Combined feedback method for designing a

free-form optical system with

complicated illumination patterns for an
extended LED source
Wenchang Situ,1 Yanjun Han,1 Hongtao Li,1 and Yi Luo1,2,*
1
State Key Lab on Integrated Optoelectronics/Tsinghua National Lab on Information Science and Technology,
Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
2
Semiconductor Lighting Laboratory, Graduate School at Shenzhen of Tsinghua University, Shenzhen 518055, China
*luoy@mail.tsinghua.edu.cn

Abstract: A combined feedback design method based on variable

separation mapping is proposed in this paper to design free-form optical
systems for an extended LED source with complicated illumination patterns.
In this method, macro energy division and micro illuminance distribution
feedback modifications are carried out according to the deviation between
the simulated illumination results and the target requirements. The free-form
optical system is then regenerated, and the deviation could be minimized
through multiple iterations. Results indicate that free-form optical system
designed by this method could achieve precise energy distribution, high
regional illuminance uniformity (89.7%), and high light output efficiency
(94.9%) simultaneously.
©2011 Optical Society of America
OCIS codes: (080.4225) Nonspherical lens design; (080.4298) Nonimaging optics; (220.2945)
Illumination design; (230.3670) Light-emitting diodes.

References and links

1. Y. Luo, Z. X. Feng, Y. J. Han, and H. T. Li, “Design of compact and smooth free-form optical system with
uniform illuminance for LED source,” Opt. Express 18(9), 9055–9063 (2010).
2. R. Winston, J. C. Miñano, and P. Benítez, eds., with contributions by N. Shatz and J. C. Bortz, eds., Nonimaging
Optics (Elsevier, 2005).
3. Y. Luo, X. Zhang, L. Wang, Y. Yang, F. Hu, K. Y. Qian, Y. J. Han, W. Lee, O. Zhang, and G. Deng, “Non-
imaging optics and its application in solid state lighting,” Chin. J. Lasers 35(7), 964–971 (2008).
4. Y. Yi, K. Y. Qian, and Y. Luo, “A novel LED uniform illuminance system based on nonimaging optics,” Opt.
Technol. 33(1), 110–112 (2007).
5. H. Ries and J. A. Muschaweck, “Tailored freeform optical surfaces,” J. Opt. Soc. Am. A 19(3), 590–595 (2002).
6. A. Timinger, J. Muschaweck, and H. Ries, “Designing tailored free-form surfaces for general illumination,” Proc.
SPIE 5186, 128–132 (2003).
7. V. Oliker, “Geometric and variational methods in optical design of reflecting surfaces with prescribed
illuminance properties,” Proc. SPIE 5942, 594207 (2005).
8. J. Bortz and N. Shatz, “Generalized functional method of nonimaging optical design,” Proc. SPIE 6338, 633805
(2006).
9. L. Wang, K. Y. Qian, and Y. Luo, “Discontinuous free-form lens design for prescribed irradiance,” Appl. Opt.
46(18), 3716–3723 (2007).
10. Y. J. Han, X. Zhang, Z. Feng, K. Qian, H. Li, Y. Luo, G. Huang, and B. Zhu, “Variable-separation three
dimensional freeform nonimaging optical system design based on target-to-source mapping and micro belt
surface construction,” Sciencepaper Online 1–9 (2010).
http://www.paper.edu.cn/en/paper.php?serial_number=201002-443

1. Introduction
Free-form optical systems designed by nonimaging optics [1–8] are usually adopted on LED
modules to improve the lighting performance by controlling the light distribution.

