Measuring, simulating and optimizing current LED phosphor systems to enhance the visual quality of lighting
© Bois et al.; licensee Springer. 2014
Received: 27 December 2013
Accepted: 20 March 2014
Published: 23 April 2014
Lighting engineering aspects of light emitting diodes (LEDs) with down converting phosphors (phosphor converted LEDs or PC-LEDs) were investigated by building sample PC-LEDs, measuring their spectral emission physically and simulating them with a software. This article describes the influence of the phosphors’ chemical, physical and technological properties on some well-established and widely used measures of visual lighting quality including luminous efficacy and the color rendering index. The dependence of these measures on YAG and LuAG phosphor particle size and activator concentration in the PC-LEDs is dealt with. Light sources of superior lighting quality can be obtained by mixing several phosphors. Hence, the mixture of a green and red phosphor system was also studied for a set of real PC-LED light sources and their simulations. Two green and two red phosphors are considered in this study. To predict and optimize the lighting quality of PC-LEDs with a reasonable accuracy, a usable input shall be provided for the simulation software assuming the understanding of the emission mechanisms. To validate our simulation results, the PC-LEDs’ measured spectral emission characteristics were compared with their simulated counterparts. The measurement and simulation of spectral power distributions contribute to the understanding of new phosphor mixtures. The present simulation method turned out to be usable to optimize the target parameters of the PC-LED (luminous efficacy, white tone quality and color quality of the illuminated colored objects).
Energy saving and environmental considerations have oriented the efforts of light source manufacturers towards alternative illumination solutions. The development of energy efficient lighting devices that also provide a high visual lighting quality for general lighting user constitutes an important challenge. In the last decades, research in the area of solid state lighting achieved considerable advances. LEDs proved to be an innovative and promising technology designed to revolutionize the lighting market. Today’s high-end LED light sources provide high luminous efficacy, high color quality, stability and long lifetime.
White light can be obtained by adding the emission of blue, green and red LEDs. The association of a pumping UV LED and phosphors is an alternative technique. The combination of powerful blue LEDs with broad emitting phosphors enables to create so-called phosphor-converted LEDs (PC-LEDs) of promising light output and quality at an affordable price. As the latter approach has gained momentum in industry, PC-LEDs were chosen to be investigated in this paper.
The modelling of PC-LED emission spectra requires considering the physical interactions between the constituting elements, the LED chip and the phosphor. The PC-LED’s emission spectra and their visual lighting quality properties are strongly influenced by phosphor characteristics, its chemical composition or synthesis method. In this article, phosphor particle size in narrow particle size distribution YAG and LuAG phosphors were first considered to optimize the PC-LED’s visual quality of lighting. The impact of YAG phosphor particle size on PC-LED light output[6–8] was reported in literature. However, these investigations were conducted on phosphors from different manufacturers hence it is difficult to compare them. Also, phosphor particle morphology in these previous studies was highly non-spherical.
The influence of phosphor particle size distribution (PSD) on the PC-LEDs’ properties made of YAG phosphor was discussed in detail. The present paper is intended to extend these considerations by measuring a set of dedicated PC-LEDs built from narrow phosphor particle size distributions and extracted from single synthesis batches of YAG and another garnet phosphor, LuAG. In the current study, a further chemical parameter, cerium activator concentration was also varied in the LuAG LEDs together with simultaneously varying particle size. The effect of these phosphor variations on the spectral power distributions of the PC-LED light source will be reported in this paper.
Then, four different phosphor mixtures of high visual lighting quality containing a green phosphor and a red-orange phosphor will be considered. Here again, sample PC-LEDs were built, measured and simulated. Accuracy limitations of the simulation will beanalyzed in terms of the differences of the measures of visual lighting quality between measurement and simulation.
Particle size distribution represents a significant physical parameter influencing the PC-LEDs’ lighting properties. Particle size distribution was measured with a single-particle optical sizing system (AccuSizer 780/SIS, PSS Nicomp). The influence of particle size on YAG and LuAG LEDs was studied. Phosphors were classified by particle size following a sieving process. Blue chip-on-board LED light sources were used as a basis to build the PC-LEDs. Blue LEDs emit with a peak wavelength at 441 nm and a luminous flux of 5 lm. LEDs are operated at constant forward current of 350 mA, with an active cooling system imposing 25°C. The phosphor was mixed with a transparent silicone binder (OE 6550, Dow Corning) and stirred to obtain homogeneous slurry. The latter still liquid was finally poured in a ring surrounding the blue LED. The volume to fill up with phosphor and silicone binder mixture is 40 mm3. After one hour curing at 150°C the silicone binder formed a solid layer enclosing phosphor particles resulting in the creation of conventional coated PC-LEDs. Precise weighting of slurry masses for each PC-LED ensures identical and reproducible phosphor layers and their luminescent properties in the PC-LEDs.
