tutorial

ENGINEERING AN
OPTICAL SYSTEM
Choosing the right starting
point opens up a world of designs.

ILLUSTRATION BY GEORGE ABE

By Warren Smith, Kaiser Electro-Optics

esigning the basic format for an optical almost always done using the thin-lens concept.

D system is a process very different from

lens design. It involves determining the
components of the system and their locations in
A thin lens has a thickness of zero, an obvious
impossibility but a very valuable simplification
of the process because a thin component of a
order to produce a system that will meet a set of system can be represented simply by a power
required characteristics. It is the layout of the sys- and a location. Both principal planes in a thin
tem up to, but not including, the lens-design lens are coincident with the lens. The thin lens
stage (in which the lens designer determines the is a concept, not a reality. The lenses are later
detailed component configurations that are expanded to match real components with physi-
needed to do the job.) cal thickness.
The initial layout of an optical system is

july 2002 | s p i e ’s oe m a g a z i n e 49
step by step as a checklist:
The first step in the process should always be to establish the • The purpose of the system (obvious but often overlooked).
requirements and specifications. One should try to collect all • Wavelength, bandwidth, spectral response, or distribution.
of the specifications before beginning the design. The follow- • Aperture diameter.
ing, which will probably be included in a typical set, can serve • Focal length, or magnifying power if afocal. Magnification
and track length if the conjugates are both finite. For finite
conjugates the track length (object-to-image distance) and mag-
nification are sufficient; a focal length specification is
redundant and may constrain the final lens design.
• Numerical aperture (NA) or f-number (f/#). This may be
the infinity f/#, or NA, or the “working” f/#, but it should be
so defined.
• Object size and distance, image size, angular fields of view,
= image orientation.
• Performance: resolution, modulation transfer function
(MTF) at prescribed spatial frequencies, radial energy distribu-
tion, encircled or ensquared energy.
• Sensor characteristics: dimensions, spectral response, pixel
size and number, the aerial image modulation (AIM) curve of the
sensor, system type (visual, photographic, projection, laser, etc).
Figure 1 In this set of MTF curves, the best possible • Physical requirements: spatial limitations, size and location
image contrast for an ordinary optical system is shown in of entrance and exit pupils, cold stop, glare (Lyot) stop, bends
curve A. Curves B through F show the image contrast for or folds needed.
systems with wavefront deformations of λ/4, λ/2, 3λ/4, λ, • Ambient conditions.
and 2λ A system with a quarter wave deformation is often • Thermal stability requirements.
called diffraction limited. • Illumination and vignetting.
Once you’ve assembled this list, the second step, and an
important one, is to question or challenge the specifications.

(b) (a)
(a) (b)

D B
F D
B
F
(b)
(a)
(a) (b)

F
D B D F
B

(b) (a)

(a) (b)

D B (-)F D (-F)
B

Figure 2 Two-component lens systems include (a) telephoto, (b) retrofocus, and (c)
relay, in which a negative focal length provides an erect image. The mirror equivalents
are (a) Cassegrain, (b) Schwarszchild, and (c) Gregorian.

