Sie sind auf Seite 1von 12

Achieving color point stability in RGB multi-chip LED modules

using various color control loops



P. Deurenberg
*
, C. Hoelen, J. van Meurs, J. Ansems
Central Development Lighting, Philips Lighting, Eindhoven, the Netherlands


ABSTRACT
The continuing research effort in high power LEDs will allow their use in high quality lighting systems in the (near)
future. There are still a number of issues to tackle, for instance the LEDs (strong) temperature dependence. This
dependence will change the emitted flux and the spectral distribution of the LED. In addition, these parameters will also
change as the LED ages. When creating white light by mixing red, green and blue LEDs, the temperature effects
described above will already result in a visible color difference after a small rise in temperature.
To overcome this issue, a number of LED color control loops have been developed. These loops can be based on: the
heat sink temperature, flux measurements of each primary color, a combination of these last two and an integrated color
point. For this purpose, an RGB test set up has been built, equipped with a temperature sensor and various photo-sensors.
The appropriate color control loops have been implemented and tested in software. Some control loops use empirically
determined LED parameters (d/dT or T
0
), the value of these parameters has been determined for a different set of LEDs.
In addition, initial optical LED (and sensor) calibration has been performed at a single temperature only.
The color stability of the various color control loops has been measured for a temperature increase of about 50
degrees Centigrade. In this range, we find that, on short term, all color control loops show a significant improvement in
the color error, except for the color control loop based on flux measurements of each primary color, which performs
nearly as mediocre as open loop. However, the color control loop based on the heat sink temperature cannot offer color
stability when the LED ages, which is expected to be significant. The color control loop based on an integrated color
point seems the most expensive one.

Keywords: LED, RGB, color control, color feedback, color feed forward
1. INTRODUCTION
LEDs for lighting have great future potential. Their luminous efficacy has already reached the luminous efficacy of the
compact-fluorescent lamps. Apart from the advantages with respect to efficiency and lifetime, using LEDs in general
lighting also provides the opportunity to instantaneously change color and dimming level.
Two approaches can currently be discerned to create white LED light: 1) by using phosphor based white LEDs, and
2) by using multi-color LEDs (e.g. by additive mixing of red, green and blue)
2
. An advantage of the second solution is a
theoretically higher efficiency, however, this option still has many challenges, especially with respect to color stability:
LEDs have the unfortunate property that their output varies with their temperature. Not only does the flux output change,
but also their peak wavelength of emission increases for rising temperatures. In addition, for option 2, the non-constant
maintenance causes color point shifts during ageing. An RGB LED system without compensation for any of these effects
would already produce clearly visible color differences after a temperature rise of about 20 C. The CIE1976 uniform
color space (UCS)
5
is usually used to analyze just-noticeable-color differences. In this color space, color differences,
between two color points (u
1
, v
1
) and (u
2
, v
2
), can be calculated by equation (1).
( ) ( )
2
'
2
'
1
2
'
2
'
1
' ' v v u u v u + = (1)
The maximum allowed color deviation (or target color deviation) depends strongly on the application. In this research,
the target is set at uv=0.010, because lamps are usually specified within 10 points in the CIE 1931 color space
5
and
because this target also coincides with display specifications. However, a color difference of uv=0.0035 is assumed as
a just noticeable difference
5
.

*
Peter.Deurenberg@philips.com; phone +31 40 27 57990; fax +31 40 27 56693
Fifth International Conference on Solid State Lighting, edited by Ian T. Ferguson,
John C. Carrano, Tsunemasa Taguchi, Ian E. Ashdown, Proc. of SPIE Vol. 5941
(SPIE, Bellingham, WA, 2005) 0277-786X/05/$15 doi: 10.1117/12.623020
Proc. of SPIE Vol. 5941 59410C-1