Absolute spectral power distributions were measured by the aid of an LED spectrometer (CAS 140, Instrument Systems) equipped with a 250 mm integrating sphere. It was possible to conclude about the influence of phosphor particle size on the spectral power distributions of our sample YAG and LuAG PC-LEDs. From the spectral power distributions, values of luminous efficiency and numeric descriptors of visual color quality of the light of the LED were calculated. The impact of the concentration of the Cerium activator was also studied for the case of our LuAG PC-LEDs as well as for the PC-LEDs with the green and red-orange phosphor combination.
Concerning our simulation method, the optical software LightTools® was used based on Mie calculation to simulate our sample PC-LEDs measured in the experiment. Spectral power distributions (emission spectra) of the sample PC-LEDs were simulated using the measured photoluminescent data of the phosphors as input data for the simulation software. Simulated and measured spectral power distributions, color quality parameters and luminous efficacy values were compared to validate the simulation and determine its accuracy. The relevance of simulation is that it offers the possibility to optimize the technological parameters of the PC-LED for high color quality and high light output for the human light source user with a reduced number of experiments.
Results and discussion
Photometric values and color quality characteristics of measured and simulated YAG PC-LEDs
As can be seen from Figure 4, luminous efficacy depends on phosphor particle size significantly. Luminous output decreases with a reduction in phosphor particle size possibly due to the fact that small particles scatter the light more strongly than large particles in every direction which results in backscattering. Backscattered radiation is absorbed in the blue LED chip which deteriorates luminous efficacy. Hence phosphor particle size is a determining factor of great impact on the light output of these PC-LEDs.
The difference between the simulated and measured emission spectra represented by the dotted line in Figure 5 enables one to identify at which wavelength discrepancies of simulation occur. Simulated and measured phosphor down conversion as well as the emission around 480 nm at the junction point in the spectrum between blue emission and phosphor exhibit a good fit: simulated and experimental curves overlap and the difference is small. The simulation differs from the measurements at and around the position of the blue peak. The spectral position of the blue peak wavelength itself is well predicted by the simulation. Differences arise concerning the shape of the blue emission peak. Simulations of the blue LED chip without phosphor perfectly reproduced the blue emission peak but simulations did not match any more when the phosphor was introduced into the program. The mismatch can be ascribed to the phosphor itself. The measured absorption data of the phosphor we implemented in the simulation as an input parameter may not have been appropriate: they might have been too high in the wavelength interval of the blue emission.
Turning back to Table 1, there are different ways to evaluate the difference between simulated and measured LED emission spectra. The extent of the error can be evaluated by the emission intensity difference for each wavelength (see Figure 5). Another way to evaluate this error is the computing of a standard measure widely used in lighting engineering which characterizes the perceived quality of white tones i.e. whether the white tone of the PC-LED can be considered as an acceptable or preferred white tone (e.g. without any disturbing greenish shade) when it is assessed visually be the human user. This measure is the white tone chromaticity difference Δu’v’wtq (the index “wtq” stands for white tone quality) between the chromaticity point of the PC-LED and its so-called reference light source in the CIE u’,v’ chromaticity diagram. The reference light source is a phase of daylight above 5000 K CCT or a blackbody radiator below 5000 K. The reference light source has the same CCT as the PC-LED. These Δu’v’wtq values are listed in Table 1.
Because of the not complete matching of the simulated and measured spectral power distributions, the two correlated color temperature (CCT) values do not match. Such differences (ΔCCT = 200 K - 500 K) are visually well noticeable. The general color rendering property of the PC-LEDs represented by the value of the CIE color rendering index (CIE CRI Ra) exhibits quite similar values for simulation and measurement. All these values (in the range Ra = 71-76) represent only a moderate color rendering property which are, in general, not suitable for general indoor lighting. R9 values of measurement and simulation are also similar in the sense that all of them (R9 = -3 to 26) represent visually unacceptable (bad) color rendering for red object colors. The reason of the R9 differences is that the simulation does not agree exactly with the measurement in the spectral range between 500 nm and 560 nm.
