View metadata, citation and similar papers at [Link].

uk brought to you by CORE

provided by DSpace at University of West Bohemia

LED application specific calibration method

for color measurements
Thomas Nimz, Christian Bodenstein and Mladen Berekovic

Abstract—More and more light applications use multi-

ple light emitting diodes to generate defined light spectra. 1 2° observer
Often reaching defined brightness levels and an exact TCS002

relative sensitivity
color coordinate or color temperature is desired. The 0.8
LEDs change their light spectra over time or under
influence of different environmental temperatures. One 0.6
possibility is to compensate the resulting color shift of
the illumination by integrating a color sensor. Before the 0.4
light can be re-adjusted or regulated, a precise color
measurement must be guaranteed. This paper studies 0.2
the color measurement accuracy, based on specific filter
curves of sensors and models of red, green, blue and 0
white LEDs fitted to measure data. 400 500 600 700
λ [nm]
I. I NTRODUCTION
The spectrum of light emitting diodes (LEDs) Fig. 1. Characteristic curves of 2◦ standard observer and True Color
change depending on the ambient temperature, the uti- Sensor.
lized current or the age. Even spectral changes due to
pulse width modulation (PWM) can be detected. When errors caused by infrared light. The second sensor type
marketing LED illuminations, the question about long- is called True Color Sensor [4] and the characteristic
term availability of the LEDs or similar future batches curves correspond to the 2◦ standard observer. Due to
arise. In [7] four solution approaches to compensate the tolerance in manufacturing process, it is not possible
problems of LEDs are addressed. A promising solution to produce filter curves without any deviation. Fig. 1
is direct feedback control by using the tristimulus shows the remaining differences in spectral sensitivity
values of the emitting light source. One presented idea between the standard observer and the sensor. Because
is to use a color sensor to measure the LED light of this it is also necessary to calibrate the True Color
color and regulate it actively via LED control features. Sensor. For the determination of the influence of the
This concept is for example mentioned in [1], [6] and manufacturing tolerances the spectral sensitivities were
[5]. The accuracy of the feedback control system is measured for 10 RGB and 10 True Color Sensors.
significantly influenced by the measurement accuracy
of the sensor. Thus, one important part is the sensor III. LED M ODELLING
calibration. [2] demonstrates a calibration method for
a RGB sensor along the black-body curve. The developed LED models serve as a basis for
In this paper the measurement accuracy over the com- the color measurement simulations. The goal of this
plete LED system-specific color space is considered. development was to find a function for each LED,
The authors of this paper decided to develop a simula- because the LED datasheets includes no information
tion environment to compare multiple sensors using an about the spectral shift depending on the current or
RGB and RGBW LED system. First we measured the PWM. With the indication of the PWM, current and
spectral sensitivities of the chosen color sensors. The the ambient temperature the function should calculate
second step was the generation and fitting of a model the LED spectrum. We decided to build a test setup and
of the red, green, blue and white light emitting diodes. measure the RGBW light emitting diodes. In future it
The following measurement simulation is based on the will be possible to compare the simulated results with
selected sensor filter curves and a theoretical model of the actual results.
the LEDs.
A. Experimental Setup
II. C OLOR S ENSORS
On the one hand the experimental setup, shown
Two different types of color sensors are considered in Fig. 2, allows the measurement of the LEDs. On
in this paper. Both sensors have three characteristic the other hand the test setup contains a color sensor
curves. The first sensor is a typical RGB sensor with to validate the theoretical results in a future study.
the highest sensitivity of the channels and peak values The ambient temperature of the LEDs can be adjusted
at 454, 560 and 660 nm [3]. The sensor has an inte- between 30 and 80◦ C. The LED currents can be
grated infrared blocking filter to prevent measurement increased stepwise up to 700 mA. Furthermore the
LEDs can be dimmed by using a 12 bit PWM signal.
c
ISBN 978-80-261-0386-8, University of West Bohemia, 2015 IEEE
      !   influence of the pulse-width modulation. The resulting
     color deviation differs between the LEDs. The red LED
has the highest color shift of 0.0016∆u’v’ compared
   to 0.0015∆u’v’ for the blue LED, 0.0008∆u’v’ for the
green LED , and 0.0006∆u’v’ for the white LED.
  
IV. C OLOR M EASUREMENT S IMULATION

    
 
" # $ %
  
 
 
       
     
  
     

Fig. 2. Experimental setup.    