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1022
 The variable separation mapping design method has been widely used to design free-form
optical systems with point sources [9,10]. Here rectangular coordinates (x,y) and spherical
coordinates (u,v) are employed to denote target cells and source energy divisions,
respectively. Correspondence between (u,v) and (x,y) can be established according to the
variable separation energy mapping between source energy divisions and target cells [9].
Then, the free-form surface of the optical system can be constructed. However, for extended
source with dimensions comparable with the optical system, a real lighting result will deviate
from the target requirement [1].
Recently, a feedback modification method was introduced to minimize this deviation [1],
where the target plane is equidistantly separated into sufficiently small rectangular cells, and
the source energy divisions are adjusted to perform negative feedback modification. Repeated
iterations lead to negligible deviation. Although the optical system designed by this method is
able to distribute, in theory, the energy into different regions in the target plane precisely, it’s
difficult to reach high regional illuminance uniformity and light output efficiency due to
insufficient accuracy during construction, and the energy distribution would not be accurate as
expected either.
Another way to achieve higher construction accuracy is the feedback modification method,
which equally divides the source angle (u,v) into small angles and adjusts the area of target
cells to perform feedback modification. The optical system designed by this method could
achieve high regional illuminance uniformity and light output efficiency, but it could not
achieve precise energy distribution of different regions in the target plane.
The methods above could be used for simple illumination pattern, where the illuminance
distribution is a continuous function of (x,y), but not the complicated one, where the
illuminance distribution is a discontinuous function of (x,y). A combined feedback design
method is proposed in this paper for complicated patterns. It contains two feedback processes:
macro energy division and micro illuminance distribution feedback modifications. Results
show that the optical system designed by this method can achieve precise energy distribution,
high regional illuminance uniformity, and high light output efficiency simultaneously.
2. Feedback modification method
2.1 Feedback principle
To facilitate the optical system design, we first introduce two matrixes: energy matrix E,
denoting the source energy division weight with mesh grid (u,v), and area matrix A, denoting
target cells area weight with mesh grid (x,y). The design process is then simplified to the
adjustment of these two matrixes instead of complicated calculation of the correspondence
between (u,v) and (x,y). This makes the feedback process easier and clearer. Relationship of E
~(u,v) and A ~(x,y) are shown in Eqs. (1)–(4):
k n

u
uk 1
J (u ) cos udu 
i 1 j 1
E (i , j )
u
min
 m n , (1)
u  E (i, j )
max
J (u ) cos udu
min
i 1 j 1


vsk11
J (v ) cos vdv 
j 1
E (k , j )
v
vmin
 n , (2)
v
max

min
J (v ) cos vdv 
j 1
E (k , j )

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1023
 k n

x
xk 1
J ( x )dx 
i 1 j 1
A(i, j )
x
min
 m n , (3)
x
max

min
J ( x )dx 
i 1 j 1
A(i, j )

y
ysk11
J ( y )dy 
j 1
A(k , j )
y
min
 n , (4)
y  A(k , j)
max
J ( y )dy
min
j 1

where both E and A are m × n matrixes with (m + 1) × (n + 1) points on the surface, uk + 1 is

the u angle of the (k + 1)th curve Ck + 1, and vsk11 is the v angle of the (s + 1)th point on the (k
+ 1)th curve, so as xk + 1 and ysk11 . J(u), J(v) and J(x), J(y) are the Jacobian factors. Suffixes
min and max reveal the variation range of the corresponding parameter. The relationship is
shown in Fig. 1, where P is the constructed point on the free-form surface and Q is the
mapping point on the target plane.

Qsk ( xk , ysk , z0 ) Qsk1 ( xk , ysk1 , z0 )

A(k, s)
Target
k 1
Q s ( xk 1 , ysk 1 , z0 ) plane

Psk (uk , vsk ) P k (u , v k )

s 1 k s 1
E(k,s) z
Ck freeform
Psk 1(uk 1, vsk 1) surface
Ck+1

O y

Fig. 1. Energy matrix, area matrix, and the corresponding (u,v) and (x,y) divisions.

