 2
p + rg2
2 rp rg cos pg + .... (1-37c)
2 zg 8z g
is the distance between the pupil point Pp and the Gaussian image point Pg (see Figure
1-4) and
(r r ) (r r ) [
P rp ; ro = A rp ; ro exp ikW rp ; ro (r r )] , inside the exit pupil
= 0 , outside the exit pupil , (1-38)
r r
is called the pupil function of the system. The amplitude function A rp ; ro , called the ( )
apodization function, represents the variation of amplitude of the wave across the exit
pupil and accounts for any nonuniform transmission of the system. The inverse-square-
law dependence of irradiance on the distance s is contained in the amplitude function. It
r r
should be evident that rp = rp , rg = rg , and pg is the angle between the position
r r r
( )
12
vectors rp and rg . Note that s = R = zg2 + rg2 when rp = 0 .
r
The irradiance at a point rp in the plane of the exit pupil due to the object element
under consideration is given by
(r r ) (r r )
2
I p rp ; ro = *Uex rp ; ro (1-39a)
(r r )
2
= P rp ; ro (1-39b)
(r r )
= A 2 rp ; ro . (1-39c)
The corresponding power in the exit pupil is obtained by integrating the irradiance across
the pupil, i.e.,
r r r r
P (r ; r )
2
Pex ( ro ) = p o d rp (1-40a)
r r r
= A ( rp ; ro ) d rp
2
. (1-40b)
r r r r r
*Ui ( ri ; ro ; zi ) =
i
zi
( ) [
P rp ; ro exp ik (l
s) d rp ] , (1-41)
where
r r 2 1 2
l = zi2 + rp
ri (1-42a)
1 r r 2 1 r r 4
= zi + rp
ri
rp
ri +K (1-42b)
2 zi 8zi3
1
( 1
) (r )
2
= zi + rp2 + ri2
2 rp ri cos pi
3 2
p + ri2
2 rp ri cos pi + .... (1-42c)
2 zi 8 zi
r
is the distance of the observation point Pi from a pupil point Pp , ri = ri and pi is the
r r
angle between the position vectors rp and ri (see Figure 1-4). We may add that if the
obliquity factor is not close to unity, it may be replaced by z R for a better
approximation.
zg zi
r r
p i
r +
zg 2 zi zg
g rp
8 zi3 zg3
1 1 1 4 1 3 ri rg
3
3 rp + 2 rp 3 cos pi
3 cos pg
8 zi zg zi zg
1 ri3 rg3
+ rp 3 cos pi
3 cos pg + ... . (1-43)
2 zi zg
Each term on the right-hand side of Eq. (1-43) depends on the location of the
observation and Gaussian image planes and represents a phase term when multiplied by k.
The first term represents a piston aberration. It is simply the distance between the two
planes. Similarly, the second term represents a quadratic phase factor. These two terms do
not affect the irradiance distribution in the defocused iamge plane. Hence, they have no
impact on the image of an incoherent extended object since it is linear in irradiance. The
third term, when exponentiated, forms a Fourier kernel in Eq. (1-41). The quantity
r
( )
zi zg rg in this term represents the position vector of a point Pig where the line joining
the center Op of the exit pupil and the Gaussian image point Pg intersects the defocused
image plane (see Figure 1-4). The fourth term represents the usual defocus aberration due
to the observation plane being different from the Gaussian image plane. The fifth term is
1.4 Diffraction Image 19
a higher-order piston aberration, which also does not affect the irradiance distribution in
the observation plane. The sixth term has the form of a primary spherical aberration, as
may be seen by comparing it with Eq. (3-34) of Part I. The next four terms have two
components each, but only one of them represents a classical aberration. The other
component in each case is only deceptively similar to a classical aberration, but it
depends on the coordinates of the observation point (rather than the Gaussian image
point). For example, the component varying as rp3ri cos pi may be called axial coma in a
manner somewhat similar to the axial coma of a perturbed system (see Section 7.2.1 of
Part I), and the component varying as rp3rg cos pg may be called Seidel coma.
The quantity
r r r 1 r r z r
( ) zi
Wd rp ; ri ; rg = l
s + rp ri
i rg
zg
(1-44)
is the aberration of the diffracted wave.5 It should be noted, however, that this aberration
is related to the propagation of the wave and has nothing to do with the characteristics of
the imaging system (other than its dependence on the location of the Gaussian image
point). It would be convenient to call the last five terms of Eq. (1-43) spherical aberration,
coma, astigmatism, field curvature, and distortion of the diffracted wave, respectively.
The aberrations that are radially symmetric in pupil coordinates vanish when the
observation plane coincides with the Gaussian image plane, i.e., when zi = zg . Hence,
they are aberrations of the diffracted wave associated with defocus. Only the first four
terms are retained in the Fresnel approximation, and only the first three are retained in the
Fraunhofer approximation. These approximations limit the region of observation space
that accurately satisfies the Fourier-transform relationship to near the optical axis and
away from the pupil. When the observation plane coincides with the Gaussian image
plane, the Fraunhofer region of diffraction reduces to the Gaussian image plane.
In practice, the aberrations of the diffracted waves are negligibly small.6 For
example, the depth of focus will be determined by the amount of defocus aberration that
can be tolerated. For systems with large Fresnel numbers, e.g., photographic systems, the
depth of focus for 8 defocus aberration tolerance is F 2 . Here, F = zg D is the focal
ratio of the image-forming light cone, where D is the diameter of the exit pupil. In that
case, the corresponding spherical aberration term in Eq. (1-43) is approximately equal to
( )
3 128 F 2 , which is negligibly small. The spherical aberration is also small for a
system with a small Fresnel number. This may be seen by considering, as discussed in
Section 2.5, a laser transmitter with an aperture diameter of 25 cm focusing a beam of
wavelength 10.6 mm at a distance of 1.47 km so that its Fresnel number is unity. Based on
the defocus aberration, a Strehl ratio of 0.8 is obtained at a distance of 3 km, showing a
large depth of focus. The magnitude of the spherical aberration term in this case is only
about 10 -2 -m . For the angle-dependent terms, such as coma, it is important to note that
the region of interest in the image of a point object is where there is a significant
r
( )
illumination. This region corresponds to ri , pi values that lie, say, within the Airy disc
20 IMAGE FORMATION
(discussed in Section 2.1) that is centered at the Gaussian image point. Thus, even for
rg >> F , we have ri . rg and pi differs from gi by no more than approximately D.
Hence, the coma term for a field angle = rg zg is approximately equal to 16 F 2 , ( )
which is also negligibly small. The value of the axial coma is equal to 16 F 2 .
Similarly, the astigmatism, field curvature, and distortion terms are approximately equal
( ) ( ) ( )
to 2 8F , 3 2 16 , and 3 4 , respectively. Their axial values are extremely
( ) ( )
small and equal to 8zg , 16zg , and ( 2 D) . Thus, the aberrations of a
2
diffracted wave are negligibly small. The expressions for their values are summarized in
Table 1-1. It should be noted that in high-quality imaging systems, the value of is often
quite small (few degrees or less) since it is limited by the aberrations of the system itself.
We may neglect all terms in Eq. (1-43) beyond the defocus term if
1 1 4
3
3 a < 8 , (1-45)
zi zg
where a is the maximum value of rp , i.e., if the peak aberration contributed by the
primary spherical aberration is < / 8 . We will see, for example, in Section 2.3 that such
an aberration has a negligible effect on the quality of an image. Hence, Eq. (1-41) for the
complex amplitude may be written
r r " 1 r 2 rg $"
2
*Ui ( ri ; ro ; zi ) =
i
zi
( )
exp !ik zi
zg + i
%
"# 2 zi zg "
&
r r 2 i r r z r r
( )
× P rp ; ro ; zi exp
zi
rp ri
i rg d rp
zg
, (1-46)
where
Table 1-1. Aberrations of a diffracted wave for an on-axis and an off-axis point
object. The field angle of the image point is indicated by the angle .
Coma 16 F 2 ( 16 F ) 2
r r r r ik 1 1
(
P rp ; ro ; zi ) ( )
= P rp ; ro exp
rp2
2 zi zg
(1-47)
is the defocused pupil function signifying that the image is observed in a plane other than
the Gaussian image plane, i.e., zi / zg .
So far we have been concerned with the phase variations across the exit pupil for a
wave from a given point object and ignored the optical path length of the chief ray from
the object point to its Gaussian image point. There is an additional phase factor that needs
to be considered when comparing the waves from different point objects. It represents the
difference in the optical path lengths of their chief rays from the object points to their
Gaussian image points. The difference in the optical path lengths of a chief ray for an
r
object point located at ro at a distance zo from the entrance pupil and that for an axial
point object, up to the quadratic approximation (see Figure 1-2), is given by
(r r )
W ro ; rg = ( Po Oen + z )
Oex Pg
zg ( )
2 r2
~
ro + g , (1-48)
2 zo 2 zg
where zo is numerically negative according to our sign convention given in Section 1.3.2
of Part I. It is assumed that the optical path lengths of the chief rays from the entrance
pupil to the exit pupil are the same. This must, of course, be verified for the system under
consideration by ray tracing, or the phase must be adjusted if the assumption is not valid.
Equation (1-48) is evidentally valid in the case of a thin lens when its aperture stop and,
therefore, its entrance and exit pupils are located at the lens. Adding the phase
contribution from Eq. (1-48), Eq. (1-46) is modified to
r r " 1 r 2 r 2 $"
*Ui ( ri ; ro ; zi ) =
i
zi
( )
exp !ik zi
zg + i
o %
"# 2 zi zo "
&
r r 2 i r r z r r
(
× P rp ; ro ; zi exp
)
rp ri
i rg d rp . (1-49)
z zg
i
The additional phase factor does not impact imaging of an incoherent object, but does
impact coherent imaging, which is discussed in Section 1.12.
In obtaining Eq. (1-49), we have in effect propagated the wave from the exit pupil to
the observation plane in the Fresnel approximation, as may be seen by substituting Eq. (1-
36) into Eq. (1-22a) and retaining only the first two terms on the right-hand side of Eq.
(1-37b). Indeed, for a thin lens of focal length f, Eq. (1-49) can also be obtained by
considering Fresnel propagation of the spherical wave from a point object to the entrance
pupil, its modification by the lens, i.e., introduction of the quadratic phase factor
( )
exp
ikrp2 2 f , and Fresnel propagation to an observation plane (see Problem 4).
22 IMAGE FORMATION
r r r
= Pex (ro ) PSF ( ri ; ro ; zi ) , (1-50)
where
2
rr r r 2 i r r z r r
PSF ( ri ; ro ; zi ) = r
1
Pex ( ro )2 zi2
( )
P rp ; ro ; zi exp
zi
rp ri
i M ro d rp
zg
(1-51)
is called the incoherent point-spread function of the imaging system, as we shall see in
the next section. It is proportional to the modulus square of the inverse Fourier transform
r
of the defocused pupil function of the system (Theorem 4). Hence, substituting for Pex (ro )
from Eq. (1-32) into Eq. (1-50) and integrating over the object, we obtain
r rr
Ii ( ri ; zi ) = * I ( ri ; ro ; zi )
r r r r
(
= , Sen zo2 ) B ( ro ) PSF ( ri ; ro ; zi ) d ro
object
, (1-52)
where we have neglected the dependence of Sen and the solid angle it subtends at an
object element on the location of the element. Thus, the object is assumed to be small, far
1.4 Diffraction Image 23
from the system, and at a small angle from the axis of the system so that the cosine-to-
the-fourth-power dependence on the chief ray angle in the object space is negligible, and
vignetting of its pupil for the off-axis object elements (discussed in Section 2.6 of Part I)
is either zero or negligible.
r (r r
)
The pupil function P rp ; ro ; zi represents the actual complex amplitude at a point
r r
rp in the plane of the exit pupil due to an object element * ro centered at ro . Evidently, it
r
depends on the location ro of the object element. We now introduce a relative pupil
r r
( )
function G rp ; ro ; zi that represents the complex amplitude per unit amplitude of the
object element. The two pupil functions are related to each other according to
(r r
P rp ; ro ; zi ) r
(r
= U ( ro ) G rp ; ro ; zi
r
) , (1-53)
r r
where U ( ro ) is the amplitude at a point ro . It should be evident that the two pupil
r
functions have identical dependence on rp . Substituting Eq. (1-53) into Eq. (1-40a), we
obtain
r r r r r
G (r ; r ; z )
2
Pex ( ro ) = U ( ro )
2
p o i d rp . (1-54)
For a small object, we may neglect any changes in the nonuniform transmission or
r r r
aberrations introduced by the system as a function of ro . Hence, G rp ; ro ; zi may be
r ( )
assumed to be independent of the position vector ro . [The assumptions of a small field of
view and a small object are the same as those made in obtaining the Gaussian image
distribution of Eq. (1-35).] An object of small extent for which the relative pupil function
r r
( )
G rp ; ro ; zi may be considered independent of the position of a point on it is called
isoplanatic (or the system is said to be isoplanatic for the small object under
r r r
( )
consideration). Of course, the pupil function P rp ; ro ; zi will vary as U ( ro ) varies with
r
ro for such an object.
