You are on page 1of 8

USING COMPUTERS TO DESIGN NONIMAGING

ILLUMINATION SYSTEMS
David Jenkins and Mark Kaminski
Breault Research Organization, 6400 E. Grant Rd. Suite 350, Tucson, AZ 85715

ABSTRACT

With the development of faster computers, the ability to design and optimize complex optical systems has
been dramatically improved. This directly translates into faster product development cycles with less need
to build costly prototypes. Systems using light pipes, faceted Fresnel lenses, and nonimaging optics
demand non-sequential raytracing, generalized surface modeling, and scattering and/or ray-splitting off of
surfaces. Addressing these issues slows computation, resulting in time constraints that, in the past,
prevented the use ofsoftware codes to do much more than analyze complex systems. Now, a system's
radiometric performance can be evaluated in minutes instead ofhours, allowing more exotic computer
aided design and optimization techniques to be used. We present rules-of-thumb on how to design,
optimize, and tolerance illumination systems. Examples of systems include faceted light pipes and Tailored
Edge-Ray Concentrators that create uniform illuminance. Applications for such systems are broad and
include automotive, appliance, and room lighting.

KEYWORDS: Nonunaging Optics, Computer Design, Tolerance Analysis, Illumination Engineering

1. THE CHALLENGE OF ILLUMINATION DESIGN

Illumination system design involves complex and varied phenomena that occur in real life.
The physical laws of optics determine whether a design will work, but it is not always easy to calculate all
effects. While there are many well defmed algorithms in traditional lens (classical optical) design,
illumination systems follow relatively few rules. Illumination design is an art that blends real-world effects
to create a desired pattern.

The challenge comes from the variability of problems encountered when modeling a typical illumination
system. From automotive headlights to LCD backlights to overhead projectors, everything we see is lit up
differently. In general, light rays follow many different ray paths that do not follow a prescribed or
sequential order. At various surfaces, light can be transmitted, reflected, scattered, or diffracted.
Polarization and coherent effects may also be crucial in many systems, making the analysis of an
illumination system more complex. Generally, illumination system geometries are also more complex.
Figure 1 shows examples of common illumination surfaces, such as a nonimaging compound parabolic
concentrator (CPC), reflective Fresnel lens, and a planar array of pillow lenses.

Figure 1 - Examples of Illumination Surfaces. Illumination surfaces are generally more complex than
those found in classical lens design.

196 SPIE Vol. 3130. 0277.786X1971$1O.O0

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/04/2013 Terms of Use: http://spiedl.org/terms


Design Verification

A design concept is verified via analytic calculations, computer analysis and prototyping. Analytic
calculations check out the first order properties of a design ensuring that a system does not violate any
optical principles, such as brightness conservation. Computer modeling takes the optical presciption of the
source and physical geometry and calculates the system's performance. The model is checked by
measurements and analytic calculations. Many flawed designs are eliminated before time-consuimg and
costly prototyping.

Imaging versus Nonimaging Systems

The phenomena encountered in an illumination system is outside the realm of traditional lens design ray
trace code. Often more versatile and powerful ray tracing programs are necessary, since illumination
systems typically have both imaging (lens) and nonimaging parts.

Imaging systems deal with the first-order properties, aberrations, and radiometry of classical optical
systems. Classical systems are lens systems that map light on a point-to-point basis from source to target.
To first-order, each point on a source image corresponds to another point on the target image with the
deviations of these systems described by aberration theory. Design techniques for these systems are well
known and traditional lens design codes handle these problems efficiently. Imaging systems require a well-
defmed optical train with sequential propagation oflight through the system.

In the real world, light reflects, refracts, and scatters off of any interface based on where it exists
physically. With nonimaging optics, engineers also deal with first-order properties, but the emphasis is on
the radiometry of illumination systems. The goal is not to produce a sharp image, but to accurately transmit
light from an extended source to a target screen, meeting general design constraints. An optical systems
analysis code is needed to handle this type of problem. The Advanced Systems Analysis Package (ASAP)
from Breault Research Organization (Tucson, AZ), is one such package that accurately calculates light
propagation through general optical systems.

This paper describes a general design process for illumination systems. Specific examples include an LCD
backlight and a fluorescent illuminator system using a tailored edge-ray concentrator to provide uniform
illumination on a target plane.