As illuminance = luminous flux / cell area, if the illuminance of a target cell is lower than
expected, the cell area can be reduced or its received luminous flux increased to increase the
illuminance of the cell, and vice versa. Thus, once the simulated illuminance distribution
matrix P is obtained, energy matrix E and area matrix A can be calculated separately with
inverse and direct proportion to perform negative feedback.
2.2 Feedback method with fixed area matrix
In this method, energy matrix E is adjusted when the area matrix A is maintained to perform
negative feedback modification. The process can be expressed as Eq. (5):
k 1 k 2 0
Ppre (i, j ) Ppre (i, j ) Ppre (i, j )
Ek (i, j )  k 1
 k 2
  0
 E0 (i, j ), i  1...m, j  1...n, (5)
Psim (i, j ) Psim (i, j ) Psim (i, j )

where Ek is the energy matrix of the kth feedback iteration and E0 is the energy matrix of the
k k
initial design. Ppre and Psim are the prescribed illuminance distribution matrix and the
simulated illuminance distribution matrix interpolated with the mesh grid of the area matrix of
the kth iteration, respectively.

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1024
 Owing to the Lambertian intensity distribution property of the LED source, there will be
large u and v angle divisions near the edge of the free-form surface. This may lead to low
construction accuracy, because the accuracy of the approximation between infinitesimal plane
and curved surface decreases when the plane enlarges. Furthermore, it would also cause
energy leakage and low uniformity of regional illuminance, as shown in Fig. 2. The surface
P1P2P3P4 constructed with smaller source angle α1, α2, α3 in Fig. 2(a) has a better accuracy of
construction than P1P2' with larger angle β in Fig. 2(b). As a result, ray r1 is controlled more
accurately than ray r2 .
Target plane Q Target plane Q Q'
P1 P1
P2 P2 r2
r1
1 P3 P3
2

3
O O P2'
P4 P4
(a) (b)

Fig. 2. Different accuracy of construction: (a) high construction accuracy with small source
angle division, (b) low construction accuracy with large source angle division.

2.3 Feedback method with fixed energy matrix

In this method, area matrix A is adjusted when the energy matrix E is maintained. The process
can be expressed as Eq. (6):
k 1 k 2
Psim (i, j ) Psim (i , j ) P 0 (i , j )
Ak (i, j )  k 1
 k 2   sim
0
 A0 (i, j ), i  1...m, j  1...n, (6)
Ppre (i, j ) Ppre (i, j ) Ppre (i, j )

where Ak is the area matrix of the kth feedback iteration and A0 is the area matrix of the initial
k k
design. Ppre and Psim are the prescribed and simulated illuminance distribution matrixes
interpolated with the mesh grid of the area matrix during the kth iteration, respectively.
Source angles u and v are equally divided into small angles, which could maximize the
accuracy of the surface construction according to the drawer principle. But since the area
matrix is not fixable, there will be a mesh grid mismatch between the area matrix and the
prescribed illuminance distribution matrix, or the boundaries of the regions divided by the
area matrix will not cover the boundaries of the regions divided by the prescribed illuminance
distribution matrix, as shown in Fig. 3. It would cause nonaccurate energy distribution.

Mesh grid of prescribed

illuminance matrix
Mesh grid of area matrix

Fig. 3. Mesh grid mismatch between the area matrix and the prescribed illuminance matrix.

This feedback method is especially suitable for designing free-form optical systems with
uniform illumination distribution pattern for an extended source.

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1025
2.4 Combined feedback method
Owing to the disadvantages mentioned above, either of these two methods could not deal with
complicated illumination patterns very well. Hence, a combined feedback method is proposed
in this paper for designing optical systems with complicated illumination patterns. Two
feedback processes are used: one is used to adjust the macro energy distribution through Ew,
and the other is used to adjust the micro regional illuminance distribution through As, so it can
ensure precise energy distribution of different regions, high illuminance uniformity, and high
light output efficiency at the same time. This method is performed as follows:
1. Divide the prescribed illumination pattern into specified regions with uniform
illuminance distribution or other simple illuminance distribution.
2. Obtain the prescribed macro energy matrix Ew0 according to the prescribed
illuminance distribution of each region.
3. Perform the feedback process to adjust the energy weight of the divided regions
according to