B. LED Modelling && &&

The model of each LED is based on 756 spectra,
'&&(' 
measured with the spectrometer. Using the presented
test setup the light emitting diodes were considered
Fig. 4. Workflow of the simulation process.
at six temperature levels between 30 and 80 de-
grees Celsius. At each temperature level the LEDs
Figure 4 shows the workflow to simulate the mea-
were measured with different PWMs between 0 and
surement process and compare the result with the
100% at an interval size of 5% at several current
observer as reference. In the first step the illumination
levels. At the end of each measurement process we
spectrum is calculated by combining the LED models.
receive a quadruple consisting of temperature, current
It is possible to set the current level, the PWM and
step, PWM step and the associated value for each
the temperature for each LED. The resulting spectrum
wavelength. So it is possible to determine a func-
is the input to calculate the tristimulus values XYZ
tion f (temperature, current, P W M ) for a specific
via the standard observer. On the other hand the
wavelength. 756 quadruples were used to determine
convolution between the illumination spectrum and the
the coefficients for the chosen linear interpolation. The
sensor characteristics result in the simulated sensor
subsequent validation showed almost no differences
response values. It is necessary to multiply the result
between the input spectra and the generated spec-
vector with the calibration matrix M to receive the
tra. Figure 3 shows several simulated spectra of the
XYZ values (shown in equation 1).
four used LEDs. The black lines in the figure show ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
measured spectra taken by a spectrometer. The blue X m1,1 m1,2 m1,2 Ch1
spectra between the corresponding two black spectra ⎝ Y ⎠ = ⎝m2,1 m2,2 m2,2 ⎠ ∗ ⎝Ch2⎠ (1)
are calculated by the interpolation function f. The Z m3,1 m3,2 m3,2 Ch3
4X
u′ = (2)
1 X + 15Y + 3Z
increase current 9X
v′ =
relative intensity

increase (3)
X + 15Y + 3Z
Temperature 
0.5 ∆u′ v ′ = (u′2 − u′1 ) + (v2′ − v1′ ) (4)
increase
PWM Because of the perceptual uniformity the CIELUV
color space is recommended to measure color devi-
ation. In the last step both results are converted to the
0 color space (Eq. 2 and 3) and the color as well as
450 500 550 600 650 700 brightness deviation is calculated (Eq. 4).
λ [nm]
M = (AXY Z ∗ A′sensor ) ∗ (Asensor ∗ A′sensor )−1 (5)
Fig. 3. Spectra interpolation [blue spectra] between two measured
led spectra [black spectra]. First we use the standard calibration with pseudo-
inverse (see equation 5). This calibration method is
following simulation of color measurement used the possible for RGB and True Color Sensor, however it
LED model with a constant temperature and current. is required to calibrate each sensor individually. At the
Thus the remaining cause of the spectral shift is the beginning the calibration matrix is calculated within
a calibration process. The matrix Asensor consist of specific filter curves with only little deviation to the
the sensor values for each LED at full PWM. AXY Z standard observer reached a ∆u’v’ smaller than 0.001
includes the corresponding reference values XYZ of (data not shown). During the actual simulation the
the LEDs under a 2◦ standard observer. Both matrices LED ambient temperatures was constant. However, if
have the same size, for example 3x3 with a RGB or a small spectral shift over the PWM results in a color
3x4 with a RGBW LED system. deviation - it is also necessary to review the influence
of the LED shift during temperature changes. Usually,
A. RGB LED System
the calibration matrix is calculated at 30◦ C ambient
To compare the measurement accuracy between the temperature of the LEDs. Subsequently, we increased
RGB sensor and the True Color Sensor it is necessary the LED temperature and simulated the color measure-
to review several LED input combinations. For the ments for all LED input combinations. In this scenario
RGB LED system we chose all combinations of the we ignored the temperature influence of the sensor.
three LEDs from 0 to 100% PWM with a 12.5% Fig. 7 shows the influence of the ambient temperature.
interval size. Figure 5 shows the color measurement The result is that the measurement deviation increases
deviation for one of the RGB and True Color Sensor at with the temperature. That means that the accuracy
the resulting 728 target points. The average deviation is worse the higher the spectral shift is between the
of the RGB sensor ∆u’v’ 0.0017 is 11 times higher calibration spectra and spectra of the LEDs at the
than ∆u’v’ 1.55 · 10−4 for the True Color Sensor. The operating temperature.
boxplot Fig. 6 shows the accuracy for all twenty color
sensors. Both color sensors have the same behavior, )3
but the True Color Sensor is much more accurate. All x 10
sensors measured the brightness better than +/- 0.5%
∆Y. The worst case is a LED combination where only 8
6
∆u’v’

