IEEE JOURNAL ON SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 8, NO. 2, MARCH/APRIL 2002
`
`333
`
`Red, Green, and Blue LEDs for
`White Light Illumination
`
`Subramanian Muthu, Associate Member, IEEE, Frank J. P. Schuurmans, and Michael. D. Pashley
`
`Abstract—The rapid improvement of the white light efficacy
`achievable with light-emitting diodes (LEDs) opens up new oppor-
`tunities in the general illumination market. An LED light source
`made of red, green, and blue LEDs (RGB-LEDs) can provide the
`unique feature of color variability, allowing the user to select the
`desired color point of the lamp. The white light color accuracy re-
`quired in the general illumination market is a challenge for LEDs.
`The variation in lumen output and wavelength for nominally iden-
`tical LEDs and the change in these parameters with temperature
`and time result in an unacceptably high variability in the color
`point of white light from RGB-LEDs.
`In this paper, we show that these problems can be overcome with
`suitable feedback control schemes that can be implemented in a
`practical LED lamp. We present results of experiment and theo-
`retical modeling that shows the performance that can be achieved
`with a number of different control schemes.
`Index Terms—Color accuracy, feedback control, light-emitting
`diodes, white light illumination.
`
`I. INTRODUCTION
`
`T HE RAPID development of light-emitting diodes (LEDs)
`
`over the last few years has opened up new opportunities in
`the general illumination market [1]. The efficacy of white light
`from LEDs is now over 20 lm/W, which already exceeds that
`of incandescent lamps [2]. By 2005, it is forecast that LED ef-
`ficacy will reach 50 lm/W [3], which approaches that of com-
`pact fluorescent lamps. In addition, higher power packages are
`becoming available that enable compact lighting systems with
`LEDs. However, additional challenges remain. The general illu-
`mination market has strict requirements on the quality of white
`light—lamps of the same type must all appear to have the same
`color point. In this paper, we discuss these requirements, the is-
`sues with LEDs that make these requirements a challenge, and
`how to meet these requirements.
`There are several approaches using LEDs to achieve white
`light [4]. One approach is to use a blue or UV LED to excite one
`or more phosphors to give white light. In this paper, we focus on
`the use of red, green, and blue LEDs (RGB-LEDs) to produce
`white light. The advantages of RGB-LEDs are that they pro-
`vide a light source that can have a variable color point, and the-
`oretically can provide the highest efficiency LED-based white
`light. The ability to change the color point of the lamp provides
`a new feature to general illumination that has the potential to
`generate new applications and hence new market opportunities.
`
`Manuscript received December 11, 2001; revised January 23, 2002.
`S. Muthu and M. D. Pashley are with Philips Research, Briarcliff Manor, NY
`10510 USA (Subu.Muthu@philips.com; Michael.Pashley@philips.com).
`F. Schuurmans is with Philips Research, Eindhoven, The Netherlands
`(Frank.Schuurmans@philips.com).
`Publisher Item Identifier S 1077-260X(02)03763-2.
`
`Fig. 1. The 1964 CIE (u; v) coordinate system showing the coordinates of
`InGaN and AlInGaP LEDs. Also shown is the blackbody line over a range of
`color temperatures from 2 000 K to 10 000 K.
`
`A key challenge for RGB-LEDs is to maintain the desired white
`point within acceptable tolerances. This arises from the signif-
`icant spread in lumen output and wavelength of manufactured
`LEDs, and the changes in LED characteristics that occur with
`temperature and time. Maintaining the desired white point can
`only be achieved with feedback schemes to control the relative
`contributions of red, green, and blue to the white light.
`
`II. WHITE LIGHT REQUIREMENTS
`
`It is well known that red, green, and blue LEDs can be com-
`bined to produce white light. This can be represented on a chro-
`maticity diagram. The most common chromaticity diagram is
`the CIE 1931 coordinate system
`[5]. However, the just no-
`ticeable color difference is not a constant length over
`space.
`By applying a linear transformation a new coordinate space can
`be generated where the just noticeable color difference is ap-
`proximately uniform. A number of such transformations exist.