Ew0 (i, j ) Ew0 (i, j ) Ew0 (i, j )

Ewk (i, j )  k 1
 k 2
 0
 Ew0 (i, j ), i  1...mw , j  1...nw , (7)
Ewsim (i, j ) Ewsim (i, j ) Ewsim (i, j )

where Ewk is the macro energy matrix of the kth iteration feedback process, and there are
mw × nw divided regions on the target plane.
4. Calculate the macro (u,v) division corresponding to the divided regions using Eqs. (1)
and (2), and build the frame curves of the free-form surface as shown in Fig. 4(a).
5. Divide the source angle (u,v) equally in each region. Obtain the initial regional area
matrix As0 according to the prescribed regional illuminance distribution. Perform the
feedback process to adjust the illuminance distribution of each region according to
Psimks 1 (i, j ) Psimks 2 (i, j ) Psim0s (i, j ) 0
Ask (i, j )     As (i, j ), i  1...ms , j  1...ns , s  1...mw  nw , (8)
Ppreks 1 (i, j ) Ppreks 2 (i, j ) Ppre0s (i, j )

where the subscript s denotes the sth region on the target plane.
6. Build the regional curves as shown in Fig. 4(b). Generate the free-form optical system
model by integrating all the curves, then use the actual LED source dimensions to
obtain the simulated illuminance distribution matrix of kth iteration.
7. Repeat steps 3–6 until the performance of the free-form optical system satisfies the
requirement.

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1026
 Target Target
plane plane Frame
curve
Regional
curve

Freeform Freeform
surface surface

O Source O Source

(a) (b)
Fig. 4. Combined feedback process: (a) frame curves established, (b) regional curves
established.

3. Design example
The example is to perform a complicated illuminance distribution pattern in a 30m × 30m
quadrate region with the source mounted at a height of 7m. Design parameters and prescribed
illuminance distribution pattern are shown in Fig. 5, where region 1, region 3, and region 5 are
three rectangular regions with uniform illuminance distribution. The total energy ratio
between the white region and the blue region is 8:1. Dimension of the LED source immersed
in the optical system is 1mm × 1mm. The central height and the refractive index of the optical
system are 5mm and 1.59, respectively.

7m Total
Region1 8m
Energy
Region2 3m Weight

8
Region3 8m
1
Region4 3m
30m

30m Region5 8m

30m

(a) (b)
Fig. 5. Design example: (a) design parameters, (b) prescribed illuminance distribution pattern.

Simulation results of the initial and the final optical models designed by three feedback
methods are shown in Fig. 6 for comparison. Results of the initial designs are far from the
prescribed illuminance distribution pattern, while the results of the final designs are much
better after several iterations of feedback modification. Besides, the initial results are different
because energy matrix and area matrix obtained by different feedback design method would
lead to different construction results. Furthermore, a final result of the feedback method with
fixed A shows low regional illuminance uniformity and severe energy leakage. On the other
hand, final results of the other two methods show good control on energy leakage and regional
illuminance distribution.

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1027
 Y, m Illuminance, Lux Y, m Illuminance, Lux
20 20
10 0.1 10 0.1
0 0
-10 -10
-20 0.0 -20 0.0
-20 0 20 -20 0 20 -20 0 20 -20 0 20
X, m Position, m X, m Position, m
(a) (b)
Y, m Illuminance, Lux Y, m Illuminance, Lux
20 20
10 0.1 10 0.1
0 0
-10 -10
-20 0.0 -20 0.0
-20 0 20 -20 0 20 -20 0 20 -20 0 20
X, m Position, m X, m Position, m
(c) (d)
Y, m Illuminance, Lux Y, m Illuminance, Lux
20 20
10 0.1 10 0.1
0 0
-10 -10
-20 0.0 -20 0.0
-20 0 20 -20 0 20 -20 0 20 -20 0 20
X, m Position, m X, m Position, m
(e) (f)
Fig. 6. Simulated results: (a) initial result of feedback with fixed A, (b) final result of feedback
with fixed A, (c) initial result of feedback with fixed E, (d) final result of feedback with fixed
E, (e) initial result of combined feedback design, (f) final result of combined feedback design.