The calculated measure of the visual quality of the white tone represented by the value Δu'v'wtq is very similar for both cases (measured and simulated). Although white tone differences in the range Δu'v'wtq =0.001 - 0.003 is visually noticeable, this accuracy of the simulation is still very usable to be able to design a phosphor mixture of an intended white tone. Luminous flux differences are considerable, up to 15%. But as the tendency of the influence of phosphor particle size is well reproduced by simulation, the simulation method presented here is able to show the right way for a comprehensive PC-LED light source design (see Figure 1) including all technological and visual aspects.
The values of the CQS metric represent either the combined color rendering and color preference property of the light source (Qa is a general index and Q9 characterizes the visually important red colors) or a measure of color gamut, Qg, the so-called relative gamut area. The concept of color gamut means the extent of the set of all possible object colors that appear under the current light source (some light sources tend to suppress saturated object colors while other light sources are able to accentuate saturated object colors). For all CQS measures, there was a good agreement between measurement and simulation.
Several sample PC-LEDs of similar chromaticities u’ = 0.1722 ± 0.0008 and v’ = 0.4430 ± 0.0011 for LuAG #1 and u’ = 0.1562 ± 0.0008 and v’ = 0.4503 ± 0.0016 for LuAG #2 were built. These chromaticity points were chosen in order to be able to reach cool white LEDs of at a usable color rendering level at a later stage, in combination with a red-orange phosphor, see Section A phosphor system.
Figure 9 shows that luminous efficacy increases with increasing particle size. The reason is the same as for the YAG phosphor (see Section YAG): small particles exhibit more backscattering and the backscattered radiation is sent back and absorbed by the blue LED chip. Moreover, smaller particle surfaces are also more seriously affected by material defects in their microstructure. Latter effect is attributed to a stop in crystal formation. This effect results in increased reflectance for the case of smaller particles. Therefore, the luminous efficacy of the PC-LED decreases.
It can also be seen from Figure 9 that LuAG #2 (with 0.63% Ce) exhibits higher luminous efficacies than LuAG #1 (with 2.7% Ce), at least for smaller particle sizes. Chen reported the enhancement of crystal splitting by increasing activator concentration. Ce3+ relaxes by electron-phonon interaction to lower energy sites which leads to a red-shift emission and, consequently, a lower energy[13, 14].
The dotted line shows the difference between the simulated and the measured PC-LED emission spectra. Simulated and measured PC-LED spectra exhibit a good match from 520 nm on i.e. in the main spectral range of the phosphor part of the PC-LED’s emission curve. It should be noted that the phosphor’s photoluminescence data given as an input parameter in the software are of great influence and may enable a consistent simulation of down conversion emission. It should be emphasized at this point that the success or failure of simulation does not only depend on the software itself (LightTools® in the case of the present paper) but much on the way how it is used by the software user, especially how carefully the input data are prepared. The software itself is a good tool but it is up to the user to provide a good input for the software (see also below). Matching problems of the spectral radiant flux distributions are situated at about 430 nm and also at the transition point between the blue pump LED and the green phosphor at 470 nm.
At this spectral transition point, the simulated emission is too low. This is probably caused by an erroneous consideration of phosphor absorption characteristics in our measurement. We measured the photoluminescence properties of the phosphor as the input parameter of the simulation software in reflection mode on the surface of the phosphor sample. However, multiple reflections might possibly have occurred and changed the shape of the spectral absorption curve. A better way of absorption measurements would have been to use transmission mode to eliminate this possible error source. But this option was not possible due to the lack of such laboratory equipment in our research.
Radiant flux and chromaticity differences between measurement and simulation
LuAG 1 #1
LuAG 1 #2
LuAG 1 #3
LuAG 1 #4
LuAG 1 #5
LuAG 1 #6
LuAG 1 #7
As can be seen from Table 2, measured radiant flux and simulated radiant flux agree up to about 7%. The difference in the shape of the spectral emission curves implies discrepancies in the chromaticity point of the LuAG PC-LEDs up to Δu’v’=0.0047 (see the fourth column of Table 2) which is a visually well noticeable difference. In addition, a spectral difference measure designated by D was defined here as the sum of the spectral radiant flux differences between simulation and measurement at each wavelength. The computed D values demonstrate that the agreement between the simulated and the measured emission spectra varies among the different PC-LEDs (#1-#7) considerably.