Unlike earlier publications
2, 3, 4
discussing a single color control loop, this paper gives an overview of four possible
LED color control loops to counter the effects of temperature (and ageing). These loops are described and tested on an
experimental set up (with an LED source consisting of two red, six green and two blue Lumileds Luxeon emitters
7
);
measurement results are presented and discussed. This overview can be used as a basis for selecting a suitable color
control loop. The actual choice depends on the application and its requirements.
2. MEASUREMENT OPTIONS
The color control loops introduced here, strongly relate to the optical variations that can be expected for LEDs when
their temperature changes. As can be found in literature
6, 7
, a rising LED junction temperature causes the flux output to
decline (e.g.
-1
d/dT = -0.2 K
-1
), the peak wavelength to increase (e.g. d/dT = 0.10 nm/K for a red LED) and the
forward voltage to drop (dV
F
/dT -2 mV/K). If also the LEDs forward current is varied, the flux output (d/dI > 0) and
the peak wavelength (d/dI < 0 nm/K) change. To simplify LED color control loops, the LEDs are operated using a
PWM (pulse-width-modulated) forward current, driving them at a constant current and changing their on-time (duty
cycle) to vary the average flux.
Keeping these temperature dependent optical variations of LEDs in mind (e.g. d/dT or
-1
d/dT), the color control
loops below can maintain a constant color point up to a certain degree. For each color control loop, a color error
prediction is formulated, based on an RGB LED system at a junction temperature of 60 C and 3500 Kelvin on the black
body line, gathered from literature
2
. As will be shown in subsequent sections, not all measurement possibilities will offer
an equally stable color point.
2.1 Temperature measurement (TFF)
Most of the LED output variations are caused by a change in junction temperature. Therefore, temperature measurements
form a reasonable basis for a compensation scheme. Unfortunately, it is not practical to directly measure the junction
temperature and for this reason, an indirect measurement of the junction temperature is made through the heat sink
temperature. As it is not the junction temperature that is controlled, this color control loop is called temperature feed
forward (abbreviated as TFF).
The flux output and peak wavelength of an LED both change as a function of temperature. Starting from a correct
color point at an initial temperature and a suitable model with accurate prediction of the LED output characteristics, this
color point can be maintained. Unfortunately, these dependencies are not precisely known and may vary over batches
and manufacturers. This can result in significant color errors. In addition, this scheme cannot correct for maintenance
issues. Given the variability in LED ageing, a simple counter cannot yet adequately address this issue. Simulations
indicate that long-term color error uv can be much larger than 0.005
2
.
2.2 Flux measurement (FFB)
A single photodiode can be used to obtain the LED flux of each color component. The controller can subsequently
maintain the preset flux to preserve the color point. This measurement scheme will be able to correct for flux variations
due to temperature and ageing. Unfortunately, it cannot correct for peak wavelength shifts caused by temperature
changes. This can already result in a color point deviation uv > 0.005 for temperature changes of T=20 C
2
. This
color control loop is called flux feedback (abbreviated as FFB).
2.3 Flux and temperature measurement (FFB&TFF)
Combining the last two measurements, thus flux feedback combined with temperature feed forward (abbreviated as
FFB&TFF), can make a significant improvement, as the combination is able to determine all variations in LED output:
flux changes due to temperature and ageing through the photodiodes, and wavelength changes through the temperature
sensor.
However, it still relies on data describing the relation between wavelength shift and temperature (d/dT). Therefore,
it also suffers from uncertainties in this relation. The degree of uncertainty will determine the accuracy improvement
over flux feedback only. However, the color error uv will be smaller than 0.01 for temperature changes up to 50 C
2
.
2.4 Color coordinates measurements (CCFB)
An LED color control loop can also be achieved by directly feeding back the color coordinates of the mixed light using a
color sensor. Hence, this color control loop is called color coordinates feedback (abbreviated as CCFB). In this case, the
spectral response of the sensors must match the CIE 1931 2 color matching functions
5
. The feedback signal would result
in X, Y and Z color values. In principle, the sensors could be photodiodes covered by an appropriate optical filter.
Proc. of SPIE Vol. 5941 59410C-2


As the system would directly control the white light, a high degree of color accuracy is possible. All above-
mentioned variations in LED output are measured and can thus be compensated for. Errors are mainly introduced by
mismatches between the sensor sensitivity curves and the color matching functions. Simulations indicate that with
adequate sensors it should be possible to achieve a color accuracy uv of about 0.006
2
. Using color sensors that are
identical to the color matching functions, this can drop to a color difference uv of about 0.002
2
.
3. CALIBRATION ISSUES
One of the key issues for all color control loops is the calibration of the unit. Calibration is needed for accurate color
setting because the LED binning and sensor characteristics do not provide enough detail. However, it also should not be
required to determine all LED parameters for a range of temperatures. The calibration can be split into two parts: 1) the
optical characteristics of the LEDs (C-matrix), and 2) the optical characteristics of the (color) sensors used in CCFB (S-
matrix). With these matrices, target color points can be translated to LED output power and sensor set points.
A target color point and light level with tristimulus values
5
X
T
, Y
T
and Z
T
can be expressed as:
( ) [ ]

B
G
R
T
T
T
D
D
D
T C
Z
Y
X
(2)
where matrix C describes the CIE set points of the LEDs as a function of the duty cycle D
i
for LED color i. The C-matrix
contains the CIE 1931 tristimulus values for each LED color (X
i
, Y
i
and Z
i
) on a column basis:
( ) [ ]

=
B G R
B G R
B G R
Z Z Z
Y Y Y
X X X
T C

(3)
Unfortunately, this matrix is temperature dependent as the flux output and peak wavelength change as a function of
temperature. The inverse of the C-matrix, also called Calibration matrix, can be used to determine the required duty
cycles for a certain target color point. Similarly, the (target) sensor values SA
T
, SB
T
and SC
T
can be expressed as:
( ) [ ]

B
G
R
T
T
T
D
D
D
T S
SC
SB
SA
(4)
where the S-matrix contains the sensor output values (SA
i
,

SB
i
and SC
i
) for each LED color (again on a column basis):
( ) [ ]