Figure 7 shows the comparisons between these simulated results on three parameters. As
shown in the figure, improvements in total energy ratio and illuminance uniformity are
obtained at the expense of decrease in light output efficiency. The low construction accuracy
and energy leakage of the feedback method with fixed area matrix A cause a deviation in the
total energy ratio, as well as low regional illuminance uniformity and low light output
efficiency. The feedback method with fixed energy matrix E has a good performance in both
regional illuminance uniformity and light output efficiency, but its total energy ratio doesn’t
satisfy the requirement. The final optical system model designed by the combined feedback
method has the best results on all three parameters shown in Fig. 7, which has a total energy
ratio of 7.92:1, an average regional illuminance uniformity of 89.7% and a light output
efficiency of 94.9% (Fresnel loss ignored).

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1028
 Total Energy Ratio
8

7
Initial
Final
6

5
A fixed E fixed Combination
(a)

Average Regional Illuminance Uniformity

90

80
Initial
Final
70

60
A fixed E fixed Combination
(b)

Light Output Efficiency

100

95

90 Initial
Final
85

80
A fixed E fixed Combination
(c)

Fig. 7. Comparisons between simulated results on three parameters: (a) total energy ratio
between white and blue regions, (b) average regional illuminance uniformity, (c) light output
efficiency (Fresnel loss ignored).

Profiles of the final optical system models designed by three feedback methods are shown
in Fig. 8. Although the dimensional differences between these models are small, the
illumination results are quite different due to comparable dimensions of the source and the
optical system. The 3-D geometry of the final optical system designed by combined feedback
method is shown in Fig. 9. The dimensions (length, width, and central height) of this model
are 13.71mm × 12.8mm × 5mm.

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1029
 6 6

4 4

A fixed A fixed
2 2
E fixed E fixed
Combination Combination
0 0
-8 -6 -4 -2 0 2 4 6 8 -8 -6 -4 -2 0 2 4 6 8
(a) (b)

Fig. 8. Profiles of the final optical system models designed by three feedback methods: (a)
cross-sectional profiles in the y-z plane (x = 0), (b) cross-sectional profiles in the x-z plane (y =
0).

Fig. 9. 3-D geometry of the final optical system model designed by the combined feedback
method.

4. Conclusions
A combined feedback modification method is proposed in this paper to design free-form
optical systems for an extended LED source with complicated illumination patterns. This
method contains two feedback modification processes: macro energy division and micro
illuminance distribution feedback modifications, which could keep high accuracy of the free-
form surface construction and have good control on the energy distribution.
The design example using the proposed method achieved precise energy distribution, high
regional illuminance uniformity (89.7%), and high light output efficiency (94.9%)
simultaneously.
Acknowledgments
This work was supported by the National Natural Science Foundation of P. R. China (under
Grant Nos. 60536020, 60723002, 50706022, and 60977022), the “973” Major State Basic
Research Project of China (Nos. 2006CB302800 and 2006CB921106), the “863” High
Technology Research and Development Program of China (Nos. 2007AA05Z429 and
2008AA03A194), Beijing Natural Science Foundation (No. 4091001), and the Industry,
Academia and Research Combining and Public Science and Technology Special Program of
Shenzhen (No. 08CXY-14).

#143541 - $15.00 USD Received 4 Mar 2011; revised 13 May 2011; accepted 6 Jul 2011; published 15 Jul 2011
(C) 2011 OSA 12 September 2011 / Vol. 19, No. S5 / OPTICS EXPRESS A1030