Substituting Eqs. (1-53) and (1-54) into Eq. (1-51), neglecting the dependence of
r r r
( )
G rp ; ro ; zi on ro , and letting zi ~ zg ~ R , for an isoplanatic object imaged by a system
with a small field of view, we obtain
24 IMAGE FORMATION
r 2 i r r r r
2
r r
( )
G rp ; zi exp
R rp
( ri
M ro ) d rp
PSF ( ri ; ro ; zi ) = r r . (1-55)
( )
2
2 R 2 G rp ; zi d rp
We note that the integral in the numerator on the right-hand side of Eq. (1-55) depends on
r r r
the position vectors ri and rg (which is equal to M ro ) of the observation and Gaussian
r r
image points, respectively, through their difference ri
rg only, and we may replace
r r r r
PSF ( ri ; ro ; zi ) by PSF ( ri
M ro ; zi ) . Thus, the PSF is shift invariant in the sense that its
form does not change as the object point is shifted; only its location changes by virtue of
r
it being centered at rg . Accordingly, Eq. (1-52) for the irradiance distribution of the image
of an isoplanatic incoherent object may be written
r r r r r
(
Ii ( ri ; zi ) = , Sen zo2 ) B ( ro ) PSF ( ri
M ro ; zi ) d ro
object
, (1-56a)
(
= , Sen zo2 M 2 ) B (rr M ) PSF (rr
rr ; z ) d rr
g i g i g (1-56b)
r r r r
= Ig ( rg ) PSF ( ri )
rg ; zi d rg , (1-56c)
where we have made use of Eq. (1-35). Thus, the diffraction image of an isoplanatic
incoherent object is equal to the convolution of its Gaussian image (which is a scaled
replica of the object) and the PSF of the imaging system (Theorem 5). It should be
evident that the magnification M of the image is assumed to be constant across the
isoplanatic region of the object. Of course, a nonuniform magnification will result in a
distortion of the image also.
Multiplying the numerator and denominator of the right-hand side of Eq. (1-55) by
r
U ( ro ) , we may also write the PSF in the form
2
r r 2 i r r r
2
PSF ( ri ; zi ) =
1
2 2 P rp ; zi exp
Pex R R
( )
rp ri d rp
, (1-57)
r r
where we have suppressed any explicit dependence on ro and the position vector ri is
r
written with respect to rg . The irradiance distribution of the image of a particular object
element is obtained by multiplying the PSF by Pex for that element.
Substituting Eq. (1-58) into Eq. (1-52), we obtain the irradiance distribution of its image,
namely,
(r r
Ii ri ; rj ; zi ) (r
= Pex PSF ri ; rj ; zi
r
) , (1-59a)
where
(
Pex = , Sen zo2 B j ) (1-59b)
represents the total power from the point object in the exit pupil of the imaging system.
Substituting Eq. (1-51) into Eq. (1-59a), we obtain
2
rr r r 2 i r r z r r
(
Ii ri ; rj ; zi ) 1
= 2 2
zi
(
P rp ; rj ; zi exp
zi
)
rp ri
i M rj d rp
zg
. (1-60)
For an isoplanatic object, we substitute Eq. (1-58) into Eq. (1-56a) and obtain
(r r
Ii ri ; rj ; zi ) (r
= Pex PSF ri
M rj ; zi
r
) , (1-61)
rr 1 r 2 i r r r
r
2
(
Ii ri ; rj ; zi ) R
(
= 2 2 P rp ; zi exp
R
rp ) ( ri
M rj d rp
) . (1-62)
It is evident from Eq. (1-62) that the images of two point objects lying within an
isoplanatic patch are identical except that they are centered at their respective Gaussian
image points. Such an imaging is referred to as shift-invariant imaging, and comes about
because of the invariance of the relative pupil function with the position of an object
point on the isoplanatic object. Integrating both sides of Eq. (1-62), we find that
r r r 1 r r 2 i r r r r
2
(
i i j i R
)
I r ; r ; z d ri = 2 2 d ri P rp ; zi exp
R
rp ( ) ( )
ri
M rj d rp
r r r r r r
= d rp P ( rp ; zi ) P * ( rp) ( rp
rp ) d rp
r r
P (r ; z )
2
= p i d rp
= Pex . (1-63)
We obtain the same result if we integrate both sides of Eq. (1-61). It represents the
conservation of power; i.e., the total power in the image is equal to the total power in the
r r r
exit pupil. It also shows that the integral of PSF ri ; rj ; zi over all values of ri in the ( )
image plane is unity. Hence, PSF given by Eq. (1-51) or Eq. (1-57) represents the
r
irradiance distribution of the defocused image of a point object located at ro per total
26 IMAGE FORMATION
power in the image. It is called the diffraction point-spread function (PSF) of the system.
Note that its dimensions are m
2 .
Unapodized Pupil
A system with a pupil that is uniformly illuminated is said to be unapodized. For
r
( )
such systems, G rp is a constant varying inversely with the distance zo of the entrance
r
( )
pupil from the object. For an unapodized pupil, let A rp be equal to a constant, say, A0 .
r
If we redefine ri , as illustrated in Figure 1-5, as the position vector of an image point Pi
r r
with respect to the Gaussian image point rg = M rj when the image is observed in the
r
Gaussian image plane, or with respect to the corresponding point zi zg rg (where the ( )
line joining the center of the exit pupil and the Gaussian image point intersect the
defocused image plane) if it is observed in a defocused image plane at a distance zi from
the plane of the exit pupil, Eq. (1-62) becomes
r 2 i r r r
2
Ii ( ri ; zi )
I r
[ ( )]
= 2 0 2 exp i kW rp ; zi exp
R R
rp ri d rp
, (1-64)
where I0 = A02 is the irradiance at the exit pupil. If, in addition, the pupil is aberration
free and the observation is made in the Gaussian image plane so that there is no defocus
r
( )
aberration, i.e., if W rp ; zg = 0, then Eq. (1-64) reduces to
r 2 i r r r
2
I (0)
Ii ( ri ) = i 2 exp
Sex R
rp ri d rp , (1-65)
where
Pex Sex
Ii ( 0 ) = (1-66)
2 R 2
r
is the aberration-free irradiance at the Gaussian image point ri = 0 . Of course,
Pex = I0 Sex in this case. The effect of aberrations and or amplitude variations on the
central irradiance is discussed in Section 1.9.
yi yg
Pi Pi
ri ri
Pig Pg
xi xg
Oi Og
r
Figure 1-5. Redefinition of ri as the position vector of an image point Pi in the (a)
defocused image plane with respect to Pig and the (b) Gaussian image plane with
respect to Pg .
1.6 Optical Transfer Function (OTF) 27
Theorem 7. The OTF of an imaging system is also equal to the autocorrelation of its
pupil function.
Theorem 8. The PSF and the OTF of a system with a radially symmetric pupil
function form a zero-order Hankel transform pair.
A relationship between the spatial-frequency spectra of the object and image may be
obtained by taking a Fourier transform of both sides of Eq. (1-56c). Let
r r r r r
B˜ ( vo ) = B ( ro ) exp (2 i vo ro ) d ro (1-67)
and
r r r r r
I˜i ( vi ) = Ii ( ri ) exp (2 i vi ri ) d ri (1-68)
be the spectral components of the object and image corresponding to the object and image
r r
spatial frequencies vo and vi , respectively, where
r r
vi = vo M . (1-69)
28 IMAGE FORMATION
Substituting Eq. (1-56c) into Eq. (1-68) and suppressing the zi dependence of the PSF
and the OTF, we obtain
r r r r r r r r
I˜i ( vi ) = d ri exp (2 i vi ri ) Ig ( rg ) PSF ( ri
rg ) d rg
r r r r r r r r r r
= d rg Ig (rg ) exp (2i vi rg ) PSF (ri
rg ) exp [2 vi (ri
rg )] d ri
r r
= I˜g ( vi ) 0 ( vi ) , (1-70)
r
where Ĩg ( vi ) is the spectral component of the Gaussian image corresponding to a spatial
r
frequency vi and
r r r r r
0 ( vi ) = PSF ( ri ) exp (2 i vi ri ) d ri (1-71)
is called the optical transfer function (OTF) of the imaging system at this frequency.
Equation (1-71) states that the OTF, which is dimensionless, is the Fourier transform of
the PSF. From Eq. (1-35), it is evident that
r r
(
I˜g ( vi ) = , Sen zo2 B˜ ( M vi ) ) (1-72a)
r
(
= , Sen zo2 B˜ ( vo ) . ) (1-72b)
r
Equation (1-70) states that the spatial-frequency spectrum Ĩi ( vi ) of the diffraction image
r
of an isoplanatic incoherent object is equal to the product of the spectrum Ĩg ( vi ) of its
Gaussian image (which in turn is equal to a scaled spectrum of the object) and the OTF
r
0 ( vi ) of the imaging system (Theorem 6). It is a consequence of the fact that if a function
is equal to the convolution of two functions, as in Eq. (1-56c), its Fourier transform is
equal to the product of their Fourier transforms.7
Substituting Eq. (1-57) into Eq. (1-71), the OTF can also be written
r r r r r 2i r r r
2
0 ( vi ) =
1
Pex 2 R 2
( )
d ri exp (2 i vi ri ) P rp exp
R
rp ri d rp
r r
=
1
Pex 2 R 2
d rp P rp ( ) d rrp P * (rrp) d rri exp 2Ri rri (rrp
rrp + R vri )
r
= Pex
1 d rp P rp (r ) d rrp P * (rrp) (rrp
rrp + R vri )
r r r r
= Pex
1 P ( rp ) P 1 ( rp
R vi ) d rp (1-73a)
r r r r r 2 r
P ( rp ) P * ( rp
R vi ) d rp P ( rp ) d rp
=
(1-73b)
r r r
( ) (
= Pex
1 A rp A rp
R vi exp iQ rp ; zi d rp ) [ (r )] r
, (1-73c)
1.6 Optical Transfer Function (OTF) 29
(r )
where the pupil function P rp is given by Eq. (1-38) when the image is observed in the
Gaussian image plane and by Eq. (1-47) when it is observed in a defocused image plane,
and
(r r
Q rp ; vi ; zi ) (r ) (r
= 2 rp ; zi
2 rp
R vi ; zi
r
) (1-74)
is a phase aberration difference function defined over the region of overlap of two pupils,
r r r
one centered at rp = 0 and the other at rp = Rvi . The phase aberration 2 is related to
the wave aberration W according to
2 = (2 ) W . (1-75)
Equation (1-73b) shows that the OTF of an incoherent imaging system is also equal to the
normalized autocorrelation of its pupil function (Theorem 7). [The autocorrelation of a
function f ( x ) is equal to its convolution with the function f * (
x ) .] The region of
integration in the numerator of the right-hand side of this equation is the overlap region of
r
the two pupils. We may note here that the region of overlap is maximum for vi = 0,
giving a value of unity for 0 (0) . Because of the finite size of the pupil, the overlap region
r
reduces to zero at some frequency vc , called the cutoff frequency, and stays zero for
r r r
larger frequencies; i.e., 0 ( vi ) = 0 for vi 3 vc . If the image is observed in a defocused
image plane such that zi is significantly different from R, then R in Eq. (1-73) is replaced
by zi , as may be seen by using the PSF obtained from Eq. (1-60). Problem 4.2 is an
example of this type, where the defocused OTF of a weakly truncated Gaussian pupil is
considered.
r
= PSF ( ri ) , (1-76)
r r
which shows simply that the functions PSF( ri ) and 0 ( vi ) form a 2-D Fourier transform
pair. The value of the PSF at the origin may be written
r r
PSF (0) = 0 ( vi ) d vi (1-77a)
r r
= Re 0 ( vi ) d vi , (1-77b)
r r
where Re 0 ( vi ) is the real part of 0 ( vi ) . Since the PSF at any point is a real quantity,
only the real part of the complex OTF contributes to the integral. The integral of its
imaginary part must be zero.