=
3 3 3
2 2 2
1 1 1
SC SB SA
SC SB SA
SC SB SA
T S

(5)
This matrix is also temperature dependent, although for small sensor temperature changes this can neglected. The S-
matrix should be mostly diagonal, determined by the degree of coupling between sensors and LEDs. When using a single
photodiode to determine the flux output of each LED color, the sensor is time-multiplexed and therefore the S-matrix
already is diagonal by design.
As the CIE 1931 x, y chromaticity coordinates are the ratio of the tristimulus values, ( ) Z Y X X x + + = etc.
5
, a
similar (calibration) matrix can be used to calculate the (sensor) set points from these color coordinates, however, the
outputs need to be normalized with respect to the maximum (sensor) set point. An additional parameter L
T
(0 100 %)
can be used to indicate the relative amount of light with respect to maximum possible light output at the chosen color
point. Flux output can either be set through a scaling factor for the (sensor) set points, or it can be implemented through
scaling the maximum flux output (being the sum of Y
R
, Y
G
and Y
B
). The flow from user set points (x
T
, y
T
, L
T
) to duty
cycles (D
R
, D
G
, D
B
) and, depending on the used color control loop, (color) sensor set points would then be
Proc. of SPIE Vol. 5941 59410C-3
U S E R D O M A I N
C a l i b r a t i o n m a t r i x
U s e r I I
I n t e r f a c e L - . 1 e 4 C I E x y , L - - >
9 O l E x y . L
L E D d u t y c y c l e s a t
_ _ _ _ _ _ L
: T _ T r 1 e r s n u u
A c T U A T O R D O M A I N
L i g h t i n g s y s t e m . . I
( m d . d r i v e r )
'
d u t y c y c l e - - >
l i g h t o u t p u t
e x p ( D C n o m , T , l f e % ) D C n o m

e f
1
T
r e f
l f e %
m o d e l a c t u a l s y s t e m


( ) [ ]
( ) [ ]

+ +

SC
SB
SA
D
D
D
Z
Y
X
Y Y Y
y
x
T S
B
G
R
T C
T
T
T
B G R
T
T
T
1
L apply
normalize
(6)
Note that there exist other, equally suitable, matrices to determine the same information.
4. COLOR CONTROL LOOPS
This section will discuss the different color control loops. Each loop tries to maintain the color point based on a
calibration matrix valid at a single junction temperature for each LED color. Depending on the characteristics of the
color control loop, it uses certain sensor values and/or (empirical) LED parameters. These empirical LED parameters
must be determined for a different set of LEDs! These systems do not regulate the system temperature!

Figure 1: Block diagram for an open loop system
4.1 Open loop (OL)
The basic building blocks, present in every color control
loop, are introduced by means of a system without
compensation. The block diagram can be found in figure 1.
Note that, without compensation, a rise in system
temperature will lead to a deviation in color point.
The target color point is set by the user in the user
domain (User interface-block) and converted to actuator
domain by the Calibration matrix-block using an inverted
C-matrix, equation (3). The Lighting system converts the
duty cycles to light. In all of the loops below, the LEDs,
PWM amplifier, optics and sensors (incl. possible AD
conversion) are unified in the Lighting system-block.
4.2 Temperature feed forward (TFF)
As mentioned earlier, the junction temperature of each LED cannot be measured directly. Therefore, these temperatures
are determined through the system (or heat sink) temperature and a thermal model of the system. The compensation can
then be implemented through an inverted LED model (model prediction control). As a model for the flux output, the
influence of temperature on an LED based unit can be split into two parts. In the first part, the light output () of each
LED color decreases as a function of temperature (if the reference temperature (T
ref
) is below present junction
temperature T
j
), for example:
( ) % exp % % , %,
0
life
T
T T
DC life T DC
ref j
ref j


= (7)
where DC% represents the duty cycle (0%-100%), T
0
a characteristic LED parameter and life% describes the LED
maintenance. In the second part, a different temperature changes the LED spectrum (e.g. peak wavelengths shifts), which
results in a different color impression. This last influence will be compensated for by a change in set point through the
Calibration matrix-block (middle of figure 2 below). The change in flux output is dealt with through the outside loop,
in which the nominal duty cycles from the Calibration matrix-block are increased with the same factor as the flux
output decreases or vice versa (see equation (7)):
(8)
If the Lighting system is well modeled by these functions, the actual system and the model in the above equation
exactly cancel and the flux output will remain constant for every temperature.
Proc. of SPIE Vol. 5941 59410C-4
u s e r