`For the purposes of this paper we use the CIE 1960 UCS system
`, as shown in Fig. 1. By combining three different color
`LEDs, it is possible to produce any color point (
`coordi-
`nate) that lies within the triangle formed by the
`coordi-
`nates of the three LEDs. In almost all white light illumination
`applications, the resultant color point must lie on, or very close
`to the locus of points that follows the line of a black-body radi-
`ator (shown in Fig. 1). An incandescent lamp has a black-body
`temperature of approximately 2700 K. Most fluorescent lamps
`are designed to have a color temperature of between 3000 K and
`5000 K, dependent on the application and preference of the user.
`Another key requirement of illumination relates to the spec-
`tral properties of the white light source. Our perceived color of
`objects depends upon the spectrum of incident light upon them.
`
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on August 04,2023 at 22:30:29 UTC from IEEE Xplore. Restrictions apply.
`
`1077-260X/02$17.00 © 2002 IEEE
`
`APPLE 1051
`Apple v. Masimo
`IPR2022-01299
`
`1
`
`

`

`334
`
`IEEE JOURNAL ON SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 8, NO. 2, MARCH/APRIL 2002
`
`TABLE I
`VALUES OF COLOR RENDERING INDEX
`(R ) REQUIRED FOR A NUMBER OF
`ILLUMINATION APPLICATIONS.
`
`TABLE II
`VALUES OF COLOR RENDERING INDEX (R ) THAT CAN BE ACHIEVED
`WITH THE COMBINATION OF TWO, THREE, AND FOUR
`DIFFERENT WAVELENGTH LEDS.
`
`Fig. 2. The calculated shift in the (u; v) color coordinates as a result of a
`change in the flux of the red, green, or blue LEDs in an RGB-LED.
`
`pends upon the application. To quantify the color error of a light
`source, we introduce the quantity
`where
`
`A red object illuminated with light that is drastically deficient in
`red will appear black. The lighting industry uses a standard color
`rendering index
`to determine the color rendition properties
`of a light source. It is based on the components of eight standard
`spectra in the white light source as compared to a black-body ra-
`diator with the same color temperature as the light source. Thus,
`an incandescent lamp has an
`value of 100. Typical fluores-
`cent lamps used in offices have an
`of 80. The required
`value depends upon the application. Typical examples are given
`in Table I.
`The illumination of goods in a retail store is typically the most
`demanding application for color rendering index. The precise
`requirements depend upon the goods being displayed. As the
`goods on display are changed, different color points may be de-
`sired. With conventional light sources, this means that the lamp
`has to be changed. RGB-LEDs will allow the desired color point
`to be achieved simply by adjusting the ratio of RGB illumina-
`tion. Typical indoor living space is illuminated with sources that
`have an
`of 80. The color temperature can vary from 2700 K
`with incandescent lamps, to the 4000 K typically used in offices
`with fluorescent lighting. RGB-LEDs will allow one lamp to
`provide a range of color temperatures. General outdoor illumi-
`nation such as street lighting puts the lowest demands on color
`rendering with
`of 40 or less being common.
`that can be achieved with LEDs depends on the white
`The
`spectrum. The white spectrum is made up of the individual LED
`spectra, and thus, depends on the wavelengths selected, and the
`number of different wavelength LEDs used to make white light.
`Table II shows the
`values that can be achieved with the
`mixing of two, three, and four different wavelength LEDs.
`RGB-LEDs can achieve the required
`values provided that
`the correct LED wavelengths are selected. Most applications can
`be addressed by the selection of three different wavelengths.
`A major requirement of many illumination applications is that
`the light source has the required color point (i.e.,
`coordi-
`nate), and that it stays at its color point over time. It is viewed
`as unacceptable if all fluorescent lamps lighting in an office
`area are not the same color. This raises the question: what is
`the required specification? There is no single answer—it de-
`
`being the color coordinates of the light source, and
`are the required color coordinates. This is simply the
`distance in
`color space of the lamp from the desired color
`point (see Fig. 1). As a point of reference, fluorescent tubes are
`usually specified to be within
`of their designed
`color point. Some discharge lamps have larger deviations of
`over
`, and are regarded as unacceptable by some
`customers. As we will show below, the demands on color
`reproducibility of the general illumination market provides a
`severe challenge for RGB-LEDs. Color point reproducibility is
`also a severe challenge for most approaches to phosphor-LEDs.
`No phosphor-LEDs on the market today meet the color point
`reproducibility requirements of the general illumination market.