Consider a system with a radially symmetric pupil function, i.e., one for which
30 IMAGE FORMATION
(r )
P rp = P rp ( ) . (1-78)
If we let
r
( )
rp = rp cos p , sin p , 0 ' p < 2 , (1-79a)
and
r
ri = ri (cos i , sin i ) , 0 ' i < 2 , (1-79b)
Noting that8
2
exp (ix cos ) d = 2 J0 ( x ) , (1-81)
0
where J 0 ( ) is the zero-order Bessel function of the first kind, Eq. (1-80) reduces to
( ) P ( r ) J (2 r r )
2
PSF(ri ) = 4 2 Pex 2 R 2 p 0 p i R rp drp . (1-82)
Thus, the PSF of a system with a radially symmetric pupil function is also radially
symmetric. Multiplying both sides of Eq. (1-82) by Pex , we obtain the radially symmetric
irradiance distribution of the image of a point object. Similarly, Eqs. (1-71) and (1-76)
reduce to
and
r
where vi = vi . Thus, the PSF and the OTF of a system with a radially symmetric pupil
function form a zero-order Hankel transform pair (Theorem 8). This result is a
consequence of the fact that the Fourier transform of a radially symmetric function is a
zero-order Hankel transform. From Eq. (1-83), we note that, since the PSF is a real
function, the radially symmetric OTF is also a real function.
By multiplying PSF(ri ) with Pex in Eqs. (1-83) and (1-84) we obtain relationships
between the irradiance distribution of the image of a point object and the OTF of a system
with a radially symmetric pupil function. For example, the irradiance distribution may be
obtained from the OTF according to
1.6 Optical Transfer Function (OTF) 31
The encircled power of the image, i.e., the power contained in a circle of radius rc
centered at the Gaussian image point, is given by
r r
Pi (rc ) = I ( r ) d ri
rr ' r i i
. (1-86)
i c
Or, substituting Eq. (1-85) into Eq. (1-87) and noting that8
z0
where J1 ( ) is a first-order Bessel function of the first kind, we obtain
Equation (1-89) gives the encircled power of the image in terms of the radially symmetric
OTF of the system.
To understand the physical significance of the OTF defined by Eq. (1-71), we derive
the following theorem.
Bo(1+ m)
B(ro)
1/ "6o "
Bo(1– m)
0
xo
lo(1+ m)
lg(ri)
lo(1– m)
0
xi
1/ "6i "
7
lo(1+ m "0 ")
li(ri)
0
xi
Figure 1-6. Image of a sinusoidal object shown along the x axis. (a) Object, (b)
Gaussian image, and (c) Diffraction image.
r r r
[
I g ( ri ) = Io 1 + m cos (2 vi ri + 4)] , (1-91)
where
(
Io = , Sen zo2 M 2 Bo ) (1-92)
is the average irradiance of the image. We note that the image is also sinusoidal but with
r r
a frequency vi = vo / M , where M is its magnification. Its modulation and phase are the
same as that of the object, respectively.
Substituting Eq. (1-90) into Eq. (1-56a), we obtain the irradiance distribution of the
diffraction image, which may be written
r r r r r r
r r r r
{ r[
Ii ( ri ) = Io M 2 [ PSF( ri
M ro ) d ro + m Re exp
i (2 vi ri + 4 )
r
]
[
× PSF( ri
M ro ) exp 2 i vi (ri
M ro )] d ro }]. (1-93)
r r r r r 2i r r r r
PSF ( ri
M ro ) d ro =
1
Pex 2 R 2
d ro P rp exp
R
( )
rp ( ri
M ro ) d rp
r 2 i r r r r
( )
× P * rp exp
R
rp ( ri
M ro ) d rp
1.6 Optical Transfer Function (OTF) 33
(
= 1 Pex 2 R 2 ) d rr P (rr ) d rr P * (rr )
p p p p
r 2 i r r
× d ro exp
R
rp
rp ( ) (rri
M rro )
(
= 1 Pex M 2 ) d rr P (rr ) d rr P(rr ) (rr
rr )
p p p p p p
( ) P (rr ) r
2
= 1 Pex M 2 p d rp
= 1 M2 . (1-94)
Following Eq. (1-71), the second integral on the right-hand side of Eq. (1-93) may be
written
r r r r r r r
PSF ( ri
M ro ) exp [2 i vi ( ri
M ro )] d ro = M
2 0 ( vi ) . (1-95)
Thus, like the Gaussian image, the diffraction image of a sinusoidal object of spatial
r r r
frequency vo is also sinusoidal with a spatial frequency vi = vo M , and its average
r
irradiance is Io . However, the modulation of this image is m 0 ( vi ) ; i.e., its modulation
r
is different from that of the object or its Gaussian image by the MTF factor 0 ( vi ) . It is
r
shown later [see Eq. (1-106)] that 0 ( vi ) ' 1. Thus, the modulation of the diffraction
image is always less than or equal to that of the object at any spatial frequency. The phase
of the sinusoidal image is also different from that of the object or its Gaussian image by
r r
the PTF 7( vi ) . Hence, the OTF 0 ( vi ) of an incoherent imaging system corresponding to
r
an image spatial frequency vi is a complex function whose modulus when multiplied by
r
the modulation of the corresponding sinusoidal object (of spatial frequency vo ) gives the
modulation of the sinusoidal image, and whose phase gives the phase of the diffraction
image relative to that of the object or its Gaussian image, as illustrated in Figure 1-6
r r r r
(Theorem 9). Note that vo ro = vi ri .
1.6.3 Properties of OTF
Now we derive certain properties of the OTF from its definition.
Theorem 10. (a) The OTF is a complex symmetric function with a value of unity at
the origin. (b) Its magnitude, the MTF, is unity or less at any other spatial frequency. (c)
The aberrated MTF at any frequency is less than or equal to the corresponding
aberration-free value.
34 IMAGE FORMATION
From Eq. (1-73b), we note that its value at the origin is unity, i.e.,
0(0) = 1 . (1-98)
as may be seen from Eqs. (1-61) and (1-63). Equation (1-98) represents the fact that the
OTF at zero spatial frequency is unity, i.e., the contrast of an image is zero for an object
of zero contrast. From Eq. (1-71), we also note that
r r
0 ( vi ) = 0 * (
vi ) , (1-100)
i.e., the OTF is complex symmetric or Hermitian (Theorem 10a). Therefore, its real part
is even and its imaginary part is odd; i.e.,
r r
Re 0 ( vi ) = Re 0 (
vi ) ,
(1-101)
and
r r
Im 0 ( vi ) =
Im 0 (
vi ) . (1-102)
r r
For two arbitrary but well-behaved functions f ( r ) and g( r ) , Hölder’s inequality 9
states that
r r r r r 1n
r n ( n
1) r ( n
1) n
f ( r ) g( r ) d r ' f ( r ) d r g( r ) dr
n
. (1-104)
r r r r r r 12
0 ( vi ) ' Pex
1
( ) ( ) d rp
2 2
P rp d rp P rp
R vi . (1-105)
Thus, the MTF at any spatial frequency is less than or equal to unity (Theorem 10b).
(r ) (r ) (r
Using Hölder’s inequality with f rp = P rp P * rp
R vi , g( r ) = 1 , and n = 1,
r
) r
Eq. (1-103) yields
1.6 Optical Transfer Function (OTF) 35
r r r r r
0 ( vi ) ' Pex
1 P (rp )P * (rp
R vi ) d rp
r r r r
= Pex
1 P ( rp ) P * ( rp
R vi ) d rp
r r r r
= Pex
1 A( rp ) A( rp
Rvi ) d rp
r
= 0 ( vi ) w = 0 ; (1-107)
i.e., the aberrated MTF at any spatial frequency is less than or equal to the corresponding
aberration-free MTF (Theorem 10c). The ratio of the aberrated and aberration-free MTFs
at a certain spatial frequency is called the Hopkins ratio for that frequency, as discussed
later in Section 1.10. Equation (1-107) shows that Hopkins ratio is less than or equal to
one.
The MTF of an aberration-free system with a nonuniform amplitude across its exit
pupil can be higher or lower than that for a uniform amplitude, depending on the spatial
frequency and the nautre of the amplitude distribution. For example, as discussed in
Section 4.3.3.2, the MTF of a Gaussian pupil is higher for low frequencies and lower for
the high. However, if the amplitude increases from the center of the pupil toward its edge,
then the MTF is lower for low frequencies and higher for the high.
We now summarize the results represented by Eqs. (1-98), (1-100), (1-106), and (1-
107) (Theorem 10).
i. OTF at the origin is unity.
ii. OTF is complex symmetric or Hermitian.
iii. MTF at any spatial frequency is less than or equal to unity.
iv. Aberrated MTF at any spatial frequency is less than or equal to the
corresponding aberration-free MTF; i.e., the Hopkins ratio is less than or
equal to one.
( )
12
vi = 8 2 + ,2 (1-108)
and
9 = tan
1 ( , 8) . (1-109)
Thus, vi is the magnitude of the spatial frequency of a periodic image and 9 gives its
orientation, corresponding to rectangular frequency components 8 and ,. It should be
evident that the OTF corresponding to a two-dimensional periodic image of spatial
36 IMAGE FORMATION
Following Eq. (1-73a), we write the OTF corresponding to a spatial frequency (8, ,)
in the form
( ) (
0 (8, ,) = Pex
1 P x p , y p P * x p
R8, y p
R , dx p dy p ) , (1-110)
r
( )
where x p , y p are the rectangular coordinates of a pupil point with a position vector rp .
To determine, the OTF slope at the origin we write the integral in Eq. (1-110) in a (U, V )
(
coordinate system whose origin is shifted from the x p , y p coordinates system by )
( )
( R 2)(8, ,) but whose axes are parallel to the x p , y p axes, as indicated in Figure 1-7.