Figure 2: Block diagram for temperature feed forward system (the main feed forward loop is indicated by the thick lines)
Describing figure 2, the user domain set points are converted to actuator domain duty cycles for each LED color
(Calibration matrix-block). This conversion now depends on the junction temperature of each LED as the peak
wavelength may have shifted (indicated by the arrows form the Calculate junction temperature for each LED color-
block to the Calibration matrix-block). Depending on the junction temperature of each LED, the decreased flux output
is compensated for by an appropriate multiplication factor supplied by the
ref
/ (T
j,i
)-blocks. This results in adapted
duty cycles, which are filtered through the Rescale duty cycle-block (described in section 4.6). After a small delay
(indicated by the z
-v
-blocks), these values are fed into the Lighting system-block, which generates the light and the
heat sink temperature based on the applied duty cycles. Using a thermal model of the system, the LED junction
temperatures can be derived from the dissipated power and the heat sink temperature. These are passed to the

ref
/ (T
j,i
)-blocks, which completes the loop.
Essentially, a quasi-static situation is assumed every time the duty cycle is changed. This assumption is valid when
the sample period of the feedback is much shorter than the thermal time constant of the lighting system. Comparing this
situation to a classic approach, the controller actually has proportional feedback with a variable gain.
4.3 Flux feedback (FFB)
A system using optical feedback with a single optical sensor multiplexed over multiple LED colors is depicted in figure 4
below. With this approach, it is possible to detect and compensate for flux changes caused by ageing and junction
temperature changes. Unfortunately, wavelength shifts due to temperature changes, cannot be detected and will generate
a color error.

Figure 3: Implemented measurement approach featuring a
single measurement for each LED color and an additional
measurement to determine the additive background level
As said, a single sensor is used to determine the light
output of each LED color; this is achieved by time-
multiplexing the sensor over all LED colors. This means
that the instantaneous radiative power is determined of all
switched-on LED colors during each measurement. There
is one measurement for each LED color and an additional
one to determine the additive background light component
falling onto the sensor. This multiplexing requires a large
sensor signal bandwidth, because the measurements are
performed in rapid succession, see figure 3. The
multiplexing and the inherent decoupling are indicated by
the Time multiplexer- and Color signal extractor-
blocks, both in the top right corner of figure 4.
Proc. of SPIE Vol. 5941 59410C-5
w S ) L E D
b l L E O
C O L O R E D P A R T S A R E
A C T U A T O R D O M A I N


The main control loop (indicated by the thick lines), starts with a sensor set point for each LED color at a certain
reference temperature (T
ref
). The difference between set point target (S-matrix, equation (5)) and actual state is calculated
and passed to a Proportional-Integral-Differential-controller (PID), which determines the change in duty cycles. As the
measurements determine the light output amplitude, as a function of the applied forward LED current and junction
temperature, the measured signals no longer depend on the duty cycle. Therefore, in order to facilitate the feedback loop
with a duty cycle dependent sensor signal, the sensor values are multiplied with the previous iteration of the PID
controllers output (indicated by the z
-1
-blocks). As such, the PID controller is designed to determine the relative
amount of power, which needs to be applied to maintain the flux amplitude at a desired level.
In principle, these values should now be transformed to the actuator domain; however, this is not necessary because
the sensor measurements are already completely decoupled by the chosen measurement method (also indicated by the
unity gain matrix I-block. To implement the color point chosen by the user, the PID outputs are multiplied with the
nominal duty cycles for the chosen color point at the reference temperature provided by the Calibration Matrix-block
(C-matrix, equation (3)), thus providing corrected signals in the actuator domain. After passing these duty cycles through
the Rescale duty cycle-block (described in section 4.6), they are delayed and passed to the Lighting system-block,
which generates the light and the sensor values.

Figure 4: Block diagram for flux feedback system (the main feedback loop is indicated by the thick lines)
4.4 Flux feedback and temperature feed forward (FFB&TFF)
A system using optical feedback combined with a temperature feed forward utilizing a single optical sensor multiplexed
over multiple LED colors is depicted in figure 5 below. With this approach, it is possible to detect and compensate for
flux decreases and maintenance issues through the optical sensor. Color point changes due to wavelength shifts (d/dT)
can be compensated for via feed forward through the temperature sensor. Assuming an adequate model of the
wavelength shifts is available, (very) accurate color stability should be possible.
The optical feedback loop is already described in the previous section (flux feedback) and will not be discussed again
here. In this case, the Lighting system-block also provides a heat sink temperature, from which the three junction
temperatures can be derived (bottom right block). These temperatures are passed to the Calibration matrix-block to
account for the peak wavelength shifts. Additionally, the flux references for the flux feedback loop need to be altered, as
the flux sensitivity of the photodiode is wavelength dependent. Note that if the temperature of the photodiode changes as
well, this sensitivity change needs to be accounted for as well (which may require an additional temperature
measurement in the Lighting system-block).
Proc. of SPIE Vol. 5941 59410C-6
0
b I ! L E D
E l i l l i i i
A C T U A T O R D O M O I D -
A C T U A T D O M A I N



Figure 5: Block diagram for flux feedback and temperature feed forward (the main feedback loop is indicated by the thick lines)
4.5 Color coordinates feedback (CCFB)
The most general, but probably most expensive, approach to solve color accuracy issues is color coordinates feedback,
which controls the integrated color point of the mixed light. This approach measures the color properties of the mixed
light of all LED colors through separate, optically filtered, photodiodes. The relative output of the separate photodiodes
provides information regarding the change in color coordinates and light output of the mixed light. Changes due to
temperature rise, maintenance issues and even an LED failure can be detected and compensated for.