As seen above, despite these differences, our simulation provided usable results concerning radiant flux. Accurate chromaticity simulation, however, remains an important challenge as the chromaticity point depends on the relative shape of the emission spectrum (measured or simulated) strongly. A spectral mismatch in the 440 nm - 470 nm range is decisive for the chromaticity match between simulation and measurement. As for the YAG simulations, the emission of the blue LED could not be simulated in a satisfying way. Phosphor absorption properties are suspected to cause this difference in the simulation of the blue part of the PC-LED’s emission spectrum. The tendency of the influence of phosphor particle size could be well simulated.
It should be mentioned that the LightTools® software is based on the theoretical assumption of spherical particles and this does not correspond to reality. Two other factors, the heat lost at the chip and the reflectance of the metal ring which contains the phosphor, are approximated only roughly. These approximations might also have caused the spectral deviations shown above.
A phosphor system
As can be seen from Figure 13, only one phosphor combination G2/R1 at 3100 K, 4300 K and 5200 K is able to approximate (Ra = 85) the Ra ≥ 87 criterion level above which the general color rendering can be considered good. This is due to the fact that the two red-orange phosphors (R1, R2) we used do not emit enough in the red – deep red spectral range (610 nm – 660 nm) hence they do not provide enough red spectral content to illuminate important reddish objects. The use of red – deep red emitting phosphors shall be considered in the future to obtain interior PC-LED light sources of high color quality. Note that, for a more comprehensive, up-to-date analysis of the color rendering – color quality issue, not only the general color rendering index but also a set of object specific color rendering indices shall be considered but this is out of the scope of this article. Also note that the CIE CRI color rendering index computation method is currently being modernized in the CIE. A new color rendering index has been developed but as it has not been officially endorsed by the CIE at the time of writing, it is not dealt with in this paper.
Warm white LEDs (2700 K, 3000 K) were obtained by adding a high amount of red-orange phosphor to the mixture. This fact leads to the higher red-orange peaks. The analysis of the photoluminescent properties of the phosphors (used as input data for the simulation) revealed that the green emission overlapped with the red-orange phosphor’s excitation and absorption ranges in the spectrum (see Figures 11 and12). Due to this interaction of the green and red-orange phosphor components, the amount of green phosphor still had to be high in order to achieve the right correlated color temperature (CCT) and a visually acceptable white tone without any disturbing color shade (e.g. a greenish tone which is not acceptable when the PC-LED is intended to be used for general lighting). Of course, both phosphors required excitation from the blue pump LED. Therefore, the warm white PC-LEDs exhibited a lower blue emission. In turn, when white PC-LEDs get a higher CCT (cool white, e.g. 6000 K) then less red-orange phosphor is necessary but there must be more green emission and more blue emission.
As can be seen from Figure 14, an increasing amount of green phosphor reduces the amount of blue emission because the blue rays are used to excite the green phosphor. The red-orange phosphor also absorbs the blue emission but it absorbs the green emission, as well. To increase red-orange emission intensity by keeping green emission intensity at the same level, red-orange phosphor concentration must be increased but green phosphor concentration must also be increased. This fact considerably decreases blue light output. The point is that this behavior could be simulated successfully, as can be seen from the comparison of the measured and simulated spectral emission characteristics of Figure 14. For this, here again, appropriate phosphor input photoluminescent data had to be provided for the simulation software. A similar fit between the simulated and experimental spectral emission curves was obtained for the other phosphor mixtures (G1/R2, G2/R1, G2/R2) as well. Proportions among the peak blue, peak green and peak red-orange spectral radiant flux values agreed well between the simulations and the measurements across the different CCTs.
Differences of radiant flux (ΔΦ e ), luminous flux (ΔΦ v ), correlated color temperature (ΔCCT), chromaticity (Δu’v’) and general color rendering index (Δ R a ) between the simulated and the measured spectral emission curves for the four phosphor mixtures
Δ CRI Ra
Δ Φ v /lm
CRI, R a,meas
Δ CRI Ra
Δ u'v' wtq
Δ u'v' m
Δ u'v' s
Δ Φ e /%
Δ Φ e /W
CRI, R a,meas
Δ CRI Ra
CRI, R a,meas
Δ CRI Ra
Δ u'v' s
Δ Φ e /%
Δ Φ e /W
As can be seen from Table 3, measured and simulated luminous flux and CRI values agree reasonably. The measured white tone characteristics (Δu’v’wtq,meas) remain less than 5 · 10-3 (mean: 2,1 · 10-3) which confirms that the sample PC-LEDs we built yielded reasonable white tones. Nevertheless, some white tone discrepancies between simulation and measurements exist. The measure of white tone quality of the simulated white PC-LED spectra (Δu’v’wtq,sim) reached values up to 1.23 · 10-2 (mean: 5,2 · 10-3). The maximum value (Δu’v’wtq,sim = 1.23 · 10-2) corresponds to a visually not acceptable white tone. Simulated chromaticity points also deviate from measured ones (see the Δu'v'ms values). As a consequence, measured and simulated CCT values differ, as well and this difference is visually well noticeable.