Thus, letting
xp = U + R8 2 (1-111a)
and
yp = V + R, 2 , (1-111b)
0 (8, ,) = Pex
1 P(U + R8 2, V + R, 2) P * (U
R8 2, V
R, 2) dU dV . (1-112)
Equation (1-38) [or Eq. (1-47) with zi dependence suppressed] for the pupil function may
be written
yi
v
q
6i
6i–1 p
(R/2)(8,,)
9
yp
9
xi
u
xp
(–R/2)(8,,)
-m /0/)
I (1
oo
R6:
0/)
-m/
Ioo(1
(a) (b)
Figure 1-7. (a) Sinusoidal object yielding image of spatial frequency (vi , 9) or (8, ,) ,
mean irradiance I0 , and modulation m 0 . (b) Geometry for evaluating the OTF of
an optical imaging system. The centers of the two pupils are located at (0, 0) and
( )
R (8, ,) in the x p , y p coordinate system, m ( R 2) (8, ,) in the (U , V ) coordinate
system, and m ( R 2) (vi , 0) in the ( p, q ) coordinate system, where vi = 8 2 + ,2 ( )
12
and 9 = tan
1 ( , 8) . The shaded area is the overlap area of the two pupils.
1.6 Optical Transfer Function (OTF) 37
( ) ( ) [ (
P x p , y p = A x p , y p exp i 2 x p , y p )] ,
(1-113)
[
P(U + R8 2, V + R, 2) = A(U + R8 2, V + R, 2) exp i2(U + R8 2, V + R, 2) ] .
(1-114)
Substituting Eq. (1-114) into Eq. (1-112), the real part of the OTF may be written
[ ]
Re 0 (8, ,) = Pex
1 I (U , V ; 8, ,) cos Q(U , V ; 8, ,) dU d V , (1-115)
where
I (U, V ; 8, ,) = A (U + R8 2 , V + R , 2) A(U
R8 2, V
R , 2) (1-116)
and
Q(U , V ; 8, ,) = 2 (U + R8 2 , V + R , 2)
2 (U
R8 2 , V
R , 2) . (1-117)
The region of integration in Eq. (1-115) is the overlap area of two pupils centered at
m ( R 2)(8, ,) in the (U , V ) coordinate system. Expanding 2(U ± R8 2 , V ± R , 2) in
a Taylor series about the point (U , V ) , e.g.,
2 (U , V ) 2 (U , V )
2 (U + R8 2 , V + R , 2) = 2 (U, V ) + ( R 2) 8 +,
U V
we find that
2(U , V ) 2(U , V )
Q(U , V ; 8, ,) = R 8 +,
U V
3
( R 2)3 8 + , 2(U, V ) + ... .
1
+
3! U V (1-119)
x2 x4
cos x = 1
+
K , (1-120)
2! 4!
we may write
[ ( )]
Re 0 (8, ,) = Pex
1 I (U , V ; 8, ,) 1 + O 8 2 , ,2 , 8, dU d V . (1-121)
Thus, Re 0 (8, ,) does not consist of any aberration-dependent terms that depend linearly
on 8 or , alone. Hence, we obtain
38 IMAGE FORMATION
8 Re 0 (8, ,) = Pex
1 8 I (U , V ; 8, ,) dU d V , (1-122)
8 = , = 0 + 8 = , = 0 +
and a similar equation for the derivative with respect to h. Thus, the derivative of the real
part of the OTF with respect to a spatial frequency component x or h evaluated at the
origin is independent of the aberration, irrespective of the shape of the pupil. It does,
however, depend on the amplitude variations across the pupil. If the amplitude is
uniform, then the integral on the right-hand side reduces to the power contained in the
overlap region of the two pupils whose centers are separated by R(8, ,) from each
other. When divided by the total power Pex , it yields the fractional overlap area of the
two pupils.
Re 0 (8, ,) = Re 0(
8 ,
,) . (1-123)
0 (8, ,) [
= Re 0 (8, ,)] + [Im 0 (8, ,)]
2 2 2
(1-124)
0 (vi , 9) = Pex
1 P( p + Rv i 2 , q ) P * ( p
Rvi 2, q ) dp dq . (1-125)
[
P( p, q ) = A( p, q ) exp i2( p, q ) ] , (1-126)
r
[ ]
cos Q( p, q; vi ) = 1
1
2
[
( Rvi )2 2 ( p, q) ]2 + O (vi4 ) . (1-131)
Taking the derivative of both sides of Eq. (1-132) with respect to vi and evaluating at
vi = 0, we obtain
r
Re 0 (vi , 9) = Pex
1 I ( p, q; vi ) dp dq . (1-133)
iv vi = 0
iv vi = 0
That is, the derivative of the real part of the transfer function with respect to the
radial frequency v evaluated at the origin is independent of the aberration, irrespective of
the shape of the pupil. It does, however, depend on the amplitude variations across the
pupil. This result is to be expected from the component derivatives given by Eq. (1-122).
As discussed later in Section 1.7, the asymptotic behavior of the irradiance distribution of
the image of a point object depends on the value of this derivative. Once again, the
imaginary part of the transfer function depends on the sine of the aberration difference
40 IMAGE FORMATION
which, when expanded in a power series, will contain a linear term in vi . Consequently,
its derivative with respect to vi evaluated at vi = 0 will depend on the aberration.
However, the derivative of the modulus of the transfer function evaluated at vi = 0 is also
independent of the aberration. Indeed, the derivatives of the real part and the modulus of
the transfer function evaluated at vi = 0 are equal. This may be seen by evaluating the
derivative of both sides of 0 = ( Re 0)2 + (Im 0)2 at vi = 0 , and noting from Eq. (1-
2
We summarize the results of this section in that the derivative of the real part of the
OTF of a system with respect to a spatial frequency component 8 or , or a radial
frequency vi evaluated at the origin is independent of its aberration irrespective of the
shape of its exit pupil. The value of this derivative is equal to the corresponding
derivative of the MTF of the system.
Theorem 12. The diffraction OTF approaches the geometrical OTF as the
wavelength approaches zero.
Substituting Eq. (1-114) into Eq. (1-112), the diffraction OTF can be written
{ }
0 (8, ,) = Pex
1 I (U , V ; 8, ,) exp iQ(U , V ; 8, ,) dU d V , (1-134)
where the region of integration is the overlap area of two pupils centered at
m( R 2)(8, ,) . Substituting Eq. (1-75) into Eq. (1-119), we obtain
W (U , V ) W (U , V )
Q(U , V ; 8, ,) = 2 R 8
U
+,
V
+ O 2 ( ) , (1-135)
( )
where O 2 consists of terms with third and higher odd-order derivatives of W (U , V ) .
Substituting Eq. (1-135) into Eq. (1-134) and letting 0, we obtain
W (U , V ) W (U , V ) $
[0 (8, ,)] 0 = Pex
1 A 2 (U , V ) exp !2 iR 8
U
+,
V &
% dU d V , (1-136a)
#
or
[0 (8, ,)] 0 (
= Pex
1 I x p , y p )
" W x p , y p
× exp !2i R 8 +,
( )
W x p , y p( ) $" dx
% p d yp , (1-136b)
"# x p y p "
&
where
1.6 Optical Transfer Function (OTF) 41
( ) (
I x p , y p = A2 x p , y p ) (1-136c)
( )
is the irradiance at a pupil point x p , y p , and the region of integration is the area of the
exit pupil. Equation (1-137) may also be obtained directly from Eq. (1-110) in a similar
manner. However, the equation corresponding to Eq. (1-135) in that case will consist of
terms with odd and even-order derivatives of W x p , y p . ( )
We show in the next section that the OTF obtained according to geometrical optics is
also given by Eq. (1-136b). Hence, it yields the result that the diffraction image reduces
to the geometrical image as the wavelength approaches zero (Theorem 12). It should be
noted that, in practice, the limit of zero wavelength is equivalent to a very large
aberration. This is why the diffraction PSF may be expected to be similar to the
geometrical PSF for large aberrations.
The geometrical PSF (not the Gaussian image) is given by [see Eq. (4-10) of Part]
1
( x, y)
(
I g ( x, y) = I x p , y p ) , (1-137)
(
x p , yp )
( )
where x p , y p are the coordinates of a ray in the plane of the pupil and ( x, y) are its
coordinates in the image plane with respect to the Gaussian image point. Its Fourier
transform for unit total power gives the geometrical OTF:
[
0 g (8, ,) = Pex
1 Ig ( x, y) exp 2 i (8 x + ,y) dx d y ] (1-138a)
1
( x, y)
= Pex
1 (
I x p , yp ) [ ]
exp 2 i (8 x + ,y) dx d y
(
x p , yp )
( ) [
= Pex
1 I x p , y p exp 2 i (8 x + ,y) dx p d y p ] . (1-138b)
Note that it is properly normalized since 0 g (0, 0) = 1. Like the diffraction OTF, it is also
Hermitian.
Substituting for the ray aberrations ( x, y) in terms of the slope of the aberration
function [see Eq. (3-11) of Part I], namely,
42 IMAGE FORMATION
( x, y) = R
( ,
)
W x p , y p W x p , y p ( ) , (1-139)
x p y p
we obtain
" W x p , y p ( )
W x p , y p ( ) $" dx
0 g (8, ,) =
( )
Pex
1 I x p , y p exp !2 i R 8
x p
+,
y p
%
"
p d yp ,
"# &
(1-140)
which is the same as Eq. (1-136b). The refractive index of the image space is assumed to
be unity here, which is generally the case in practice. If this is not true, then the right-
hand side of Eq. (1-139) is divided by the refractive index. Also in that case, the
wavelength of the object radiation is divided by the refractive index in all of the
diffraction equations.
To obtain an approximate expression for the geometrical OTF for small spatial
frequencies, we write Eq. (1-138b) in terms of the centroid ( xc , yc ) of the aberrated PSF
discussed later [see Eq. (1-177)] in the form
[
0 g (8, ,) = Pex
1 exp 2 i(8 xc + ,yc ) ]
( ) { [
× I x p , y p exp 2i 8( x
xc ) + ,( y
yc ) dx p d y p ]} . (1-141)
Expanding the exponential under the integral in a power series and retaining only the first
three terms for small values of (8, ,) (the second term vanishes upon integration), we
obtain
0 g (8, ,) ~ Pex
1 exp[2 i(8 xc + ,yc )]
(
× I x p , yp ) {1
2 2 [8( x
xc ) + ,( y
yc )] 2 } dx p d y p . (1-142)
~ 1
2
2
0 g (8, ,) I ( x p , y p ) [8( x
xc ) + ,( y
yc )]
2
dx p d y p (1-144a)
Pex
and
We note that the geometrical PTF depends on the pupil irradiance only through the
centroid of the PSF. The PTF for a symmetric aberration is zero, as expected, since the
centroid for such an aberration lies at the origin.
It is evident from Eq. (1-144a) that the slope of the geometrical MTF at the origin is
zero:
0 g (8, ,) 0 g (8, ,)
= 0 = . (1-145)
8 ,
8 = 0 = , 8 = 0 = ,
This result can also be obtained directly from Eq. (1-138b). As shown in Section 1.6.4,
the slope of the diffraction MTF at the origin, which is equal to the corresponding slope
of the real part of the OTF, is nonzero.
If we write the wave and ray aberrations in the ( p, q ) coordinate system, Eq. (1-
138b) reduces to
( )
0 g (vi , 9) = Pex
1 I ( p, q) exp 2 ivi x9 dp d q , (1-146a)
where
W ( p, q )
x9 = R (1-146b)
p
is the component of the ray aberration along the p axis. It may also be obtained from the
ray aberration ( x, y) according to
( ) [ (
0 g (vi , 9) = Pex
1 exp 2 ivi x 9 I ( p, q ) exp 2 ivi x9
x9 )] dp d q , (1-148a)
where
x9 = Pex
1 I ( p, q ) x9 dp d q (1-148b)
is the mean value of x 9 . For small values of the radial spatial frequency vi , we expand
the exponential under the integral in Eq. (1-146a) and retain only the first three terms (the
second term vanishes upon integration) and obtain
( )
2
0 g (vi , 9) ~ 1
2 2 vi2 x9
x9 (1-149a)
and
44 IMAGE FORMATION
7(vi , 9) ~ 2 vi x9 . (1-149b)
Thus, the variance of the ray aberration along the p axis determines the reduction of the
MTF along that axis, and its mean value determines the PTF. It is evident from Eq. (1-
149a) that the slope of the MTF 0 g (vi , 9) with respect to the radial frequency vi
evaluated at the origin is zero.