Figure 6: Block diagram for color coordinates feedback (the main feedback loop is indicated by the thick lines)
By integrating the sensor signal over multiple PWM periods, high frequency noise and turn-on/turn-off driver
behavior can be neglected, as all these signal disturbances are accounted for in the (integrated) measurement. However,
the integration or filtering, introduces low frequency sensor dynamics. This makes it necessary to control in sensor
domain and reduces the MIMO (multi input, multi output) system to multiple SISO (single input, single output) systems.
Each SISO system can simply be controlled by a PID-controller, if the coupling between the different SISO systems is
not too big (indicated by a diagonally dominant S-matrix (5). Controlling in the sensor domain also implies converting
Proc. of SPIE Vol. 5941 59410C-7


the user inputs (e.g. in CIE 1931 2 x, y chromaticity coordinates
5
and relative luminous output) to corresponding sensor
set points (arrows from the Calibration matrix to the Color sensor references-blocks).
The block diagram can be found in figure 6. It shows that the user domain user inputs are converted to the actuator
domain by the Calibration matrix-block. The set points for the feedback loop are converted from the duty cycles to the
sensor domain through equation (4). The sensor measurements are compared to the set point and the difference is passed
to a PID controller. The resulting PID action (still in sensor domain) must then be converted to actuator domain (duty
cycles for the driver) through the decoupling matrix (the inverted sensor matrix, S
-1
described in equation (5)),
implemented by the Sensorduty cycle- block. After multiplication with the nominal duty cycles (for the chosen color
point at the reference temperature provided by the Calibration Matrix-block), the duty-cycles can be filtered through
the Rescale duty cycles-block (described in section 4.6). After a short delay, the duty cycles are converted to light and
corresponding sensor values by the Lighting system-block.
4.6 Some general properties
The above-presented methods have at least three things in common. First, all color control loops explicitly result in
positive feedback with respect to the system temperature (this is required by the LED characteristics). This implies that
limits must be set to the power dissipation and measures should be taken to cope with unusual circumstances (e.g. a
considerably elevated ambient temperature). Second, the chosen block layout and by implementing the PID output as a
multiplication factor, all color control loops show (nearly) open loop behavior with respect to chosen color point and flux
level when looking from the user point of view through the system. Consequently, the color point and flux level of the
system can be changed very dynamically.
Third, in order to ensure a constant color point in different circumstances, information exchange between the
independent color loops is required as each loop individually regulates its state to set point. If one of the color loops can
no longer reach its set point, the desired color point cannot be maintained. A situation like this can readily occur and
some measures need to be taken. In such circumstances, the set points and present duty cycles are scaled down in
conjunction with the eyes sensitivity to color differences versus flux differences. The required algorithm thus prioritizes
the correct color point in case of insufficient flux. This is implemented and indicated by the Rescale duty cycle-blocks,
the Scale factor-arrow provides the Color sensor references-block with the required reduction factor.
Also, a similar approach must be applied when transforming the user input to the sensor domain. The user can request
a flux output at a certain color point, which is simply not possible for the light fixture, as its maximum flux output
strongly depends on the set color point. For instance, compare the flux outputs at fully saturated blue and white light.
Therefore, the Calibration matrix-block also prioritizes the color point above the flux output when transforming the
user set points to the sensor domain.
5. EXPERIMENTAL SET UP
In order to verify the presented color error expectations for each color control loop, an experimental set up has been
built. This section describes some of the aspects of the experimental set up. The total set up is shown in figure 7:

Figure 7: The experimental set up used to determine the
performance of each color control loop
For accurate measurements, the LED light needs to be
mixed; this can be achieved using an integrating sphere. For
calibration and reference measurements a PHOTO
RESEARCH PR-650 SpectraScan SpectraColorimeter
(mounted on tripod) has been used. This portable colorimeter
offers an accuracy of 2nm and 0.006 x, y color accuracy for
typical CRT phosphors.
The various color control loops have been implemented on
a Texas Instruments TMS320LF2407 16-bit integer DSP with
10 bit AD converters. This processor is used to generate PWM
signal at a 14.3 bits resolution and has (more than) sufficient
calculating power to fulfill all possible requirements.
Lumileds Luxeon emitters
7
are used and their specifications at
350 mA forward current and 41 C heat sink temperature can
be found in the table 1. LED parameters T
0
and d
peak
/dT have
been determined empirically for a different set of LEDs.