`
`III. THE COLOR STABILITY OF RGB-LEDS
`
`Conventional light sources (fluorescent, incandescent, etc)
`can be manufactured very reproducibly such that the lumen
`output and color points are highly consistent (a few percent
`in flux and a
`of less than 0.003). As a result, the general
`illumination market has grown to expect this level of consis-
`tency. The manufacturing process for LEDs, on the other hand,
`does not provide this level of consistency. Nominally identical
`LEDs can vary in light output by over a factor of two, and the
`wavelength can vary by many nanometers. Lumen output and
`wavelength also change with temperature [6] and lumen output
`changes over time in a way that cannot be accurately predicted.
`These factors all influence the color point that is obtained by
`mixing the light from a combination of different wavelength
`LEDs. We now discuss the quantitative effect of these LED
`characteristics based on white light from RGB-LEDs.
`The largest impact on color point of RGB-LEDs comes from
`changes in light output of the individual LEDs. This can be as
`a result of aging, or from the initial spread in the performance
`of the LEDs used in the lamp. Fig. 2 shows a calculation of the
`color error that occurs if any one of the red, green, or blue com-
`ponents changes in intensity. At a color temperature of 3000 K, a
`change of less than 10% in intensity of either green or red moves
`the color point by
`, already outside the specifica-
`tion of a fluorescent lamp. This is a very small intensity change
`
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on August 04,2023 at 22:30:29 UTC from IEEE Xplore. Restrictions apply.
`
`2
`
`

`

`MUTHU et al.: RED, GREEN, AND BLUE LEDs FOR WHITE LIGHT ILLUMINATION
`
`335
`
`electrical, or optical sensors. The output of these sensors is fed
`to a feedback controller, which adjusts the current to the red,
`green, and blue LEDs to produce the desired white light output.
`In this section, we describe a number of different approaches
`to feedback control: temperature feed-forward compensation,
`flux feedback, combined temperature and flux control, and
`feedback of the color coordinates of the white light.
`
`A. Temperature Feed Forward Compensation
`The simplest measurement to implement is temperature. It is
`not practical to directly measure the junction temperature of the
`LED, and therefore, the temperature of the heatsink on which
`the LEDs are mounted is measured. Thus, only an indirect mea-
`sure of the junction temperature is made. As discussed above,
`the light flux and wavelength of an LED both vary with temper-
`ature. If the color point is correct at an initial set temperature,
`then the white color point can be maintained as the temperature
`changes provided that the temperature dependence of the flux
`and wavelength of each color LED is known. At each tempera-
`ture the required fluxes of the red, green, and blue LEDs must
`be calculated based on the calculated wavelengths for that tem-
`perature. The currents required to produce that flux must also
`be calculated for each color LED. A problem with this method
`of compensation is that the temperature dependence of the flux
`and wavelength are not precisely known. These LED param-
`eters have a significant distribution just as the efficiency and
`wavelength do (see Section III). This introduces significant er-
`rors in the resulting white color point. An additional problem
`with this simple compensation scheme isthat it does not correct
`for changes in LED flux with time. Given the variability in the
`aging characteristics of LEDs, adding a simple correction based
`on hours of operation would not adequately address the issue.
`
`B. Feedback Control of the LED Flux
`Photodiodes can be used to measure the LED flux of each
`color component directly. The feedback controller simply has to
`maintain the preset flux from each color component to roughly
`maintain the white point. This can be done with a set of three
`photodiodes, each photodiode placed to detect only a single
`color component. It is also possible to use pulsing techniques
`such that only a single photodiode is needed to monitor each of
`the three color components. Feedback control of the fluxes of
`each of the color components will correct for the LED aging and
`the variation of LED flux with temperature. This provides im-
`proved white light control compared to that obtained with tem-
`perature feed forward alone, and does correct for the aging of
`the LEDs. The disadvantage of this approach is that it does not
`correct for the shift in wavelength with temperature. Calculation
`shows that temperature changes of 20 C can result in a color
`point shift
`of more than 0.005.
`
`C. Combined Temperature and Flux Feedback
`An improved feedback control system can be achieved by
`combining both temperature feed forward and flux feedback.