It should be noted that the diffraction OTF depends strongly on the wavelength of the
object radiation. The cutoff frequency, for example, is inversely proportional to the
wavelength. Since the wave aberration of a ray as an optical path length error in units of
wavelength will be different for different wavelengths, the aberrated OTF will also vary
with the wavelength. However, the geometrical OTF depends on the ray aberrations,
which are independent of the wavelength as long as the optical path length errors are
independent of it. If the path length errors of a system vary with the wavelength, i.e., if it
has chromatic aberration, then the ray aberrations will also vary with it. Thus, for
example, the diffraction OTF of a system consisting of mirrors will vary with the
wavelength, but the geometrical OTF will not. In a system using lenses and suffering
from chromatic aberration, there will be some variation of the geometrical OTF as well
with the wavelength.
1.7 Asymptotic Behavior of PSF 45
Theorem 13. The PSF of a system with a radially symmetric pupil function behaves
asymptotically as the inverse cube of the distance from its center, independent of the
aberration.
For an imaging system with a radially symmetric pupil function, its PSF may be
obtained from its OTF according to [see Eq. (1-84)]
where 0 (vi ) is a real function. Willis12 has shown that for a function f ( x ) whose
derivatives exist
;
1 f (0) 3 f iv (0)
m f ( x ) J 0 ( mx ) dx ~ f (0)
2 + 2 K (1-151)
2 m 2 2! m 4
0
[
f ( n ) (0) = n f ( x ) x n ] x=0
. (1-152)
Letting x = vi , f (vi ) = vi 0 (vi ) and m = 2 ri , Eq. (1-151) can be written for large values
of ri :
0 ( 0 ) 30 (0)
PSF (ri ) ~
2 3 +
... . (1-153)
4 ri 32 4 ri5
Equation (1-153) gives an asymptotic representation of the PSF in terms of the properties
of its Fourier transform, the OTF, at the origin.
PSF (ri ) ~
0 (0) 4 2 ri3 . (1-154)
However, as shown in Section 1.6.4, 0 (0) is independent of aberration. Hence, for large
values of ri , the PSF is independent of an aberration and varies with ri as ri
3 (Theorem
13). Tatian’s results for circular pupils show that Eq. (1-154) holds even for rotationally
nonsymmetric aberration.13 This result is plausible, because the effect of an aberration is
to reduce the irradiance at the center of the image and increase it at nearby points. This
does not imply that the minimum value of ri above which the PSF is (approximately)
described by an ri
3 dependence is the same whether or not the imaging system is
aberrated, but that for large enough values of ri , the point-spread function depends on ri
through ri
3 even when the system is aberrated.
1.7 Asymptotic Behavior of PSF 47
Theorem 14. The encircled power for large circles is independent of the aberration.
where I (ri , i ) is the irradiance at a point (ri , i ) in the image plane. Following Eqs. (1-
61) and (1-76), we may substitute for I (ri , i ) in terms of the corresponding transfer
function 0 (vi , 9) , and thus write Eq. (1-155) in the form
rc 2 2
Pi (rc ) = Pex ri dri d i vi dvi
0 0
0 (vi , 9) exp [
2 iri vi cos (i
9)] d9 . (1-156)
0
Changing the order of integration, the integration over i gives 2 J 0 (2 ri vi) , and noting
that
x0
J ( mx ) x dx = x 0 J ( mx ) ,
0 1 0 (1-157)
m
0
the integration over ri gives (rc vi ) J1 (2 rc vi ) , where J1 ( ) is the first-order Bessel
function of the first kind. Thus, Eq. (1-156) reduces to
2
Pi (rc ) = Pex rc d9 0 (vi , 9) J1 (2 rc vi ) dvi . (1-158)
0
Since the left-hand side is real, the integral over the imaginary part of 0 (vi , 9) must be
zero. Hence, we may consider only the real part of 0 (vi , 9) in Eq. (1-158) and write it in
the form
2
Pi (rc ) = Pex rc d9 Re 0 (vi , 9) J1 (2 rc vi ) dvi .
0 (1-159)
for large positive values of m. Therefore, for large values of rc , Eq. (1-159) can be
written
2
" Re 0 (0, 9)
Pi (rc ) ~ Pex d9 ! +
Re 0 (0, 9)
[ ] [
Re 0 (0, 9)
+ ...
$"
% ,
] (1-161)
" 2 4 2 rc 32 3r 3c "&
#
0
is the nth radial derivative of Re 0 (vi , 9) evaluated at the origin. Equation (1-162) gives
an asymptotic representation of the encircled power in terms of the radial derivatives of
the real part of the OTF at the origin. Since the transfer function is normalized to unity at
the origin, the first term on the right-hand side is Pex . Note that the equation is properly
normalized since Pi (rc ;) = Pex . For large values of rc , we may neglect higher-order
terms and write
2
Pi (rc ) (
~ Pex 1 + 1 4 2 rc
) [Re 0 (0, 9)] d9 . (1-163)
0
Since the integrand in Eq. (1-163) is independent of aberration [see Eq. (1-133)], we
conclude that, up to the first order, the encircled power for large circles is also
independent of an aberration (Theorem 14).
(r )
We now show that, for an unapodized system [i.e., one for which A rp = A0 across
its exit pupil], the integral on the right-hand side of Eq. (1-163), and therefore the asymp-
totic behavior of the encircled power, is related simply to the ratio of the perimeter length
and the transmitting area of the exit pupil.14 From Eq. (1-73b), we note that the OTF of an
r
unapodized aberration-free system corresponding to a spatial frequency vi is equal to the
r
fractional area of overlap of two pupils whose centers are separated by R vi . Thus, from
Figure 1-8, we note that the OTF may be written
r
0 ( vi ) = 1
( R 2 Sex ) vi cos ds , (1-164)
s
where ds is a differential length segment along the perimeter of the pupil and is the
r
angle between vi and the normal to the segment. The factor of 2 accounts for the fact that
the nonoverlap area is counted twice in the integral in Eq. (1-164). For very small values
of vi (since we are interested in the slope of the OTF at the origin), it may be considered
constant and brought outside the integral. Hence, for such values, we may write
r
0 ( vi ) = 1
( Rvi 2 Sex ) cos ds . (1-165)
s
1.7 Asymptotic Behavior of PSF 49
,
ds cos
n^
O
R6 i 9
8
O
Figure 1-8. Geometry for calculating the OTF of a system for very small spatial
r
frequencies. The displaced pupil is centered at O , whose position vector is Rvi
making an angle 9 with respect to the pupil centered at O. Note that n̂ is a unit
vector along the normal to the pupil at the point where the line joining O and O
intersects it. The overlap area of the two pupils is shown shaded.
Since d = d9 , we obtain
2 2
0 ( vi , 9) d9 = 2
( RLvi 2 Sex ) cos d
0 0
= 2
2 RLvi Sex , (1-166)
where L is the length of the perimeter of the exit pupil. Taking the derivative of both sides
with respect to vi , we obtain for very small values of vi and in the limit vi 0 :
2
2 RL
0 (0, 9) d9 =
S . (1-167)
0 ex
Since the slope of the real part of the OTF at the origin is independent of aberration, Eq.
(1-167) is valid for aberrated but unapodized systems as well. Substituting this equation
into Eq. (1-163), we obtain
(
P (rc ) ~ Pex 1
RL 2 2 rc Sex ) . (1-168)
Thus, we see that the asymptotic behavior of encircled power depends on the ratio of the
perimeter length L of the exit pupil and its transmitting area Sex (Theorem 15). The
effect of an aberration is to increase the value of rc for which Eq. (1-168) is valid. The
larger the aberration, the larger the value of rc required for the validity of Eq. (1-168).
50 IMAGE FORMATION
Theorem 16. (a) The centroid of the diffraction PSF is given by the slope of the
imaginary part of the its diffraction OTF at the origin. (b) It is the same as the centroid of
the ray geometrical PSF. (c) The centroid of an aberration-free system lies at the
Gaussian image point regardless of the amplitude variations across its exit pupil. (d) For
a system with an aberrated but unapodized exit pupil, the centroid can be obtained from
its aberration only along its perimeter.
R R
) (
x p xi + y p yi dx p dyp
) , (1-169)
where ( xi , yi ) are the coordinates of a point in the image plane with respect to the
( )
Gaussian image point, and x p , y p are the coordinates of a point in the plane of the exit
pupil. By dividing both sides of Eq. (1-169) by Pex we obtain the function PSF ( xi , yi )
for the system.
xi = Pex
1 xi Ii ( xi , yi ) dxi dyi (1-170a)
and
yi = Pex
1 yi Ii ( xi , yi ) dxi dyi . (1-170b)
From the definition of the PSF of an imaging system, it should be evident that its centroid
is synonymous with that of the irradiance distribution of the image of a point object.
0 (8, ,) = Pex
1 Ii ( xi , yi ) exp [2 i (8 xi + ,yi )] dxi dyi , (1-171)
r
where (8, ,) are the rectangular components of a spatial frequency vector v in the image
plane. Differentiating both sides of Eq. (1-171) with respect to x and evaluating the result
at 8 = , = 0 , we find that
1 0
xi = . (1-172a)
2 i 8 8 = , = 0
1.8 PSF Centroid 51
Similarly,
1 0
yi = . (1-172b)
2 i , 8 = , = 0
However, since xi and yi are real, only the slope of the imaginary part of the OTF at
the origin contributes to the centroid. Thus, we may write
1 Im 0
xi = (1-173a)
2 8 8 = , = 0
and
1 Im 0
yi = . (1-173b)
2 , 8 = , = 0
Thus, the centroid of the PSF of an optical system is given by the slope of the imaginary
part of its OTF at the origin (Theorem 16a). It can not, for example, be obtained from a
knowledge of only the MTF of the system.
( ) (
0 (8, ,) = Pex
1 P x p , y p P * x p
R8 , y p
R , dx p dy p ) . (1-174)
R P * x p , y p ( ) dx
xi =
Im P x p , y p
2 Pex
( x p
)
p dy p (1-175)
and a similar equation for yi . The pupil function given by Eq. (1-38) or Eq. (1-47) may
be written in the form
( ) ( ) [
P x p , y p = A x p , y p exp (2 i ) W x p , y p ( )] . (1-176)
W x p , y p ( )
xi =
R
Pex
(
Ip x p , yp )
x p
dx p dy p , (1-177a)
( ) ( )
where I p x p , y p = A 2 x p , y p is the irradiance at a pupil point x p , y p . Similarly, ( )
W x p , y p ( )
yi =
R
Pex
I x ,y
p p p ( y p
) dx p dy p . (1-177b)
52 IMAGE FORMATION
( ) ( )
From Eq. (1-142), R W x p and R W y p represent the ray aberrations, i.e., the
( )
image-plane coordinates of a ray passing through the pupil point x p , y p . Hence, Eqs.
(1-177) show that the centroid of the PSF according to wave diffraction optics is identical
with that according to ray geometrical optics (Theorem 16b).
From Eqs. (1-177), we also note that amplitude variations across the pupil affect the
centroid only if it is aberrated. In the absence of aberrations, the PSF centroid lies at (0,
0); i.e., it lies at the Gaussian image point where the center of curvature of the spherical
wavefront lies, regardless of the shape of the pupil and/or the amplitude distribution
across it (Theorem 16c). This may also be seen from Eqs. (1-169) and (1-176). We note
( )
from these equations that if W x p , y p = 0 , then Ii ( xi , yi ) = Ii (
xi ,
yi ) . Hence, the
symmetry of the aberration-free PSF yields its centroid at the Gaussian image point.