Proc. of SPIE Vol. 5941 59410C-8
1 < 2
T I F T 6 I D I
R I
P W M 1 n
M 2 0 . O p n 1 2 . 5 M R R S O . 0 0 0 5 0
T s R R u n R o n r u n e n o n n o u n O O j n i S R 0 0 0 0 0 0
P u s i n i n
0 0 . 0 0 0
R i o n 0 0 0 0 0 5 0 n o
I I I I I I I I I I I I ! I ! !
7
l R 5 0 ! I ! ! ! I ! ! ! ! ! ! ! I ! ! .
T s R R u n A s e r u n e 5 6 0 R U n s
O R 1 1 : 4 1 : 3 6 ! ! ! s l T n
P u s i n i n
2 4 . 4 1 1
I I I I I I I I I I I I I I I I I I I u s g s
M R R . R n g l R . 5 M R R R R . R T S T I


LED color # x y

peak
[nm]

dom
[nm]
d
peak
/dT *
[nm/K]
FWHM
[nm]

[lm]
T
0
*
[K]
P
rad

[W]
P
el

[W]
R
th

[K/W]
Red 2 0.7005 0.2981 640 626 0.15 20 31 90 0.25 2.17 18
Green 6 0.2027 0.7225 526 533 0.05 35 37 300 0.26 7.60 15
Blue 2 0.1458 0.0449 457 463 0.02 25 24 1300 0.17 2.25 15
Table 1: Overview of LEDs properties (at 350 mA forward current) used in experimental setup at 35 C heat sink temperature
(* determined empirically using a different set of LEDs)
5.1 Drivers
LEDs are essentially current driven light sources. Unfortunately, the LEDs peak wavelength tends to shift as a function
of this forward current. If not taken into account, this can provide substantial color errors. To deal with this problem,
LEDs are driven with a PWM current signal. The LED driver can therefore be seen as a PWM amplifier.

Figure 8: PWM amplifier for the LED color control loops
In particular for the time-resolved
measurements, a fast and well-defined turn-
on and turn-off behavior is required, on the
other hand, when the sensor signal is
integrated over the PWM period deviations
will be within the measurement sample.
As no out-of-the-box driver solution is
available, three independent current sources
are created. Each current source is based on
the TI PT6101 buck converters. TI
guarantees that these converters produce a
fully regulated output voltage within 1 ms of
either the release of the Inhibit pin or the
application of power. As this is too slow, the
buck converters have been modified.
The new driver circuitry is displayed in figure 8. The resulting rise and fall times are depicted in figure 9.

Figure 9: I-source rise time (left) and fall time (right) for an LED load in experimental set up
(Ch1: LED current, Ch3: PWM input signal)
The rise and fall times are typically about 5 s, however these times are load dependent.
5.2 Sensors
Depending on the activated color control method, a certain type of sensor is required. Color coordinates feedback
requires filtered sensors, which (more or less) match the Color Matching Functions, whereas flux feedback simply
requires a photodiode. Other color control loops, like temperature feed forward and the combination of the latter two also
require a temperature sensor. Therefore, the experimental set up has been equipped with several optical sensors (attached
to the integrating sphere, but on the opposite side of the LEDs) and a temperature sensor on the heat sink of the LEDs.
This position of the light sensors makes this setup independent of the types of LEDs (packaged or chip-on-submount).
Proc. of SPIE Vol. 5941 59410C-9
3 5 0 4 0 0 4 5 0 5 0 0 5 5 0 6 0 0 6 5 0 7 0 0 7 5 0
W a v e l e n g t h ( n m ]


There is a shield inside the integrating sphere to prevent direct LED light from reaching the sensors. The optical
measurements are oversampled extensively to overcome the effects of the drivers switching-frequency.
5.2.1 Temperature sensor
A simple and effective measurement of the LEDs junction temperature can be obtained through a temperature dependent
resistor (NTC) thermally connected to the MCPCB that the LEDs are mounted on. An NTC sensor is recommended
because it has a suitable sensitivity curve, is available in quite small sizes, has reasonable accuracy (1% thru 10%) and a
wide range of resistances (100 Ohm thru 470 kOhm).
As the NTC is directly connected to the MCPCB (or heat sink), the junction temperature (T
j
) can simply be estimated
as the sum of the heat sink temperature (T
b
) and the thermal resistance (R
j2b
) multiplied by the dissipated power (P
diss
)
(assuming the LED junction heats up almost instantaneously with respect to PWM period):
diss j b b j b b j
P R T T T T + + =
2 2
(9)
5.2.2 Flux sensor
A cheap solution for flux sensors is a photodiode; these are widely available in many different packages. A photodiode
can usually be modeled as a number of parallel current sources (e.g. a signal (I
s
), leakage (I
l
) and noise (I
n
) current
source) and a parallel shunt resistor (R
d
). This resistance (R
d
) is very large (R
d
>>1 M), so the current of the parallel
current sources (I
s
, I
l
and I
n
) add up and I
l
and I
n
can be considered as additive noise.
Unfortunately, the photodiodes sensitivity is not uniform over the visible spectrum: very sensitive for red, but quite
insensitive for blue. On the positive side, the spectral sensitivity is almost linear within the visible range. Combined with
the shifting peak wavelength of the LEDs as their junction temperature rises, the set point can be adapted using a linear
fit of the sensors sensitivity. In figure 10 below, a spectral overview of the LED output and a standard photodiode
response is depicted. The used photodiode is a Siemens SFH213. This is a Silicon-based photodiode with a very short
switching time. It is especially suitable for radiation between 400 nm and 1100 nm. Note that the spectral sensor
sensitivity will change when the sensor temperature changes.
5.2.3 Color-filtered flux sensors
For color coordinates feedback (CCFB), the red, green and blue ratios of the mixed light must be measured. To get an
accurate reading of human color impression, the CIE1931 2 color matching functions (CMF)
5
must be used.
Unfortunately, color filters adjusted to this specific response are available, but rather expensive. In a future lighting
application, cheaper filters must be used and, therefore, an Agilent HDJD-S831-QT333
9
sensor array is used in this
experimental set up. Agilent also utilizes this sensor for their color management system HDJD-J822-SCR00
11
.