2. FAST AND ACCURATE COMPUTER METHODS

Using optical software can greatly reduce the development time and cost of a project compared to a
development process that relies completely on prototyping. However, a typical illumination analysis can
require millions of rays. Even with the fastest computer one must use an efficient algorithm during the
design process. Further, each design has a large number of free parameters, making it impractical to check
each in depth. This is the same problem as trying to model a chess game by trying to determine every
possible combination of moves right from the start. The algorithm to a successful design does not always
follow a step-by-step procedure. Table 1 lists the keys to illumination system design to maximize the
chances for success and speed up the development cycle.

Build Fixed Geometry

Often an illumination system must be designed around existing mechanical or optical constraints, which
have been designed previously. Existing geometry is imported via a CAD translator utility or lens design
program translator or by entering the geometry using the program's interface. By saving the fixed
geometry separately, one can concentrate on the essential parts that may change. Saving this geometry as a
macro (a sequence of commands) that can be reused at any time speeds up the design process. Any initial

197

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/04/2013 Terms of Use: http://spiedl.org/terms


geometry that may impact performance in the presence ofmanufacturing imperfections should be
separated from the fixed geometry.

Table 1 - Keys to Illumination System Design


Build fixed geometry
Accurately model the source
Determine goals and merit function for design evaluation
Brainstorm and evaluate simple concepts
Optimize selected concept
Verify system performance at less than ideal conditions
Develop Tolerances
Prototype
Test and Measure

Accurately model the source

Creating a good source model is the key to design success. Ifthe source model is incorrect, the design will
be flawed. Ifthe source model is too intricate, then the ray trace will slow the design process.

Though point sources are often used in lens design, all sources are extended. Further, many problems arise
from figuring out whether the source is in the near-field or far-field. The general rule-of-thumb is to
calculate the ratio ofthe separation between the source and the nearest geometry versus the source size. If
this is greater than 10, then a far-field model is assumed.

The far-field model will always produce some level of inaccuracy since the source size is not considered.
Setting up a near-field source model will always produce the most accurate results, but this can be very
time consuming and slow. A near-field model must consider every little detail, such as the actual helical
filament structure within a bulb. An intermediate model is sometimes best as it is computationally efficient
and much more accurate than a far-field model. Figure 3 shows a far-field, intermediate, and near-field

Intermediate

Figure 3 - Pictures of a 9006 bulb models. A far-field point source model does account for a source's extent. An
intermediate type model often give results similar to the more accurate near-field model.

pictures of a typical 9006 automotive light bulb. For the near-field model, the bulb tube geometry and
helical emitting filament is shown in Figure 4. The bulb geometry was created using the Rhinoceros CAD
package and imported into ASAP via an IGES translator. A helical filament was created using ASAP. The
filament emits rays in a Lambertian pattern and its helical structure is self-obscuring. Rays are traced to the
outside of the bulb geometry and the flux is normalized to match measured values. A comparison with
actual measurements is extremely good and is shown in Figure 5. The model, however, is computationally
inefficient as only 40% of the rays created make it out of the system and each ray must be traced through
the bulb and filament geometry. The complexity of the helical filament slows the ray trace so that initial
ray set of 1,000,000 rays takes about 40 minutes to trace on a Pentium Pro 200 MHz PC. The far-field

198

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/04/2013 Terms of Use: http://spiedl.org/terms


model uses a point source with rays weighted to match the far-field intensity pattern and ignores the bulb
tube geometry. The intermediate model uses an emitting cylinder of roughly the same size as the filament
where each point radiates to match the far-field pattern. Not long ago, the far-field point source model
would be the only practical model. Today, however, both the cylindrical and point source setup ray trace
quickly. Given today's computer technology the intermediate model is often fast and sufficiently accurate.
The near-field model should be used when one needs to evaluate the effects from the Fresnel reflection of
the bulb tube.

Figure 4 - Near-field model of 9006 Bulb. Created using the Rhinoceros CAD package.

The accuracy of the source model depends on the actual radiance (both spatial and directional
characteristics at each point) ofthe source. As measurements of source radiance become available, one can
expect to see accurate and computationally efficient near-field source models.

Determine Goals and Merit Function

Merit functions are the best way to compare two designs. Sometimes it is fairly easy to decide which
design is best. An automotive headlight must meet certain candela requirements and it is easy to set up a
merit function that indicates how well they are satisfied. Less obvious cases arise from subjective
constraints. For instance, LCD backlights demand a balance between uniform radiance, low cost, and high
efficiency all inside the smallest possible package. In these cases, setting up a simple merit function and
evaluating how it works is required. Feedback about subjective issues should be used to refine the merit
function as time goes on. The importance of various constraints should be refmed also and fed back into
the merit function. Ideally, the engineer would like to simply define the best results, press a key, and get
back the best design.