50 s p i e ’s oe m a g a z i n e | july 2002
Are they really necessary? Can the tough ones be eased? Has the about one minute of arc; performance falls off as scene bright-
bar been set higher than necessary, just to be safe? ness decreases. Thus the resolution limit imposed by the eye on
The third step is to ascertain that the specifications are self- the eye-telescope combination is one minute divided by the
consistent. In an afocal system, for example, the magnifying telescope magnification. Remember, although the final perfor-
power must equal the beam expansion factor, the ratio of mance of any system will be largely determined by the quality
apparent field to real field, and the ratio of entrance-pupil to of the lens design, at this point we are only concerned with the
exit-pupil diameter. In a telescope, the eye-lens diameter is limits imposed on the system by general layout.
determined by the eye relief The magnification of a
and the apparent field. In a D B fa fb sum of abs magnifier or a compound
finite conjugate system, the powers microscope is, by conven-
magnification equals the tion, given as 10 in. divided
ratio of image distance to 0.1 0.7 +0.333 -0.35 5.857 by its focal length f in inches.
object distance, and it also 0.2 0.6 +0.5 -0.6 3.667 This convention assumes a
equals the ratio of object side 0.3 0.5 +0.6 -0.75 3.0 comparison with the object
NA to image side NA (or 0.4 0.4 +0.667 -0.8 2.75 viewed from a distance of 10
image side f/# to object side 0.5 0.3 +0.714 -0.75 2.733 in. Under circumstances in
f/#). Check to be sure that 0.6 0.2 +0.75 -0.6 3.0 which one wishes to obtain
your system requirements 0.1 0.7 +0.778 -0.35 4.143 magnification as a compari-
don’t contradict one another. son with the view from some
Next, resolve any incon- other distance, for example
gruities. Compare the performance specs with known limits to D, the magnification factor is simply D/f. A positive magnifica-
determine whether they are reasonable. The Rayleigh limit for tion produces an erect image; a negative magnification, as in a
point resolution of a perfect lens is 0.61(λ)/NA; the Sparrow compound microscope, indicates an inverted image.
and line resolution limits are 0.5(λ)/NA. For an infinitely dis-
tant object, the Rayleigh angular resolution limit is 1.22(λ)/D getting specific
radians, where D is the entrance-pupil diameter, and (λ) is the An optical system usually falls into one of the following
wavelength. For a visual system, the Rayleigh resolution limit categories:
in seconds of arc equals 5.5 divided by the pupil diameter in • Single component
inches; for the Sparrow limit use 4.5/D. • Two-component (telephoto, retrofocus, relay, etc.)
The MTF cut-off frequency is 2NA/(λ), or 1/(λ)(f/#). In the • Afocal (e.g. telescope)
visible region, the cut-off frequency is about 1800/(f/#) lines • Afocal plus a prime lens
per millimeter. Plotting the MTF, or image contrast, against • Afocal plus a scanner
spatial frequency shows how varying the optical path difference • Periscope (relay, fiber optics, grin rod)
can affect a system’s imaging capabilities (figure 1). Under • Three (or more) component.
ideal, bright conditions, the resolution of the human eye is

cover to cover, reviewed my class notes, and proceeded

designer by default to learn by making mistakes and asking questions of my
betters,” he says.

I didn’t want to be an optical designer,” says Warren

Smith, chief scientist and consultant for Kaiser Electro-
Optics (Carlsbad, CA). “When I took the lens-design
Years later, Smith was invited to write a couple of chap-
ters for a handbook on military IR technology. “I hemmed
and hawed, but eventually I agreed,” says Smith. Those
course at the University of Rochester, I didn’t even buy chapters led to McGraw-Hill asking him to write an entire
the textbook.” Yet today, Smith is the author of Modern book on optical design. But Smith took some convincing.
Optical Engineering, a fundamental text in optical design. “Then they came back and mentioned money, and I
When he graduated in the early 1940s, he found him- couldn’t refuse,” he says with a laugh.
self as part of a top-secret project in Oak Ridge, TN. “At The rest, as they say, is history. “It turns out that the
the time, all I was told was that this was the most impor- approach I took at the time and the development of the
tant thing I could do for the war effort,” says Smith. It field were all very fortuitous,” says Smith. Originally pub-
wasn’t until a couple of weeks later that he realized he lished in 1966, Modern Optical Engineering is in its third
was part of the project to build an atomic bomb, develop- edition. Smith also wrote Modern Lens Design and a sys-
ing equipment to separate U235 from U238 by mass tems layout book. “All told, I guess the books add up to
spectrograph. about 50,000 copies sold,” says Smith. In addition, he has
Smith got into lens designing after World War II, when authored more than 34 papers, holds five patents, and
he went to work for an optical manufacturer in Chicago. “I serves as an expert witness in patent cases.
went out and bought a copy of Conrady, [the textbook he “But the book is my prized accomplishment,” he says.
was supposed to buy for his course], read it three times —Laurie Ann Toupin

july 2002 | s p i e ’s oe m a g a z i n e 51
Single component most widely used in all system layout work. Note that for mir-
The single-component system is simple because it is completely ror systems the mirror radius is simply twice the component
defined by its focal length, aperture, magnification, and field of focal length; a concave mirror has a positive focal length; a con-
view. With an object at infinity, the magnification is zero, but vex mirror, negative.
the image size is the focal length times the angle subtended by There are several widely known special configurations for the
the object. two-component system. If the focal length F is positive and
longer than the overall system length (D+B), the system is
Two component called a telephoto. This gives a long focal length (and the corre-
The two-component system is the most widely encountered, spondingly large image) in a small package (figure 2 on page
and a few very simple expressions serve to handle the layout 50). The telephoto ratio is (D+B)/F; if this ratio is less than
of this type. For the object at infinity, or at a large distance, one, the system is regarded as telephoto. The mirror equivalent
they are: of the telephoto is the Cassegrain configuration.
fa = DF/(F-B) If the focal length is positive and the back focus B is longer
fb = -DB/(F-B-D), than the focal length, the result is a reversed telephoto, or retro-
focus. This sort of system is used when a long working distance
where fa and fb are the focal lengths of the two components, F is B is needed to accommodate prisms or mirrors in this space.
the desired focal length of the combination, D is the spacing The mirror equivalent, called the Schwarzschild system, is
between the components, and B is the distance from the second rarely used except in small systems such as microscope objec-
component to the focal plane. These equations are probably the tives because the concave mirror diameter must be several times

non-imaging optics point objects at representative locations in the field of view for
different conjugates and different wavelengths.