0.02
RGB Sensor 0002 4
∆u’v’

0.01 True Color 0001

2
0
0 30 (Cal) 40 50 60 70 80
0 200 400 600
T [°C]
LED combinations LEDs

Fig. 5. Color deviation for True Color and RGB sensor. Each Fig. 7. Change of the measurement deviation by increasing the
measurement series include 728 target points (RGB LED System). ambient temperature of the RGB LED system. The True Color
Sensor was calibrated at 30◦ C ambient temperature of the LED.
the red LED is used and dimmed via PWM. This is due
to the fact that the red LED has the highest color shift B. RGBW LED System
over the PWM. The simulation shows that the small
LED color shifts due to the PWM decrease the sensor For the measurement simulation with the RGBW
accuracy. It is possible to increase the accuracy by ex- LED system we chose all combinations of the LEDs
tending the calibration. For example the measurement from 0 to 100% PWM with 25% interval size. First
accuracy can be doubled via additional calibration sets we simulated the measurement process with the same
for each LED at several PWM stages. Furthermore, the calibration method as used in the RGB system. But
the results showed that the simple way to extend
)3 the calibration with an additional data set for the
x 10 white LED is not accurate enough. For example the
8
best True Color Sensor has a average deviation of
6 0.003∆u’v’ and in the worst case of 0.0182∆u’v’.
Furthermore the sensor shows a strong brightness
∆u’v’

4 deviation (+/- 4% ∆Y ). One solution is to increase

the number of data sets during the calibration process.
2 On the one hand it is necessary to find the right
combination of data sets and on the other hand it is
0 important to use minimal effort. In [8] a optimization
True color sensors RGB sensors program is used to find the optimum between the
number of input data and accuracy. This approach
Fig. 6. Comparison of the measurement accuracy between both requires many initial measurements.
sensor types. Each box include one measurement series about 728
led configuration (RGB LED System).
In this paper we looked for a solution without least
color deviation between the ten color sensors of one calibration effort. An other way to get better results is
type differ greatly. The quality of the sensors spectral to integrate further information about the actual LED
sensitivity are the reason for these differences. Using control sets. If the sensor logic has information about
 The calculated data set is integrated as a fifth data set
(additional to the four LED data sets) during the calcu-
lation of the calibration matrix (Eq. (5)). Therefore, it
is required to calculate the calibration matrix for each
measurement process. Figure 9 evaluates the three cali-
bration methods. With the extended calibration method
(CT) all True Color Sensors had a color measurement
deviation of ≤ 0.002∆u’v’. The brightness can be
measured at deviation levels better than +/- 0.5% Y.
The comparison between the two sensor types shows
that for an RGBW LED system the usage of an RGB
sensor is inferior. The measurement accuracy of the
RGB sensor is greater than 0.0121 ∆u’v’ and +/-3.1%
∆Y (data not shown). The first measurements with the
Fig. 8. Distribution of the color measurement deviation of a
test system confirmed the simulation. In future studies
True Color Sensor and a RGBW LED system by using the PWM we would also include the temperature dependence of
calibration. the chosen sensor.

the actual PWM of the LEDs it is possible to calculate 0.03

a customized correction matrix for each measurement
process during the LED operation. At the beginning in
0.02
the calibration phase it is only necessary to measure the
∆u’v’