Brainstorm and Evaluate Simple Concepts

If you do not have an exact path to achieving a goal, brainstorm and setup simple concepts where only a
few parameters can be changed. Most of the choices can be eliminated because they violate optical
principles, such as brightness conservation. Choosing between competing designs can usually be done after
a few simple ray traces. The evaluations use fast source models with fewer rays, so that many different
cases for each type can be looked at. Generally, concept ray traces should take less than one hour.
Otherwise, you will end up optimizing a design concept that may change later. Be careful to avoid
statistical uncertainty with a limited ray set. Averaging and simple data manipulation can be done at this
stage with less risk because any mistakes can be caught during the optimization phase.

199

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/04/2013 Terms of Use: http://spiedl.org/terms


Si mul aled Radi an Si miii aled Radi ani ml ensi ly of Syl sani a 9006 bulb

ANGLE from axis

165
150
135
120 125. 1401
105
iu. 6708
go 6: -_.:; :,:.: 111 1322
15
60
+5
30
_____ ___ 8+. 9165
16. 9119
6?. 0393
15 68. 1806

30 60 90 120 150 180 210 240 270 300 330


31 . 2850
ANGLE aboul axis
22. 3+64
13. 4019
3. +63

ASAP P1 us uS. 1 6.20. 199? 11:44

4sjrd Radiae( lI1tQeity 1 Sylvmeim 9008 Bil b'

AIIGLE(,om

1 8S

1 50
1 35
1 20 114.2178
I OS
110. 1391
go 101. 9023
75
80
77.5118
'S 89. 3S50
30 81. 1982
is 53. 0 1
-150 -120 -90 -80 -30 0 30 80 90 120 iSO
28. 5709
AIIGLE mbait zi 20. '1'0
12. 2S72
. 1004

ASAP P1 si VS. I 8-20-1997 10:35

Figure 5 - Comparison of measured bulb data with the near-field model of 9006 automotive bulb.

200

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/04/2013 Terms of Use: http://spiedl.org/terms


Optimization Selected Concept

After deciding on the best concept to pursue it is important to setup a complete model ofthe system with as
many free parameters as possible. A flexible macro (a sequence of commands) allows the user to rerun the
system for many different case with little or no effort. By setting up the design process this way, ASAP can
loop through various cases and report performance for each. At the beginning ofthe design all the
parameters may appear to be independent of each other. As the optimization proceeds one fmds dependent
parameters and the most sensitive parameters (those that most effect the design). Optimize the most
sensitive parameters first and reduce the number ofparameters by defming dependent parameters in terms
of independent ones. Start with fewer rays and gradually increase as the optimization becomes more
sensitive to statistical noise. Global optimization algorithms for illumination systems are usually
impractical because ray trace times are prohibitive. Ifthe design is not optimizing well, the engineer may
have to brainstorm again.

Verify System Performance at Less than Ideal Conditions

The optimized system is designed within a perfect operating environment. The next step is to add in effects
that degrade system performance. These effects can include displacing or tilting optical components,
adding roughness to surfaces to simulate slope errors, and adding scattering for unclean or poorly formed
surfaces. If performance is extremely sensitive to a particular perturbation, revisit the optimization phase
and redefme the merit function to reduce the sensitivity. The slight loss in time and performance is small
compared to the effort involved in a redesign because the prototype is a failure.

Develop Tolerances

Determining the system tolerances is fairly easy. From the "degraded system" runs described above select a
set of reasonable tolerances. Randomly varying the parameters within their tolerance ranges is the last step
to checks for any problems. The key is to avoid cliffs where performance drops substantially. Redefining
the base design values is sometimes necessary to reduce chances for a disaster.

Prototype
Prototype must accurately reflect the design. Translation via IGES files is relatively easy and lessens
miscommunication. Spare no effort to ensure good data transfer.

Test and Measure

Whenever possible measure before modeling. Fully characterize sources before starting the design process.
measured. The greater the accuracy of the model, the more confidence one can have that the system works.
After prototypes are built, measure surface quality and other optical properties again. This knowledge will
aid the next generation design. Measure the fmal performance.