Similarly, since in that case the aberration-free OTF is real [see Eq. (1-174)], Eqs. (1-173)
also give the centroid at the Gaussian image point.
Equations (1-170), (1-173), and (1-177) give the centroid in terms of the PSF, OTF,
and the aberration function, respectively. In practice, given an imaging system, the most
convenient expression to use would be Eqs. (1-170), since the PSF can be measured by
using a photodetector array. In optical design and analysis, the simplest way to obtain the
centroid would be to use Eqs. (1-177) since the aberrations must be calculated even if the
other two expressions were used. Thus, one may trace rays all the way up to the image
plane and determine the centroid of the ray distribution in this plane with appropriate
( )
weighting I p x p , y p of each ray.
( )
A x p , y p = A0 (1-178)
( )
I p x p , y p = A02
(1-179)
= Pex Sex ,
=
R
(
W x p , y p )
xi dx p dy p (1-180a)
Sex x p
and
=
R
(
W x p , y p )
yi dx p dy p . (1-180b)
Sex x p
1.9 Strehl Ratio 53
Using Stokes theorem,16 the surface integrals in Eqs. (1-180) involving the derivative of
the aberration function can be written in terms of its line integral along the curve
bounding the surface. Thus, we may write
r
W ( x , y ) xˆ d s
R
xi = p p p (1-181a)
Sex
and
r
yi
( )
= ( R Sex ) W x p , y p yˆ p d s , (1-181b)
r
where x̂ p and ŷ p are unit vectors along the x p and y p axes, respectively, and d s
represents an element of arc length vector along the perimeter of the pupil. It is evident
from Eqs. (1-181) that, in the case of an aberrated but an unapodized pupil, the centroid
of the PSF can be obtained from the value of the aberration function only along the
perimeter of the pupil (Theorem 16d). Accordingly, in that case, to calculate the centroid
the knowledge of the aberration across the interior of the pupil is not needed.
Theorem 17. (a) The central irradiance for an apodized-aberrated system is less
than or equal to the corresponding value for an unapodized-unaberrated system. (b) For
a given total power, any amplitude variations reduce the central irradiance and any
phase variations further reduce it. (c) The peak value of an unaberrated PSF lies at the
center, regardless of the amplitude variations across the pupil.
54 IMAGE FORMATION
The irradiance distribution of the (defocused) image of a point object is given by Eq.
r
(1-60). For simplicity of notation, we let ri be the position vector of the observation point
r
( )
with respect to the point zi zg rg lying on the line joining the center of the exit pupil
and the Gaussian image point in the observation plane at a distance zi . Thus, the
irradiance distribution given by Eq. (1-60) may be written
2
r 1 r 2 i r r r
I ( ri ; zi ) = 2 2 P rp ; zi exp
zi zi
(rp r d rp
) , (1-182)
(r )
where P rp ; zi is the defocused pupil function given by Eq. (1-47). The irradiance at
the center of the distribution for an apodized and aberrated pupil is obtained by letting
r
ri = 0 . Thus,
r r
P ( rp ; zi ) d rp
1 2
Iaa (0; zi ) = . (1-183)
2 zi2
Similarly, for the same total power Pex in the image, the central irradiance at the
Gaussian image point for an unapodized and unaberrated system may be obtained from
r r r
( )
Eq. (1-183) by letting zi zg , A rp = A0 and W rp = 0 , so that P rp = A0 . Thus, ( ) ( )
we may write
(
Iuu 0; zg = ) Pex Sex
2 zg2
(1-184)
r r
( )
Sex 2
= P rp ; zi d rp . (1-185)
2 zg2
The ratio of the central irradiance in the defocused plane for an apodized and
aberrated pupil, and in the Gaussian image plane for the unapodized and unaberrated
pupil may be written
Iaa (0; zi )
2
zg
= Saa , (1-186)
(
Iuu 0; zg ) zi
where
r r
P ( rp ) d rp
2
Saa = r r (1-187)
( )
2
Sex P rp d rp
is the corresponding Strehl ratio. It represents the ratio of the central irradiances in the
defocused and Gaussian image planes, except for the inverse-square-law effect
( )
2
represented by the factor zg zi . This factor is unity if the aberrated irradiance is also
observed in the Gaussian image plane. It is practically equal to unity in imaging systems
because of their small depth of focus. However, it can be significantly different from
1.9 Strehl Ratio 55
(r ) (r ) (r )
unity in laser transmitters. Letting f rp = P rp , g rp = 1, and n = 2 in Hölder’s
inequality (1-104), we find that
Saa ' 1 . (1-188)
(r )
P rp = A rp (r ) , (1-189)
( ) ( ) [ A (rr ) d rr ]
2
Iau 0; zg = 1 2 zg 2 p p . (1-190)
(
Sau = Iau 0; zg ) Iuu (0; zg )
[ A (rrp ) rd rrp ]r .
2
= (1-191)
Sex A 2 ( rp ) d rp
Thus, any amplitude variations across the pupil of an aberration-free system reduce the
central irradiance. For example, as discussed in Section 4.3, a pupil with a Gaussian
illumination across it yields a smaller value of the central irradiance than a pupil with a
uniform illumination, the total power being the same in the two cases.
Comparing Eq. (1-191) with Eq. (1-187), and using Hölder’s inequality (1-104) with
r r r
( ) ( ) ( )
f rp = P rp , g rp = 1, and n = 1, we find that
r r 2
Saa P ( rp ) d rp
=
[ A (rrp ) d rrp ]
2 (1-193a)
Sau
' 1 . (1-193b)
56 IMAGE FORMATION
From Eqs. (1-192) and (1-193b), we note that amplitude variations reduce the central
irradiance, and phase variations (i.e., aberrations) further reduce it. Note, however, that
an irradiance reduced by phase variations alone does not necessarily reduce any further
if amplitude variations are also introduced (Theorem 17b). In fact, amplitude variations
can even increase this irradiance. The maximum value of central irradiance is obtained
when the system is unapodized and unaberrated.18 Hence, the maximum value of
irradiance at a certain point in the image space is obtained when a beam exiting from the
system is focused at it with uniform amplitude and phase at its exit pupil. Any variation
in the amplitude or phase across the exit pupil reduces the value at the chosen point.
The peak value of the aberrated irradiance distribution of the image of a point object
does not necessarily occur at the center of the reference sphere. However, the peak value
of its unaberrated image does occur at this point regardless of the apodization. The
Huygens’ spherical wavelets emanating from the spherical wavefront are equidistant from
this point. Hence, they interfere constructively, producing a maximum possible value at
r r
( ) ( )
this point. Mathematically, this may be seen by letting zi = zg , and P rp = A rp in
r r
Eq. (1-182), and comparing the irradiance at a certain point ri with that at ri = 0 . Thus,
we may write
r r r
f ( rp ) d rp
2
Iau ( ri )
=
Iau (0)
[ (r ) r
]
2
f rp d rp
' 1 , (1-194)
where
r r 2 i r r
( ) ( )
f rp = A rp exp
zg
rp ri .
(1-195)
Hence, the peak value of an unaberrated image occurs at the Gaussian image point
regardless of the apodization of the system (Theorem 17c).
Theorem 18. For small aberrations, the Strehl ratio of an image is determined by
the variance of the phase aberration across the exit pupil.
The ratio of the central irradiance at a distance zi with aberration and that at the
Gaussian image point without aberration is given by
1.9 Strehl Ratio 57
Iaa (0; zi )
2
zg
= S , (1-196)
(
Iau 0; zg ) zi
where
r r r
A (rp ) exp [ i2 (rp )] d rp
2
S =
[ (r ) r (1-197)
]
2
A rp d rp
is the Strehl ratio of the image. The right-hand sides of Eqs. (193a) and (1-197) are
identical. Hence,
S ' 1 . (1-198)
This Strehl ratio gives a measure of the irradiance reduction due to aberration in the
system and or due to defocus. It can be written in an abbreviated form
2
S = exp (i 2) , (1-199)
where the angular brackets L indicate a spatial average over the amplitude-weighted
pupil, e.g.,
r r r
2 =
A ( rp ) 2 ( rp ) d rp
[ A (rrp ) d rrp ]
2 . (1-200)
r
Since 2 is independent of rp , Eq. (1-199) can be written
[ )]
2
S = exp i ( 2
2
= cos (2
2 ) + sin (2
2 )
2 2
3 cos (2
2 )
2 (1-201)
,
equality holding when 2 is zero across the pupil, in which case S = 1. For small
aberrations, expanding the cosine function in a power series and retaining the first two
terms, we obtain the Maréchal result generalized for an apodized pupil
(1
? 2)
2
S >~ 2
2 , (1-202)
where
? 22 = (2
2 )2 (1-203)
For small values of ? 2 , three approximate expressions have been used in the
literature:
~ (1
? 22 2)
2
S1 , (1-204)
S2 ~ 1
? 22 , (1-205)
and
S3 ~ exp (
? 22 ) . (1-206)
The first is the Maréchal formula, 19 the second is the commonly used expression obtained
when the term in ? 2 4
in the first is neglected, 20 and the third is an empirical expression
giving a better fit to the actual numerical results for various aberrations21 as we shall see
in Sections 2.2.4 and 3.2.4. The simplest expression to use is, of course, S2 , according to
which ? 22 gives the drop in the Strehl ratio. We note that the Strehl ratio for a small
aberration does not depend on its type but only on its variance across the apodized pupil
(Theorem 18). For a high-quality imaging system, a typical value of the Strehl ratio
desired is 0.8, corresponding to a wave aberration standard deviation of ? w = 14 .
Second, the Strehl ratio may be determined from the PSF of the system. We calculate
its unaberrated PSF from its pupil shape and amplitude distribution, and normalize it so
that its central value is unity. Next, we integrate it in its plane to give the total power.
Finally, we measure the aberrated PSF and normalize it so that its integral has the same
value as for the aberration-free case. This step insures that the total power in the aberrated
image is the same as in the aberration-free image. The central value of the aberrated PSF
normalized in this manner gives the Strehl ratio.
Third, the Strehl ratio may be determined from the OTF of the system. Since its PSF
and OTF form a Fourier transform pair [see Eq. (1-77)], its Strehl ratio may be written
integrating the real part of the measured aberrated OTF over all spatial frequencies and
dividing it by a similar integral of the calculated unaberrated OTF. In any of these three
approaches for determining the Strehl ratio, one must take into account the effect of the
inverse-square law, i.e., the effect of zi being different from zg , unless they are
practically equal to each other (see Section 2.5).
Theorem 19. (a) For small values of the phase aberration difference function, the
r
Hopkins ratio for a certain spatial frequency vi is determined by the variance of the
difference function across the overlap area of two pupils displaced with respect to each
r
other by Rvi . (b) The mean value of the phase aberration difference function represents
the corresponding PTF.
r
Following Hopkins,22 we define a modulation ratio H ( vi ) as the ratio of the MTFs
r r r
0 ( vi ) and 0 u ( vi ) of a system at a spatial frequency vi with and without aberration, i.e.,
r r r
H ( vi ) = 0 ( vi ) 0 u ( vi ) . (1-208)
r
For obvious reasons, we call H ( vi ) the Hopkins modulation (or contrast) ratio. From Eq.