`This has all the advantages of the flux feedback discussed above,
`and uses the temperature feed forward to allow corrections to be
`made for the shift in wavelength with temperature. This scheme
`
`Fig. 3. The calculated shift in (u; v) coordinates of an RGB-LED as the
`temperature is changed in increments of 20 C. The RGB-LED has a color
`temperature of 3 500 K for a junction temperature of 60 C.
`
`compared with the variability in nominally identical LEDs. Rel-
`ative changes over lifetime between the different LEDs can be
`far greater than 10%.
`Change in temperature of the LED pn junction leads to
`changes in light output, wavelength and spectral width. These
`all influence the resulting color point of the RGB-LED. This
`is illustrated in Fig. 3, which shows the calculated change in
`color point on the
`plane as the temperature changes
`in increments of 20 C. The system is set to be on the black
`body locus at a junction temperature of 60 C, with a color
`temperature of 3500 K. The calculation is based on typical
`temperature coefficients of the LEDs. A shift in temperature
`of only 10 C moves the color point by
`. The
`largest contribution to this shift is the reduction of light output
`of the red LED as the temperature increases. As a result, the
`color point moves toward the blue-green. The red LEDs (or
`any AlInGaP-based LED), typically reduces its light output by
`10–15% for every 10 C increase in temperature. If it were
`possible to reduce the temperature sensitivity of the red LEDs,
`the stability of white light from RGB-LEDs with temperature
`could be significantly improved.
`In addition to the effects already discussed, the peak wave-
`length of an LED also shifts with current. Thus, as the inten-
`sity of RGB-LEDs is adjusted by changing the amplitude of the
`drive current to each of the LEDs, the color point of the combi-
`nation will change. While this effect limits the accuracy of the
`color point, it is typically less critical than the effects shown in
`Figs. 2 and 3.
`Changes in light output and peak wavelength with temper-
`ature, and changes in light output over time mean that factory
`calibrations will not be sufficient to produce a stable white light
`RGB-LED product. The large variability in the performance pa-
`rameters of LEDs makes compensation schemes based on tem-
`perature measurement and time inadequate. The problem can
`be solved with appropriate feedback schemes used to control
`the color point. We now discuss how this can be done, and the
`performance those feedback schemes can achieve.
`
`IV. FEEDBACK SCHEMES
`
`There are several measurable quantities that can be used for
`compensation and feedback control schemes using thermal,
`
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on August 04,2023 at 22:30:29 UTC from IEEE Xplore. Restrictions apply.
`
`3
`
`

`

`336
`
`IEEE JOURNAL ON SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 8, NO. 2, MARCH/APRIL 2002
`
`Fig. 4. Block diagram of a control system with temperature feed forward and
`LED light output feedback.
`
`still relies on knowing the temperature dependence of wave-
`length on temperature, and thus suffers from the spread in LED
`characteristics. Fig. 4 shows a block diagram of such a con-
`trol system. The compensation system supplies reference red,
`green, and blue light outputs as a function of temperature to
`three independent single-input-single-output (SISO) feedback
`controllers, which regulate the RGB-LED light outputs to the
`reference values.
`
`D. Feedback of the Color Coordinates
`Direct control of the white light from RGB-LEDs can be
`achieved by measuring the color coordinates of the white light.
`The measurement requires sensors with spectral responses
`matching the CIE 1931 color matching functions. The feedback
`signal then gives the
`color coordinates. The sensors
`would consist of three photodiodes each covered with an
`appropriate optical filter. A properly designed controller then
`directly regulates the white light to the target color point. A high
`degree of color accuracy is possible with this scheme. However,
`there can be errors in sensing the tristimulus values due to
`inaccuracies in the color filters used. This tristimulus feedback
`control overcomes the variability in LED performance since it
`directly controls the white light, and not the components that go
`to make up the white light. Such feedback schemes, therefore,
`have the potential to be more accurate than the control methods
`described above.
`Either pulsewidth modulation (PWM) or amplitude modula-
`tion (AM) can be used to supply the LED forward current with
`the feedback schemes presented above. However, PWM and
`AM driving conditions affect the spectral response of the LEDs
`differently. A change in the amplitude of the drive current of an
`LED causes a shift in its wavelength, as described above. Con-
`trol methods that do not measure the color coordinates directly
`must take account of this if AM drive schemes are used. In the
`case of PWM driving of the LEDs, the current does not affect
`the wavelength of the LED as the dc forward current is always
`at the same value.