four data sets (for each LED). For each measurement

during the LED operation the correction matrix will
0.01
be calculated with (6) and (7) as calibration input. In
→
which the vector p corresponds to the actual PWM of
the LEDs. 0
⎧ ⎫ Standard PWM CT
⎨ p r X r p g X g p b X b pw X w ⎬
AXY Z = pr Xr pg Yg pb Yb pw Yw (6) Fig. 9. Color measurement deviation for 10 True Color Sensors
pr Z r pg Z g pb Z b pw Z w depending on the chosen calibration method. Each box include one
⎩ ⎭
measurement series about xxx led configuration using a RGBW LED
System.
⎧ ⎫
⎨pr C1r pg C1g pb C1b pw C1w ⎬
ASen = pr C2r pg C2g pb C2b pw C2w (7)
pr C3r pg C3g pb C3b pw C3w
⎩ ⎭
V. C ONCLUSIONS
Fig. 8 shows the measurement deviation of one True This paper shows a simulation of different calibra-
Color Sensor at the 624 test color point. The integration tion methods for LED light measurements via color
of the PWM factor improves the measurement results, sensors. The simulation is based on the sensor charac-
for example the average deviation decreases from teristics and several experimentally fitted LED models.
0.003 to 0.0009 ∆u’v’. But it is obvious, that the Using the models it is possible to generate the LEDs
calibration has problems with LED combinations spectrum depending on current, PWM and temperature.
where all four LEDs are powered with different PWM In both selected applications, RGB and RGBW LED
levels. For example the worst case in the simulation systems, the True Color Sensor achieves better results
is the RGBW combination of [0.75 0.5 0.25 1]. than the RGB sensor. The calibration matrix calculated
with the standard method can be used to measure an
First calculations showed that the results were better RGB light source with sufficient accuracy. To measure
if the calibration matrix is calculated with additional the light of a RGBW LED system it is useful to extend
data sets near the actual LED combinations. It is nei- the calibration process with additional calculated data
ther useful to integrate data sets all over the color space sets. The extended calibration sets can be calculated at
nor all possible LED combinations. But it is possible each measurement cycle. Thus, the initial calibration
to set a focus with a theoretical data set.→Due to the process is unaltered. The results show that in the
additive color mixing, we can calculate XY Z calc with simulation the True Color Sensor reached a maximum
the actual PWM setting and the initial calibration data ∆u’v’ of 0.002 and a brightness deviation of +/-0.5%
→
sets at full PWM (Eq. 8). The associated Chcalc can ∆Y. The calibration method also improves the result
also be calculated with the actual PWM control and the of the RGB sensor. These results however are still
sensor response of the calibration data at full PWM. inferior to the accuracy of a True Color Sensor within
⎛ ⎞ ⎛ ⎞ an RGBW system.
Xcalc pr X r + pg X g + pb X b + pw X w
⎝ Ycalc ⎠ = ⎝ pr Xr + pg Yg + pb Yb + pw Yw ⎠ R EFERENCES
Zcalc p r Z r + p g Z g + p b Z b + pw Z w [1] C. Martiny-A. Hilgers B. Ackermann, V. Schulz and X. Zhu.
(8) Control of leds. In Industry Applications Conference, 2006.
 41st IAS Annual Meeting. Conference Record of the 2006 IEEE,
pages 2608 – 2615. Phillips, IEEE, 2006.
[2] J.-S. Botero Valencia ; F.-E. Lopez Giraldo ; J.-F. Vargas Bonilla.
Calibration method for correlated color temperature (cct) mea-
surement using rgb color sensors. In Image, Signal Processing,
and Artificial Vision (STSIVA), 2013 XVIII Symposium of, pages
1–6, 2013.
[3] MAZeT GmbH. Data Sheet MRGBiCS, August 2013. Data
Sheet Integral RGB Sensor - LCC8 Rev. 1.2.
[4] MAZeT GmbH. Data Sheet MTCSiCF, February 2014. Data
Sheet Integral True Color Sensor - QFN16 Rev. 1.8.
[5] Subramanian Muthu, Frank JP Schuurmans, and Michael D
Pashley. Red, green, and blue leds for white light illumination.
Selected Topics in Quantum Electronics, IEEE Journal of,
8(2):333–338, 2002.
[6] Y. M. Lai Senior Member S. K. Ng, K. H. Loo and Chi K. Tse.
Color control system for rgb led with application to light sources
suffering from prolonged aging. IEEE TRANSACTIONS ON
INDUSTRIAL ELECTRONICS, VOL. 61, NO. 4, APRIL 2014,
2014.
[7] Frank J. P. Schuurmans Subramanian Muthu and Michael. D.
Pashley. Red, green, and blue led based white light generation:
issues and control. In Industry Applications Conference, 2002.
37th IAS Annual Meeting. Conference Record of the (Volume:1
), pages 327 – 333 vol.1, 2002.
[8] [Link] [Link]. Optimierung der absolutgenauigeit von true
color farbsensoren zur regelung von n-farbigen led- leuchten.
In LICHT 2012 Tagungsband, 2012.