(1-107), this ratio is less than one. From Eq.(1-73c) it can be written
r r r
r A(r ) A(r
p p ) {[ (r )
Rvi exp i 2 rp
2 rp
R vi (r r
)] } d rrp
H ( vi ) = r r r r . (1-209)
( ) (
A rp A rp
Rvi d rp )
In the rotated ( p, q ) coordinate system shown in Figure 1-4, Eq. (1-209) may be written
r r
r
H ( vi ) =
I ( p, q; vi ) exp [iQ ( p, q; vi )] dp dq
r , (1-210)
I ( p, q; vi ) dp dq
r
where the mutual irradiance function I ( p, q; vi ) and the phase aberration difference
r
function Q( p, q; vi ) are given by Eqs. (1-128). As in Eq. (1-125), a function in the ( p, q )
coordinate system is obtained from the corresponding function in the x p , y p coordinate ( )
system by replacing x p with p cos 9
q sin 9 and y p with p sin 9 + q cos 9 . As discussed
( )
in Section 3.5 of Part I, the aberration function 2 x p , y p for a rotationally symmetric
60 IMAGE FORMATION
where the angular brackets indicate an average across the overlap region of the two
apodized pupils. For example,
r r
Q =
I ( p, q; vi ) Q(rp, q; vi ) dp dq . (1-213)
I ( p, q; vi ) dp dq
For small values of Q
Q , we may retain only the first three terms in the expansion of
the exponential in Eq. (1-212), one of which reduces to zero upon averaging. Thus, we
obtain the result
r ~ 1 2
H ( vi ) 1
?Q , (1-2150
2
where
2
? Q2 = Q 2
Q (1-215)
is the variance of the phase aberration difference function across the overlap region of the
r
[
two apodized pupils. Multiplying both sides of Eq. (1-208) by exp i 7( vi ) , we obtain]
[see Eq. (1-96)]
r r r r
[ ]
H ( vi ) exp i 7( vi ) = 0 ( vi ) 0 u ( vi )
{
= exp i [Q
Q ]} exp (i Q )
~ 1
? Q2 exp (i Q ) .
1
(1-216)
2
We noted in Section 1.9.2 that, for small aberrations, the Strehl ratio of an aberrated
system depends on the variance of the aberration function across its pupil and not on the
type of the aberration. Similarly, we note from Eq. (1-214) that, for small values of the
phase aberration difference function, the Hopkins ratio for a certain spatial frequency
1.11 Line- and Edge-Spread Functions 61
depends on the variance of the difference function across the overlap region of two
displaced pupils (displacement depending on the spatial frequency) and not on the type of
the aberration (Theorem 19a). Similarly, Eq. (2-217) shows that the mean value of the
phase aberration difference function represents the PTF (Theorem 19b).
r ~ Ê 1 ˆ
H (v i ) exp Á - s Q2 ˜ . (1-218)
Ë 2 ¯
Theorem 20. (a) The aberration-free line-spread function is symmetric about its
center, and the aberrations reduce its central value. (b) The line-spread function is equal
to the derivative of the edge-spread function.
'
= 0 PSF( xi , yi )dyi . (1-219a)
<'
For a line object along the yo axis, the LSF depends only on xi , i.e., its variation with xi
is the same regardless of the value of yi , as expected for an isoplanatic line object. Since
r
( ) ( )
its Gaussian image Ig rg = b x g , its diffraction image can also be obtained by applying
Theorem 5 as the convolution of its Gaussian image and the PSF, i.e.,
' '
<' <'
( )
LSF( xi ) = 0 0 b x g PSF( xi < x g , yi < yg ) dx g dyg . (1-219b)
It should be evident that the LSF can be obtained from the PSF by scanning it with a long
and narrow slit. If the slit is parallel to the yi axis, the LSF thus obtained is LSF( xi ) .
62 IMAGE FORMATION
The LSF of a system can be written in terms of its pupil function by substituting for
its PSF from Eq. (1-57) into Eq. (1-219) . Thus
' GG 2 /i
LSF( xi ) =
1
(
2 2 0 dyi P x p , y p exp ³ <
Pex h R < ' H H h R
) (
xi x p + yi y p µ d x p dy p
)
GG 2/i
HH
(
× P * x vp , y vp exp ³
h R
) (
xi x vp + yi y vp µ d x vp dy vp
)
GG £ 2 /i ¥ GG £ 2 /i ¥
HH
( )
= P x p , y p exp ² <
¤ h R
xi x p ´ d x p dy p P * x vp , y vp exp ²
¦ HH ¤ h R
( ¦
)
xi x vp ´ d x vp dy vp
'
G 2 /i
×
1
Pex h2 R 2
H
exp ³
h R
yi y vp < y p µ dyi
( )
<'
2
G £ 2 /i ¥
=
1
Pex h R H
(
0 dy p P x p , y p exp ² <
¤ h R
)
xi x p ´ d x p
¦
, (1-220)
where in the last step we have used the fact that the integral over yi is equal to
( )
h Rb y vp < y p and thereby carried out the integration over y vp . Hence, the LSF
normalized by its aberration-free value at the origin, is given by
2
G £ 2 /i ¥
H
(
0 dy p P x p , y p exp ² <)
¤ hR
xi x p ´ d x p
¦
LSF( xi ) = . (1-221a)
[0 ( ) ]
2
0 dy p P x p , yp d x p
The aberration-free LSF is symmetric about xi = 0 . The central value of the LSF
normalized by its aberration-free value is called the Struve ratio. It is given by
( )
2
0 dy p 0 P x p , y p d x p
LSF(0) = (1-221b)
0 dy p [ 0 P( x p , y p ) d x p ]
2
)1 , (1-221c)
where in the last step we have used Hölder’s inequality (1-104) with f x p = P x p , ( ) ( )
( )
g x p = 1, and n = 2 . Thus, aberrations reduce the central value of the LSF, as in the case
of the PSF.
If the line object lies along a yov axis (or parallel to it), making an angle ` with the
yo axis, as illustrated in Figure 1-9, then the LSF( xiv) is obtained by writing the pupil
( )
function in the coordinate system x vp , y vp by replacing x p by x vp cos ` < y vp sin ` and y p
by x vp sin ` + y vp cos ` .
1.11 Line- and Edge-Spread Functions 63
yo
yov
xov
xo
ypv
xpv
xp
yiv
xiv
xi
Figure 1-9. Coordinate system in various planes for determining the LSF. (a) Object
plane. (b) Pupil plane. (c) Image plane. The line object lies along the yov axis or
parallel to it, making an angle ` with the yo axis.
64 IMAGE FORMATION
xi '
= 0 dx 0 PSF ( x , y ) dy , (1-222b)
<' <'
where in the last step we have substituted for the LSF from Eq. (1-219a). It is evident
from Eq. (1-222a) that the LSF is the derivative of the ESF, i.e.,
d
LSF( xi ) = ESF( xi ) . (1-223)
dxi
Moreover, it is evident from Eq. (1-222b) that ESF(') = 1, as may be seen by the use of
Eqs. (1-61) and (1-63). It may also be seen by noting that the integral for xi = ' is equal
to o(0, 0) , which, in turn, is equal to unity.
Introducing the step or the edge function representing the edge object,
H ( x ) = 1 for x * 0
(1-224)
= 0 for x < 0 ,
'
= 0 PSF( x, y) H ( xi < x ) dx dy . (1-225b)
<'
Thus, ESF( xi ) is equal to a 1-D convolution of the edge object H ( xi ) with LSF( xi ) , or a
2-D convolution with PSF( xi , yi ) , as expected for an isoplanatic extended object.
PSF. The PSF may also be measured using a two-dimensional array of small
photodetectors. But, again, precise measurements of the PSF outside the bright spot may
be difficult owing to their low values. The LSF can be obtained by scanning a
corresponding PSF with a long and narrow slit. The difficulty of a low signal can be
overcome by using a line object and measuring the LSF by scanning its image with a long
and narrow slit. The slit is parallel to the line object and its scan direction is normal to it.
By removing one side of the slit used as a line source, the ESF can be measured in a
similar manner.
Theorem 21. (a) The LSF and the OTF form a 1-D Fourier pair. (b) Similarly, the
ESF and the OTF divided by a spatial frequency are related to each other by a 1-D
Fourier transf orm.
Substituting for PSF( x, y) in terms of its Fourier transform, the OTF o (j, d) [see
Eq. (1-76)], namely,
where, for simplicity, we have dropped the subscript on the image coordinates ( xi , yi ) ,
Eq. (1-219) may be written
'
[ ]
LSF ( x ) = 0 dy 00 o (j, d) exp < 2 / i (j x + d y) dj d d
<'
Thus, LSF( x ) , representing the image of a line object lying along the yo axis, is the 1-D
inverse Fourier transform of o (j, 0) (Theorem 21a). Since LSF( xi ) is a real function,
only the real part of the integrand contributes to the integral; the imaginary part yields
zero upon integration. Hence, Eq. (1-227a) may be written
Similarly, by considering a line object along the x axis, we can show that its image
LSF( y) is the inverse Fourier transform of o (0, d) , i.e.,
The LSF for a line object making an angle ` with the yo axis is given by
66 IMAGE FORMATION
where x v and j v are parallel to each other and make an angle ` with the x axis in the
image plane, and o (jv, 0) is simply the OTF along the direction j v . It should be clear
from Eqs. (1-227) through (1-229) that the LSF represents the variation of irradiance in
the image of a line object along a direction that is normal to the direction of the object.
Now we consider an edge object that is parallel to the yo axis. To relate the ESF to
the OTF, we Fourier transform both sides of Eqs. (1-225). Since the integrals on their
right-hand sides represent convolutions, their Fourier transforms are equal to the product
of the Fourier transforms of the convolving functions. From Eq (1-226), the Fourier
transform of LSF( x ) is given by
since o(j, 0)b(j) = o(0, 0) = 1 . Fourier transformation of Eq. (1-225b) also yields Eq. (1-
232). The left-hand side of Eq. ( 1-232) represents the spatial-frequency spectrum of the
image of an edge object lying parallel to the yo axis, and its right-hand side represents the
product of the spatial-frequency spectrum of the edge object (or its Gaussian image) and
the system OTF for d = 0 , as expected for an isoplanatic object according to Eq. (1-70).
Inverse Fourier transforming Eq. (1-232), we obtain the ESF in terms of the OTF:
1 1 G o( j , 0 )
ESF( x ) = < exp ( < 2 /i j x ) d j . (1-233)
2 2 /i
H j
For an incoherent object, ESF( x ) represents the irradiance distribution of the image of an
edge object. It is, therefore, a real function. Hence, only the imaginary part of the
integrand in Eq. (1-233) contributes to the integral; the real part yields zero upon
integration. Accordingly, Eq. (1-233) may be written
1 G sin(2 /j x ) G cos(2 /j x )
ESF( x ) = + Re o(j, 0) d j < Im o(j, 0) dj . (1-234)
2 H 2 /j H 2 /j
Equations (1-232) and (1-233) show that the ESF( x ) and o(j, 0) j are related to each
other by a 1-D Fourier transform (Theorem 21b). Letting x = 0 in Eq. (1-234), we obtain
1 G Im o(j, 0)
ESF(0) = < dj , (1-235)
2 H 2 /j
1.12 Shift-Invariant Imaging of a Coherent Object 67
i.e., the value of the ESF at the origin is determined solely by the imaginary part of the
OTF o(j, 0) . If the OTF is real, e.g., for a symmetric aberration, then ESF(0) = 1 2
independent of the aberration. If the edge object makes an angle ` with the yo axis, then
the ESF is given by an equation similar to Eq. (1-233), except that x and j axes are
replaced by x v and j v axes as in Eq. (1-229).
Equations (1-227) and (1-233) for the LSF and ESF of a system in terms of its OTF
are also applicable to coherent imaging, provided the OTF is replaced by the
corresponding coherent transfer function (CTF) discussed in Section 1.12.2. This is done
in Section 2.15 for systems with circular pupils.