`
`V. AN EXPERIMENTAL CONTROL SYSTEM
`
`We have carried out experimental verification of the operation
`of a feedback control system for RGB-LEDs with temperature
`feed forward and flux feedback. We now describe the experi-
`mental setup and present the results obtained.
`Fig. 5 shows a schematic of an RGB-LED white light source
`with this type of feedback control. The white light source is con-
`structed from four red LEDs, eight green LEDs, and four blue
`LEDs. The LEDs are mounted on a heat sink using thermally
`
`Fig. 5. Schematic of an RGB-LED lamp with feedback control system.
`
`conducting epoxy. A single temperature sensor (LM35 from Na-
`tional Semiconductors) is used to measure the temperature of
`the heat sink. The heatsink has a heater on it so that the tem-
`perature can be varied. A single Si photodiode (VTB113 from
`EG&G) is used to measure the light output from the red, green,
`and blue LEDs. The LEDs and the photodiode are mounted in-
`side an integrating sphere to provide ideal mixing of the light
`from the different wavelength LEDs, to ensure that the photo-
`diode sees all the LEDs equally, and to shield the experiment
`from ambient light. The integrating sphere is connected to a
`spectral lamp measurement system (spectrometer) to measure
`the chromaticity coordinates of the mixed white light. This is
`used to measure the performance of the feedback system. In an
`actual illumination system, the integrating sphere would be re-
`placed by color mixing optics suited to the application.
`Three independent flyback converters operating at a con-
`stant switching frequency of 100-kHz drive the RGB-LED
`light source. We used a PWM driving scheme operating at
`a frequency of 120 Hz for these experiments. Each flyback
`converter contains a current loop to maintain a constant peak
`current for the PWM pulses. In order to minimize the rise time
`and the fall time for the PWM current pulses, a small value
`of output filter capacitance is used. In addition, an inductor in
`series with the LEDs is used to reduce the current ripple. The
`color control system (shown in Fig. 4) is implemented in a DSP
`TMS320F240, which supplies the PWM turn-on and turn-off
`signals for the power supply.
`The photodiode measures the flux of each of the three LED
`wavelengths according to the scheme shown in Fig. 6. In this ex-
`ample, it is assumed that the duty ratio for green is largest, and
`for blue is smallest. The rise and fall times for the current pulses
`are assumed to be negligible. In Fig. 6, the start of the PWM cur-
`rent pulse for red is aligned at the start of the overall PWM pe-
`riod, the pulse for green is centered in the period, and the end of
`the pulse for blue is aligned at the end of the period. Although
`many other configurations are possible, the pulse positioning
`shown in Fig. 6 provides a complete measurement in a single
`PWM period. The light measurements are taken in a predefined
`sequence of four points during the PWM cycle. The individual
`flux components of the red, green, and blue LEDs are obtained
`by differential measurements as follows: Each photodiode mea-
`surement provides the flux from one or more colors plus any
`ambient light. Subtracting the fluxes at two of the measurement
`
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on August 04,2023 at 22:30:29 UTC from IEEE Xplore. Restrictions apply.
`
`4
`
`

`

`MUTHU et al.: RED, GREEN, AND BLUE LEDs FOR WHITE LIGHT ILLUMINATION
`
`337
`
`tribution at random? A statistical model has been developed to
`study this issue.
`An LED light source is constructed from six red, six green,
`and six blue LEDs. Each LED is characterized by a number
`of parameters including flux (lumens per amp), wavelength,
`spectral width, forward voltage
`, and temperature coeffi-
`cients of flux and wavelength. The values of these parameters
`are assigned at random based on a Gaussian distribution that
`approximates typical distributions of each parameter achieved
`in production. The LED spectrum is modeled as a second-order
`Lorentzian. All these parameters are based on a junction tem-
`perature of 25 C. The LED junction temperature at a given
`heatsink temperature is calculated from the power dissipation
`(the product of current and
`), and the thermal conductivity
`from the chip to the heatsink. We assume ideal optical mixing
`such that the white light output is a combination of all 18 LEDs
`in the lamp. The required drive currents for the red, green and
`blue LEDs to generate white light can be calculated based on
`the nominal performance of the LEDs (i.e., the mean of the
`Gaussian distributions for each of the LED performance param-
`eters).