Figure 10: LED output (normalized to 1), CMF filter, Agilent sensor (green filter
normalized to 1) and photodiode characteristics (max sensitivity 1)
5, 8, 9

As the sensors do not match the
color matching functions, they need
to be calibrated to match the CMF
responses. In figure 10 below, a
spectral overview of the 2 degree
CMF, LED output, Agilent sensor
response and a standard photodiode
response is depicted. For color
coordinates feedback, the average
light per PWM period is required, so
the sensor signals must be integrated
over each PWM period.
Proc. of SPIE Vol. 5941 59410C-10


6. COLOR ACCURACY RESULTS
The total deviation of the color point of a light source from the targeted color point comprises two components: a color
set point error, which is the error that results when aiming at a target color point with the system operating at a reference
temperature (which is the temperature at which the system is calibrated), and a temperature induced color error, which is
the error that results from temperature variations of the system. Because the influence of the color control loop is most
apparent in the latter component, we will only concentrate on this component.
Below (figure 11) you can find the color point trajectories for target color points of 2500 Kelvin, 4000 Kelvin and
6000 Kelvin on the blackbody locus and at 95% light for the chosen color point. The trajectories have been determined
through repetitive color point measurements while the unit was heating up. During all measurements, the ambient
temperature was about 25 C and this resulted in a reference heat sink temperature (or calibration temperature) of 35 C.
0,000
0,005
0,010
0,015
25 35 45 55 65 75 Theatsink
u'v'
OL TFF FFB FFB&TFF CCFB Target

0,000
0,005
0,010
0,015
25 35 45 55 65 75 Theatsink
u'v'
OL TFF FFB FFB&TFF CCFB Target

0,000
0,005
0,010
0,015
25 35 45 55 65 75 Theatsink
u'v'
OL TFF FFB FFB&TFF CCFB Target

Figure 11: The temperature induced color error trajectories uv vs. heat sink temperature (CCT 2500 K, 4000 K and 6000 K)
Table 2 provides an overview of these color point deviations.

uv
2500 K 4000 K 6000 K minimum average maximum predictions
OL 0.0227 0.0186 0.0142 0.0142 0.0185 0.0227 0.0250
TFF 0.0052 0.0036 0.0040 0.0036 0.0044 0.0052 > 0.0100
FFB 0.0189 0.0133 0.0118 0.0118 0.0154 0.0189 0.0125
FFB&TFF 0.0017 0.0043 0.0052 0.0017 0.0035 0.0052 < 0.0100
CCFB 0.0057 0.0043 0.0038 0.0038 0.0048 0.0057 0.002 0.006
Table 2: The temperature induced color accuracy errors for 2500K, 4000K and 6000K white set points
for a temperature change of 50 C (predictions are at 3500 K and include ageing effects, but measurements exclude these)
The above results show the color stability of each color control loops on short term, meaning that only temperature
effects are compensated. Long term, or ageing effects are not included, however, based on the loops sensors, one
expects only the color point difference of temperature feed forward (TFF) to increase.
From these measurements, one observes that, in general, the color error drops if the color temperature rises, however,
FFB&TFF forms an exception, which may be explained by the low blue sensitivity of the chosen Silicon sensor (see
figure 12). This also means that using a blue enhanced sensor may yield an improvement. In any case, the color point
dependent color stability shows that the performance of color control loops should not be specified for a single color
point. Temperature feed forward (TFF) performs much better than the expected numbers, indicating very good LED
modeling. However, the exclusion of ageing effects means that long-term results will be significantly different. Flux
feedback (FFB) is nearly as bad as open loop and as such cannot be recommended. On average, FFB&TFF shows the
best performance.
Earlier publications for FFB&TFF and CCFB color control loops show comparable results: For instance in IEEE
paper by S. Muthu et.al.
2
a FFB&TFF method is discussed, which features a color error of uv = 0.004 (CIE 1960) for a
temperature rise of about 45 Celsius. The same authors show in their IAS conference paper
3
that their equivalent CCFB
method manages a uv = 0.006 (at 6500 Kelvin color point) for a temperature rise of about 50 Celsius. The paper by A.
Perduyn et.al.
4
, an additional CCFB equivalent method is presented, which generates uv = 0.003 (at D65 color point)
for a temperature rise of about 45 Celsius. The performance differences between these equivalent CCFB methods are
caused by differences in the sensor sensitivity curves, where A. Perduyn
4
uses sensors closely approximating the CIE
1931 color matching functions (see figure 13).
Despite these promising experimental results, one should realize that advances in LED technology occur at very high
pace. This means that one should be aware of any changes in LED design that might influence the performance of a color
control loop. As some of the presented methods rely on certain LED parameters (e.g. d/dT or
-1
d/dT), the
Proc. of SPIE Vol. 5941 59410C-11