T he term “non-imaging optics” refers to any optical sys-

tem that is not intended to produce high-fidelity image.
Non-imaging optics encompasses a vast array of illumina-
A non-imaging system has only one object: the light
source. The source may be artificial or natural, but the
designer is free to tailor the design for that one source or
tion applications, such as residential and commercial object. In contrast to an imaging system, the requirement
lighting; automotive lighting; computer, video, PDA, and for a non-imaging system is to create a particular light distri-
telephone displays’ solar-energy collection; indicator lights; bution, usually a smooth one with no rapid spatial or
instrument panels; integrating spheres and laboratory angular variations. Forming an image is often undesirable in
instruments; medical instruments; and so on. Although a non-imaging system since many artificial sources have
some of these applications do not require optical design by structures and irregularities that would spoil the smooth-
an engineer, many of them do. ness of the image. For example, in the design of a
The design process for non-imaging optics is analogous narrow-beam illumination system such as an automotive
to that for imaging optics. Compared to the mature state of headlight or a flashlight, the naïve guess is to use a point
software for imaging optics design, though, software-design source and parabolic reflector to create the illumination pat-
tools available for non-imaging optics are still in preadoles- tern. However, real sources are not points of light, and the
cence. In particular, automatic optimization of illumination approach described causes the source geometry to be
systems is still fairly crude, with most optimization being imaged or imprinted on the illumination pattern, spoiling the
done “by hand” using parametric studies of performance. uniformity. Instead, a designer working with a non-imaging
Both disciplines use geometric ray tracing to evaluate system will often break up the reflector into facets or other-
designs, but non-imaging designs may incorporate scatter- wise perturb the shape to blur or smear the image of the
ing elements, faceted reflectors, or arrays of micro-optical source and achieve the desired pattern.
components to achieve a particular illumination pattern. Another challenge in illumination systems design is the
In the early stages of a non-imaging design, the engineer choice and accurate modeling of sources. This choice may
invents a starting design from first principles or from a previ- be influenced by spatial or angular variation of the emitter, as
ous design and traces a few rays to learn how the design well as by cost, output, package size, ruggedness, or color.
performs. They may create the design in a ray-tracing pro- The shape of the emitter and its angular output often have a
gram or in a CAD program or by combining elements from strong effect on the performance of the design. Modeling
both. Accurately evaluating the performance of a non-imag- sources accurately is a challenging but important part of non-
ing design usually requires many rays, typically thousands imaging optical design because of the sensitivity of the
or millions, traced with a Monte Carlo ray-tracing program. system to inaccuracies in the model. An associated problem
To illustrate the difference between imaging and non-imag- is establishing tolerances due to variations in source manu-
ing design, consider the performance requirements of imaging facture or assembly of the source into the illumination
versus non-imaging systems. The requirement of an imaging system. This sensitivity may make the illumination system
system is to form an accurate image of whatever object or unfeasible as a production item.
scene is presented to it. The designer (and design software) —Edward Freniere, SPIE
achieves this by minimizing an error function for a collection of Lambda Research Corp. member

52 s p i e ’s oe m a g a z i n e | july 2002
as large as the aperture. in general a change in the sign of the magnification not only
The relay system is produced if one uses a negative focal indicates an erect or inverted image but also may produce two
length in the equations above. This produces an erect image, quite different systems, one of which may be vastly preferable
and the relay lens is often referred to as an erector lens. Note to the other.
that the focal length of the combination is equal to the focal
length of (a) the lens multiplied by the magnification of (b) the Periscope
lens. The Gregorian mirror system is the reflecting equivalent. A field lens is a lens (usually positive) placed close to an internal
image to make the off-axis image rays converge so that the
Afocal systems diameters needed for the subsequent optics are not unreason-
Component focal lengths for afocal systems are given by ably large. They are commonly encountered in conjunction
with eyepieces, so arranged as to direct the edge-of-the-field
fa = MD/(M-1) rays through the clear aperture of the eyelens. A converging
fb = D/(1-M), field lens will shorten the eye-relief (the distance from the eye-
lens to the exit pupil) of a telescope. A diverging field lens will
where M is the angular magnifying power and D is the system lengthen the eye-relief, but at the cost of requiring a larger eye
length. Note that a negative M indicates an inverted image. lens to pass the rays. Another classic application of the field lens
Examples of refracting afocal systems include an ordinary is in a periscope or endoscope. The periscope consists of alter-
Keplerian telescope, a Galilean telescope, and a lens-erecting nating relay and field lenses and allows a wide angular field
telescope. image to be carried through a long narrow space. An initial
There are equally simple expressions for a system where both image is formed by the objective, then passed along by the relay
conjugates are finite: lens after the field lens directs the rays to converge. Note that
the relay lenses pass the system image, whereas the field lenses
fa = msd/(ms-md-s') relay the images of the pupils of the objective and relay lenses.
fb = ds’/(d-ms+s'),
a question of distance
where s is the object distance, s' is the image distance, d is the An optical system can, to the first order, be described by com-
space between the components, and m is the magnification ponent focal lengths and spacings (i.e., by powers and spacings,
(equal to the image size divided by the object size). Note that where power is simply the reciprocal of the focal length). If