`Once the LED performance parameters are selected, the
`actual color point of the lamp can be calculated. This can be
`compared with the designed color point, and a color error
`calculated. The color error will be dependent on the
`actual performance parameters selected, and those over a large
`number of LED lamps will have a statistical distribution. We
`typically calculate the color error for 5 000–10 000 lamps
`to determine this distribution. From the distribution of color
`errors, the product yield for a maximum acceptable color error
`can be calculated. The model can also calculate the effect
`of compensation and feedback schemes on the white light
`performance. The steady state function of a given control
`system is also included in the simulation together with a model
`for the sensors and LED drivers. The effect of AM or PWM
`driving scheme can also be modeled.
`A number of different control schemes have been modeled,
`providing the product yield as a function of color accuracy. The
`modeling results for three different control schemes are shown
`in Fig. 8. The simplest control scheme involves only tempera-
`ture feed-forward compensation (see Section IV). The simula-
`tion results show that less than 20% of products will have a color
`error of less than 0.005. It is clear that this control scheme will
`not achieve the performance required for illumination applica-
`tions. If the control scheme is extended to include flux feed-
`back of the red, green, and blue components (see Section V), a
`much improved product yield is achieved. As shown in Fig. 8,
`over 80% of products will have a color error of less than .005,
`and 100% yield is achieved with a color error of 0.01. We also
`show the result of a more complex feedback scheme that uses
`a wavelength feed-forward compensation scheme in addition to
`flux feedback. Color filters together with photodiodes are used
`to sense the wavelength shifts from the nominal value. This ap-
`proach can further improve product yield, giving 98% yield for
`a color error of only 0.005.
`The results of our simulations show that it is possible to de-
`sign feedback control systems for RGB-LEDs that are capable
`of producing the required color accuracy for illumination ap-
`
`Fig. 6. Light measurement using a single photosensor with the elimination of
`ambient light.
`
`Fig. 7. The measured color error as a function of heat sink temperature of an
`RGB-LED lamp both open loop, and with a control system using temperature
`feed forward and flux feedback.
`
`points gives the LED fluxes with the ambient light component
`eliminated. The red, green, and blue fluxes are given by the dif-
`ference between measurements three and four, two and one, and
`three and two, respectively.
`The experimental setup as described was used to examine
`the performance of this type of feedback scheme. The system
`was initially calibrated at a fixed temperature by adjusting
`the relative drive currents of the red, green, and blue LEDs
`until the spectrometer showed that the desired color point had
`been achieved. The white point was then monitored with the
`spectrometer as the temperature of the heatsink was slowly
`increased. The experimentally measured color shift for an
`open loop system is shown in Fig. 7. The result shows a color
`shift
`of about 0.005 for a temperature change of 10 C,
`comparable to the predicted shift shown in Fig. 3. Fig. 7 also
`shows the performance of the feedback control system with
`variation in temperature. The color point only changes by
`with a temperature change of 50 C. The change
`in lumen output over the same temperature change was found
`to be less than 3%. These results show that this type of control
`system can be used to produce a stable white light source from
`RGB-LEDs.
`
`VI. STATISTICAL ANALYSIS OF PRODUCT YIELD
`
`The experimental results discussed above have demonstrated
`the ability of the feedback system to maintain a precalibrated
`white point. However, what happens if no calibration is per-
`formed, and the set of LEDs is picked out of the production dis-
`
`Authorized licensed use limited to: Fish & Richardson PC. Downloaded on August 04,2023 at 22:30:29 UTC from IEEE Xplore. Restrictions apply.
`
`5
`
`

`

`338
`
`IEEE JOURNAL ON SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 8, NO. 2, MARCH/APRIL 2002
`
`variable color sources, such that the user can select the desired
`color point as well as the desired intensity from a single lamp.
`The success of this type of light source in the general illumi-
`nation market will depend on efficacy and cost. The rapid im-
`provement in LED efficiencies indicate that within the next few
`years, LED white light sources will be available that can meet
`the efficacy of current compact fluorescent lamps. The chal-
`lenge ahead is to reduce the cost of the LED lamp, including
`the LED chips and feedback control system.