distribution and average value of these parameters is very important e.g. for reliability figures. Besides the statistical
properties of these empirical parameters, the accuracy of the thermal model of the system will also influence the
performance of certain color control loops. In all cases, integrating Silicon based sensors into a lighting fixture influences
the sensors spectral sensitivity, because they also heat up, in turn, this influences the performance of the color control
loop. Therefore, an improvement can be made by also compensating for temperature-induced changes to the sensors
spectral sensitivity.
Also, ageing will have an effect on the performance of the temperature feed forward method. If ageing only changes
the lumen output, all other methods should be able to handle this. However, if also the peak wavelength changes, of the
presented methods only the color control loop based on an integrated color point (CCFB) will be able to cope with this,
although other suitable sensor solutions are likely to exist. Additionally, the effects of sensor ageing may influence the
color stability.
7. CONCLUSIONS
Although LEDs promise a new era in lighting, their color instability needs to be solved by using different LED types or
by using an additional color control loop. In this paper, four methods have been presented and compared with respect to
their short-term performance. The color errors for each of the color control loops have been determined for a temperature
range of about 50C and for color temperature between 2500 Kelvin and 6000 Kelvin. In this range, we find that, on
short term, all color control loops show a significant improvement in the color error, except for the color control loop
based on flux measurements of each primary color (FFB), which disregarding ageing effects performs nearly as
mediocre as open loop. However, the color control loop based on the heat sink temperature (TFF) cannot offer color
stability when the LED ages, which is expected to be significant. On average, FFB&TFF shows the best results. These
results have been obtained using LED tristimulus data and sensor data at a single (junction) temperature. In addition,
LED parameters d/dT and T
0
have been determined empirically on a different LED set.
It is also apparent that the color error is not constant for all colors; we already see larger differences between color
points on the blackbody line from 2500 Kelvin to 6000 Kelvin. This means that the color stability of a color control loop
must be specified for a certain color point range. In relation to this, FFB&TFF shows opposite behavior with respect to
the other methods, which is most likely caused by the chosen sensor; using a blue enhanced may result in a decreased
color error.
The two color control loops based on temperature feed forward (TFF and FFB+TFF) both require prior knowledge of
empirical LED properties, therefore a color control loop based on an integrated color point (CCFB) is the most robust
method as no empirical LED properties are required, however, it does need rather expensive sensors. FFB+TFF is
suitable as less expensive alternative for CCFB, especially when the LEDs have matured and their parameters show less
spread from batch to batch. A less expensive color sensor may also yield an acceptable color error for the CCFB method,
while still offering robustness.
REFERENCES
1 A. Zukauskas, Quadrichromatic white solid state lamp with digital feedback, Proc. Of the Soc. of Photo-
Optical Instrumentation, 5187, pp 185-198, 2004.
2 S. Muthu and F. J. P. Schuurmans, Red, green, and blue LEDs for white light illumination, IEEE Journal in
Quantum Electronics, 8, pp 333-338, 2002.
3 S. Muthu, F. J. P. Schuurmans and M. D. Pashley, Red, green and blue LEDs based white generation: issues and
control, Proc. of the 2002 IEEE Industry of Applications, 1, pp 327-333, 2002.
4 A. Perduijn, S. de Krijger, J. Claessens et.al., Light output feedback solution for RGB LED backlight
applications, SID2003.
5 G. Wyszecki and W.S. Stiles, Color science. New York: Wiley, 1982.
6 A. Zukauskas, Introduction to Solid-State Lighting. New York, Wiley, 2002.
7 Datasheet Lumileds Luxeon emitters, Lumileds website [online]: http://www.lumileds.com
8 Datasheet Siemens photodiode SFH213, Siemens website [online]: http://www.siemens.com
9 Datasheet Agilent color sensor module HDJD-S831-QT333, Agilent website [online]: http://www.agilent.com
11 Datasheet Agilent color management system HDJD-J822-SCR00, Agilent website [online]:
http://www.agilent.com

Proc. of SPIE Vol. 5941 59410C-12

Das könnte Ihnen auch gefallen