Theorem 23. (a) The complex amplitude image of an isoplanatic coherent object is
equal to the convolution of its Gaussian amplitude image and its coherent spread
function. (b) Accordingly, the spatial frequency spectrum of the diffraction amplitude
image is given by the product of the spectrum of the Gaussian amplitude image and the
coherent transfer function.
i £ r2 ¥ G r r £ r2 ¥
= < exp ³ik ² zi < zg + i ´ µ d ro Uo ( ro ) exp ² < ik o ´
h zi ³ ¤ 2 zi ¦ µ H ¤ 2 zo ¦
G r r 2 /i r £ r z r ¥ r
(
× G rp ; ro ; zi exp ³<
³ h zi
) ¤ zg ¦ µ
u
rp ² ri < i rg ´ µ d rp , (1-236)
H
r r r r
where Uo ( ro ) is the amplitude at the object point ro , rg = M ro is the position vector of
the Gaussian image point, zg is the distance of the Gaussian image plane, and
r r r r ik £ 1 1¥
(
G rp ; ro ; zi ) ( )
= G rp ; ro exp ³ ² < ´ rp2 µ
³ 2 ¤ zi zg ¦ µ
(1-237)
r
Uo ( ro ) = U j b ro < rj (r r
) , (1-238)
(r r
Ui ri ; rj ; zi ) = U j CSF ri ; rj ; zi(r r
) , (1-239)
where
r r ¨« 1 £ r 2 rj ¥ ¬«
2
(
CSF ri ; rj ; zi ) = <
i
h zi
( )
exp ©ik ³ zi < zg + ² i + ´ µ
«ª ³ 2 ¤ zi zo ¦ µ «
®
G r r 2 /i r £ r z r¥ r
(
× G rp ; rj ; zi exp ³<
³ h zi
) ¤ zg
u
rp ² ri < i M rj ´ µ d rp
¦ µ
(1-240)
H
determining the defocus aberration, which varies as rp2 . For systems with small fields of
view, we may replace zg by R, the radius of curvature of the Gaussian reference sphere
with respect to which the aberration is defined.
For coherent imaging, we must consider another approximation, namely, that the
r
variation of the quadratic phase factor depending on ro is negligible.25 We will see that
the magnitude of the CSF is significant only in a small region (on the order of the Airy
disc discussed in Section 2.2) surrounding the Gaussian image point. Therefore, at a
r
particular point of observation ( ri ; zi ) , the contributions of the quadratic phase factor will
r r r
come from a small region of the object. Hence, we may replace ro = rg M by ri M in
the object-dependent quadratic phase factor. Any phase factors that do not depend on the
location of the object point can be dropped if it is the irradiance that is measured in the
observation plane.
(r r
CSF ri ; rj ; zi ) = <
i G
hR H
r
( 2 /i r r
G rp ; zi exp ³<)
hR
r r
u(
rp ri < M rj µ d rp
) . (1-241)
r r
We note that the integral on the right-hand side depends on the position vectors ri and rg
of the observation and Gaussian image points, respectively, through their difference
r r r r r r
( ) (
ri < rg only. Hence, we may replace CSF ri ; rj ; zi by CSF ri < M rj ; zi . Thus, the CSF )
is shift invariant in that its form does not depend on the location of the object point; the
r r
whole distribution shifts by virtue of being centered at rg = M rj . Accordingly, for an
isoplanatic coherent object, Eq. (1-236) for the amplitude distribution of its image may be
written
r
(r ) (r
Ui ( ri ; zi ) = M <1 0 Uo rg M CSF ri < rg ; zi d rg
r
) r
r r r r
= 0 Ug ( rg ) CSF( ri < rg ; zi ) d rg , (1-242)
where
r i G r £ 2 /i r r ¥ r
CSF( ri ; zi ) = <
hR
H
( )
G rp ; zi exp ² <
¤ hR
rp ri ´ d rp
¦
u (1-243)
and
(r )
U g rg = M <1 Uo rg M (r ) (1-244)
is the amplitude distribution of the Gaussian image. Thus, the coherent spread function is
proportional to a Fourier transform of the pupil function (Theorem 22), and the image
amplitude distribution for an isoplanatic coherent object is equal to the convolution of the
Gaussian amplitude image (which is an exact replica of the object amplitude except for its
magnification M) and the coherent spread function (Theorem 23a). Comparing Eq. (1-55)
70 IMAGE FORMATION
with Eq. (1-242), we find that the incoherent and coherent point-spread functions are
related to each other according to
r
CSF( ri )
2
r
PSF( ri ) = r r . (1-245)
0 ( )
2
G rp ; zi d rp
where
r r r r r
U˜ g ( vi ) = 0 Ug ( rg ) exp (2/i vi u rg ) d rg (1-247)
r
is the coherent transfer function of the system corresponding to a spatial frequency vi in
the image plane. Substituting Eq. (1-243) into Eq. (1-248) and dropping the constant
phase factor < i, we obtain
r 1 G r r rG r £ 2 /i r r ¥ r
CTF( vi ) =
hR H H
u
d ri exp (2 / vi ri ) G rp exp ² <
¤ hR
( )
rp ri ´ d rp
¦
u
r
r r G r r £ r rp ¥
=
1
hR 0
d rp G rp ( )
³ ¤
u
d ri exp ³2 /i ri ² vi < µ
h R ´¦ µ
H
r
r r £ r rp ¥
=
1
hR 0 ( )
d rp G rp b ² vi <
¤ h R ´¦
r
= G(h Rvi ) . (1-249)
Thus, the coherent transfer function is simply a scaled version of the relative pupil
function (Theorem 24). If the system is diffraction limited, i.e., if it is aberration free, and
r
( )
its transmission across the pupil is uniform so that G rp is equal to a constant across the
pupil (whose value varies inversely with the distance zo of the entrance pupil from the
r
object), then the normalized CTF is unity for those frequencies for which h Rvi lies
inside the pupil.
1.13 Summary of Theorems 71
r
Fourier transforming both sides of Eq. (1-245), substituting for CSF( ri ) from Eq. (1-
243), and proceeding as we did in obtaining Eq. (1-73b), we obtain
r r r r r r r
0 G (r ) G * (r ) 0 G (r )
2
o ( vi ) = p p < h R vi d rp p d rp . (1-250)
Thus, the incoherent transfer function is the (normalized) autocorrelation of the coherent
transfer function. Since the pupil functions P and G are proportional to each other, the
OTF given by Eq. (1-250) is the same as that given by Eq. (1-73b).
where the spectra of the amplitude diffraction and Gaussian images are given by
r r r r r
u
U˜ i ( vi ) = 0 Ui ( ri ) exp (2 / vi ri ) d r (1-252)
and
r r r r r
U˜ g ( vi ) = 0 Ug ( ri ) exp (2 / vi u ri ) d r , (1-253)
respectively. Thus, the spectrum of the diffraction image is given by the product of the
spectrum of the Gaussian image and the CTF (Theorem 23b).
Theorem 3. The Gaussian image of a small object lying at a large distance from an
imaging system is an exact replica of the object, except for its magnification.
Theorem 7. The OTF of an imaging system is also equal to the autocorrelation of its
pupil function.
Theorem 8. The PSF and the OTF of a system with a radially symmetric pupil
function form a zero-order Hankel transform pair.
Theorem 10. (a) The OTF is a complex symmetric function with a value of unity at
the origin. (b) Its magnitude, the MTF, is unity or less at any other spatial frequency. (c)
The aberrated MTF is less than or equal to the corresponding aberration-free value.
Theorem 11. The slope of the real part of the OTF at the origin is independent of the
aberration, and it is equal to the corresponding slope of its MTF.
Theorem 12. The diffraction OTF approaches the geometrical OTF as the
wavelength approaches zero.
Theorem 13. The PSF of a system with a radially symmetric pupil function behaves
asymptotically as the inverse cube of the distance from its center, independent of the
aberration.
Theorem 14. The encircled power for large circles is independent of the aberration.
Theorem 16. (a) The centroid of the diffraction PSF is given by the slope of the
imaginary part of the its diffraction OTF at the origin. (b) It is the same as the centroid of
the ray geometrical PSF. (c) The centroid of an aberration-free system lies at the
Gaussian image point regardless of the amplitude variations across its exit pupil. (d) For
a system with an aberrated but unapodized exit pupil, the centroid can be obtained from
its aberration only along its perimeter.
Theorem 17. (a) The central irradiance for an apodized-aberrated system is less
than or equal to the corresponding value for an unapodized-unaberrated system. (b) For
a given total power, any amplitude variations reduce the central irradiance and any
phase variations further reduce it. (c) The peak value of an unaberrated PSF lies at the
center, regardless of the amplitude variations across the pupil.
Theorem 18. For small aberrations, the Strehl ratio of an image is determined by
the variance of the phase aberration across the exit pupil.
1.13 Summary of Theorems 73
Theorem 19. (a) For small values of the phase aberration difference function, the
r
Hopkins ratio for a certain spatial frequency vi is determined by the variance of the
difference function across the overlap area of two pupils displaced with respect to each
r
other by h Rvi . (b) The mean value of the phase aberration difference function represents
the corresponding PTF.
Theorem 20. (a) The line-spread function is symmetric about its center and the
aberrations reduce its central value. (b) The line-spread function is equal to the
derivative of the edge-spread function.
Theorem 21. (a) The LSF and the OTF form a 1-D Fourier pair. (b) Similarly, the
ESF and the OTF divided by a spatial frequency are related to each other by a 1-D
Fourier transform.
Theorem 23. (a) The complex amplitude image of an isoplanatic coherent object is
equal to the convolution of its Gaussian amplitude image and its coherent spread
function. (b) Accordingly, the spatial frequency spectrum of the diffraction amplitude
image is given by the product of the spectrum of the Gaussian amplitude image and the
coherent transfer function.
Our definition of a Fourier transform is similar to that by Born and Wolf2 , but is
different from that by the others. For example, Hopkins 22 uses a + i in the PSF integral,
and a
i in the OTF definition which he calls an inverse Fourier transform. Welford26
uses a
i in the PSF and a + i in the OTF, just as we have in Eqs. (1-51) and (1-71).
However, he refers to OTF as the inverse Fourier transform of PSF. Goodman2 and
Gaskill27 both use
i in both the PSF and the OTF. According to them, PSF is related to
the Fourier transform of the pupil function and OTF is the Fourier transform of PSF. The
OTF according to them, as the autocorrelation of the pupil function, is the complex
conjugate of the corresponding integral given in Eq. (1-73b).
r r
This ambiguity does not arise if f (r ) is radially symmetric, i.e., if f (r ) = f (r ) ,
r
where r = r , since in that case its Fourier transform is real and reduces to a zero-order
Hankel transform:
; 2
F(@) = f (r ) exp(2 ir@ cos ) rdrd
0 0
;
= 2 f (r ) J 0 (2 r@) rdr , (A3)
0
;
f (r ) = 2 F(@) J 0 (2 r@) @ d@ . (A4)
0
According to our sign convention, the wave aberration associated with an object ray
is numerically positive if it travels an extra optical path length to reach the reference
sphere compared to the corresponding chief ray. Born and Wolf2 and Goodman2 have the
Appendix B: Some Frequently Used Integrals 75
same sign convention, but Welford’s26 is opposite to ours. A change in the sign of the
wave aberration changes the sign of the imaginary part of the pupil function, and the
coherent as well as the incoherent transfer functions. It also rotates the incoherent PSF by
p as, for example, in the case of coma.
It is quite common in the optics literature to consider a point object lying along the y
axis. When using polar coordinates of a point in the plane of the exit pupil, the polar
angle q in that case is defined as the angle made by the position vector of the point with
the y axis, contrary to the standard convention as the angle with the x axis. We choose a
point object along the x axis so that, for example, coma aberration is expressed as
( ) ( )
x x 2 + y 2 and not as y x 2 + y 2 . A positive value of our coma aberration yields a PSF
that is symmetric about the x axis (or symmetric in y) with its peak and centroid shifted to
a positive value of x with respect to the Gaussian image point.
;
f ( x ) ( x ) dx = f (0) , (B2)
;
1
( ax ) = ( x ) . (B3)
a
;
f ( x ) A g * ( x ) = f ( x ) g * ( x'