`
`[2]
`
`REFERENCES
`[1] M. G. Craford, “LED’s challenge the incandescents,” IEEE Circuits and
`Devices Mag., vol. 8, pp. 24–29, Sept. 1992.
`, “Visible light emitting diode technology: High performance, more
`colors, and moving into incandescent lamp applications,” in Quantum
`Electronics and Laser Science Conf., 1996, pp. 28–28.
`[3] R. Haitz, “Another semiconductor revolution: This time it’s lighting,”
`in Proc. Ninth International Symposium on the Science & Technology of
`Light Sources, 2001, pp. 319–328.
`[4] C. Holen and G. Harbers, “LCD backlighting with high luminescent col-
`ored light emitting diodes,” in Proc. Ninth International Symposium on
`the Science & Technology of Light Sources, 2001, pp. 373–374.
`[5] G. Wyszecki and W. S. Stiles, Color Science. New York: Wiley, 1982.
`[6] A. Bergh, G. Craford, A. Duggal, and R. Haitz, “The promise and chal-
`lenge of solid-state lighting,” Physics Today, pp. 42–47, Dec. 2001.
`
`Subramanian Muthu (A’95) received a B.E. degree in electrical engineering
`from Bharathiyar University, India, the M.Tech. degree from the Indian Insti-
`tute of Technology, and the Ph.D. degree from the University of Victoria, B.C.,
`Canada, in 1985, 1992, and 1998, respectively.
`From 1992 to 1994, he was with Indian Telephone Industries. Since 1998, he
`is a senior member of the research staff at Philips Research, Briarcliff Manor,
`NY. His main research interests include digital control of power converters and
`systems, inverters, active filters, white light control of light-emitting diodes
`(LEDs), LED-based light generation, photovoltaic, and fuel cell power systems.
`
`Frank J. P. Schuurmans received the M.Sc. degree in physics from the Uni-
`versity of Utrecht, Utrecht, The Netherlands, and the Ph.D. degree in physics
`from the University of Amsterdam, Amsterdam, The Netherlands.
`In 2000, he joined Philips Research, Briarcliff Manor, NY, where he worked
`on LED illumination. In 2001, he joined Philips Research in Eindhoven, The
`Netherlands, as the project leader of the Extreme Ultra Violet Lithography Pro-
`gram of Philips Research.
`
`Michael D. Pashley received the B.Sc. degree in physics from the University
`of Bristol, Bristol, U.K., and the Ph.D. degree in physics from the University of
`Cambridge, Cambridge, U.K.
`From 1982 to 1985, he was a research fellow at the University of Cambridge.
`In 1985, he joined Philips Research, Briarcliff Manor, NY. While at Philips,
`he has worked in the areas of surface physics, semiconductor lasers, and LED
`illumination. He is now a research department head.
`
`Fig. 8. The calculated product yield as a function of maximum color error for
`three different feedback control systems.
`
`plications despite the large variability in LED characteristics.
`In the results presented in Fig. 8 we have assumed that each of
`the control measurements (temperature, flux, etc) are without
`errors. In practice, there will be some errors in the feedback sig-
`nals themselves arising from the characteristics and variability
`of the feedback sensors. These inaccuracies can also be mod-
`eled, and their effect on product yield determined. We find that
`it is important to keep such measurement errors to only a few
`percent. The performance of feedback control schemes can be
`improved with the addition of some limited factory calibrations.
`This can take the form of a measurement on the finished lamp,
`or by preselecting a smaller range of performance characteris-
`tics for the LEDs used to construct the lamp.
`
`VII. SUMMARY
`
`We have shown both experimentally and theoretically that
`practical white light sources can be produced from a combina-
`tion of red, green, and blue LEDs. The requirements of the illu-
`mination market for color accuracy are very stringent. Typically,
`the white point must be accurate to better than
`.
`Due to the variability in LED performance parameters, and the
`dependence of flux and wavelength on temperature, it is not
`possible to achieve the required color accuracy without an elec-
`tronic control system. A feedback control system using tempera-
`ture feed-forward compensation and flux feedback achieves the
`required level of color control and a relatively high product yield
`of over 80% for typical variation in LED characteristics. Further
`improvements can be made to the feedback scheme to give very
`high product yields of over 95%. Such control systems wi