july 2002 | s p i e ’s oe m a g a z i n e 53
 function with targets for the desired system proper-
ties such as length, focal length, magnification, and
other items from the specification list as appropri-
ate. Next, set up the system as a series of
zero-thickness plano-convex or plano-concave
lenses, and designate the appropriate radii and
spacings as variables. Admittedly, a bit of knowl-
edge and foresight helps at this point (figure 3).
If there are one or more extra degrees of free-
dom, add a term to the merit function that forces
the sum of the absolute values of the surface cur-
vatures (the curvature is the reciprocal radius) to
zero. This technique minimizes the powers.
Another approach is minimizing the sum of the
squares of the curvatures.
The program will not only solve the problem
(assuming that there is a solution) but will find
an optimum solution. Note, however, that if
there is more than one solution, a typical pro-
gram will seek out the one nearest to the
starting system. Obviously it is beneficial to
understand the optical principles of the basic
system configuration when selecting the start-
ing system.
It is always wise to make a sketch of the sys-
tem, including the ray bundles for the on-axis
and off-axis imagery. This helps to avoid an
utterly ridiculous layout. It is possible to calcu-
late the thin-lens ray paths using the ray tracing
equations. The change in ray direction or
1 slope is given by u' = u - y(φ), and the ray
height at the next surface is given by y 2 =
u + du', where u' is the ray slope after passing
through the component, u is the ray slope before
Figure 3 This summary table shows some basic optical configura- passing through the component, y is the height
tions classified by field angles and apertures. at which the ray strikes the component, φ is the
component power (reciprocal focal length), y2 is
the ray height at the next component, and d is
the spacing to the next component. The equa-
tions are applied iteratively, component by
there are more degrees of freedom than those required to define component.
a configuration, the extra variables may be used to reduce or One can make a rough guess as to the type of design to use
minimize the component powers. As an example, let’s look at a for each component by determining the f/# and angular field
telephoto system with a unit focal length and a telephoto ratio for each component. With these factors, and experience, it is
of 0.8. This requires that (D+B)/F be equal to 0.8, but we are possible to estimate just how complex a construction will be
free to choose D and B, subject only to the restriction that required for each component. The basics of optical-system
(D+B) = 0.8F (see table on page 51). design can be seen to be relatively simple and straightforward.
It is apparent that the minimum total (absolute) power is The real trick is to avoid asking for the impossible. oe
between systems (D=0.4, B=0.4) and (D=0.5, B=0.3)—actu-
ally, it’s pretty close to D=0.46, B=0.34. As a rough rule of Warren Smith is chief scientist and consultant
SPIE
thumb, we can estimate that because this system has the least for Kaiser Electro-Optics in Carlsbad, CA.
member
power, it likely will have the least aberration residuals, the least Phone: 760-438-9255; fax: 760-438-6875;
sensitivity to mis-spacing and misalignment, and the lowest e-mail: [email protected].
fabrication cost. This is a very crude estimation because the
final result will depend largely on what the lens designer uses References
for the individual component configurations. But at the very 1. R. Kingslake, Optical System Design, 1983, Academic Press, New
least, this approach does give the lens designer an optimum York.
starting point. 2. W. J. Smith, Modern Optical Engineering; the Design of Optical
Actually all of this can be accomplished with commercially Systems, 3rd Ed. 2000, McGraw-Hill, New York.
available optical-design software. First, create a merit (or defect) 3. W. J. Smith, Modern Lens Design, 1992, McGraw-Hill, New York.

54 s p i e ’s oe m a g a z i n e | july 2002