3. ACTUAL SYSTEMS

How to apply the design principles in practice is shown in two examples. The two examples are a Tailored
Edge Ray Concentrator that creates uniform irradiance over a desired area and a LCD backlight. These
examples were chosen because they are simple and illustrate some important points.

201

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/04/2013 Terms of Use: http://spiedl.org/terms


Tailored Edge Ray Concentrator Example

A recent development in nonimagmg optics is the design oftailored edge ray concentrators for
illumination'. These illuminators have reflectors that are designed to take light from an extended source
and create a uniform irradiance on a target plane. The two dimensional reflector profile for a far-field
system, when the distance from target to source is very large, is defmed by a simple differential equation.
For this system, the design is controlled by one free parameter and the reflectivity of the reflector. The free
parameter defmes how much light is reflected to the center ofthe output distribution. Ifthe reflector does
not redirect any light to the center, there will be a lower peak flux at the center. Uniform irradiance is
produced over a wider area, since the average flux times the illuminated area is relatively constant.

Design goal: Build a reflector with polished aluminum around a fluorescent tube to illuminate a strip 200"
wide with uniform irradiance.
Constraints: The lamp is positioned 100" above the ground and the reflector should not be too big.

DESIGN PATH

Source Modeling: A long fluorescent tube with 1" diameter is created with Lambertian emission.
Analysis: The output distribution is evaluated until it covers the desired range.
Reflector Setup: A macro is written to integrate the differential equation for the reflector profile.
Optimization: Different shaped reflectors, with higher and lower peak flux levels, are tried until a uniform
irradiance over the desired area is achieved.

RkDIA3CZ 015!Xt3UTDW ar !ABZ!

FLW / 55-SrTIWiAL ruz Ur U IWII3!JR

ASAP Phis p Phi v5. 1 7-01-199?

Figure 6 - Performance of TERC illuminator and reflector profile shape.

Figure 6 shows the system performance and shape of the illuminator system. It has excellent uniformity
over the desired output range and the reflector profile is compact vertically and has a horizontal width of
20 inches. The reflector profile can be truncated and the design reoptimized if this width is too great. Ray
traces show that the fall off of the irradiance distribution depends on the overall size of the reflector. Ideal
nonimaging reflectors touch both the source and target, and are impractical.

LCD backlight

There is a huge market for LCD backlighting systems. They are used for displays of computers, video
games, pagers, answering machines, etc. The goal is to uniformly illuminate a large display area within an

202

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/04/2013 Terms of Use: http://spiedl.org/terms


extremely thin backlightmg device. Backlights are often light pipes and more information about designing
them can be found elsewhere2.

DESIGN PATH

Source Modeling: The source is modeled to represent the output of a fiber optic illumination system. Light
emission is constrained in angle to match the fiber numerical aperture (NA) and the spatial
distribution is uniform over one edge ofthe device.
Analysis: Uniformity is calculated for the output and the distribution is plotted.
Light Pipe Setup: The light pipe uses rows of facets at 45 degrees to extract light out the exit face.
The facet size and separation can be varied.
Optimization: The facet separations are varied along the length until a uniform illumination is
seen.

. :i:

:
;,

Figure 7 - Performance ofLCD Backlight System and Light Pipe Representation

Figure 7 shows the system performance and a representation ofthe facets inside the light pipe. The facets
are larger than actual scale. The backlight has good uniformity over the desired output range and the
system is very thin. The facet sizes should be made as small as possible based upon available
manufacturing techniques. The system should be redesigned for a particular facet size.

4. SUMMARY

The design of illumination systems on computer is a relatively new field. Even with the fastest computer,
the engineer must use an efficient algorithm to minimize raytrace time. In most cases, expensive
prototyping can not be eliminated, but computers give the engineer a better chance to get it right the first
time. Failures during the computer design process can be quickly corrected. Failures during the prototyping
are more time consuming and costly to correct.

The keys for a good illumination system design are accurate source modeling and avoidance of major
performance problems from loose tolerances. Redesign should be considered if a design performance is too
sensitive to a given parameter. Verification of computer models using measurements is of paramount
importance and should be done at as many steps as possible.

REFERENCES

1D. Jenkins and R. Winston, "New Reflector Types for Illumination," AppL Opt. 35, 1669-1672 (1996).

2J. Schweyen and D. Jenkins, "Smart Lightpipe Designs," Photonics Spectra, Jan., 1997.

203

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 01/04/2013 Terms of Use: http://spiedl.org/terms

You might also like