All these deviations result from the differences of the measured and simulated spectral power distributions. In Figure14, it is possible to identify at which wavelengths the spectral power differences are the highest (for the example G1/R1). Absolute and relative radiant flux differences between the simulated and measured emission spectra (ΔΦe) vary from 1% to 12% (see Table 3). These differences are ascribed to the simulated emission discrepancies occurring around the blue peak emission (420 nm - 440 nm), in the spectral transition range between the blue LED and the green phosphor emission (460 nm - 500 nm) and in the transition between the green and red-orange phosphor emissions (550 nm - 600 nm).
The reason of the spectral mismatch between simulation and measurement can be attributed to the discrepancies of phosphor absorption data (our input parameter for the simulation program). The deviation around the blue emission peak and at the transition range about 470 nm occurs due to similar reasons as discussed for YAG and LuAG in the previous Sections YAG and LuAg. The PC-LED configuration considered in this section is represented by a complex system of two phosphors in which the red-orange phosphor’s photoluminescent properties influence the emission spectra of the green phosphors. Our absorption measurements on the phosphors to provide the input parameter for the simulation program influence the multiple phosphor interactions within these PC-LED made of the mixture of two phosphor components very strongly. In this case, down converting processes are influenced by the interaction of both phosphors resulting in the difficulty to simulate these emission spectra.
In this article, LED light sources using down converting phosphors (phosphor converted LEDs or PC-LEDs) were studied. The influence of the phosphors’ chemical, physical and technological properties on the spectral power distributions and the visual lighting quality parameters of the PC-LEDs were investigated by constructing sample PC-LEDs and by simulating them in LightTools®. The influence of YAG and LuAG phosphor particle size and activator concentration on the PC-LEDs lighting quality was dealt with. Further sample PC-LEDs with the mixture of two phosphors of a four-phosphor system were built and luminous efficacy and color quality measures were computed. These PC-LED properties were strongly influenced by the photoluminescent data of the phosphors as input parameters for the simulation program. Simulation results based on the measured photoluminescent input parameters of the phosphors and the measurement results of the sample PC-LEDs were compared. The simulation method was corroborated to be a usable and important tool to optimize the technological parameters of the PC-LEDs for general lighting to obtain high luminous efficacy and high color quality.
The method presented in this paper is an interdisciplinary and comprehensive method with the following aspects: 1. photoluminescent properties of the phosphors were measured and provided as input data for LightTools®; 2. sample PC-LEDs were built; 3. their absolute spectral power distributions were measured and simulated based on the input data; 4. lighting quality measures (luminous efficacy, white tone quality, color rendering index, etc.) were computed for both the measured and the simulated spectra; 5. measured and simulated results were compared and simulation accuracy was determined. It turned out that this simulation method is usable to optimize the visual quality measures of PC-LED light sources for general lighting purposes because the simulation method is able to predict the tendencies of spectral power distribution changes when the technological parameters of the phosphors are changed systematically.
Concerning the analysis of the possible causes of the deviations between measurement and simulation, one important issue is that the simulation is based on the assumption of spherical phosphor particles which does not hold in reality. Nevertheless, the tendencies of the PC-LEDs’ lighting properties by varying phosphor size and also with varying activator concentration could be simulated with reasonable accuracy both for YAG and for LuAG, e.g. increasing phosphor particle size resulted in higher luminous efficacy both in measurement and in simulation.
The last section described certain types of green and red-orange phosphor mixtures. The measurement and simulation of these spectral power distributions contribute to the understanding of new phosphor mixtures. Results provided by the simulation strongly depend on the phosphor absorption provided as an input parameter for the simulation software. Simulated luminous flux and radiant flux values reproduced the measured values in a satisfying way. Discrepancies in the relative shapes between measured and simulated emission spectra were found and attributed to multiple reflections within the phosphor layer. The interaction between the phosphors in PC-LEDs with phosphor mixtures was identified based on the emission spectra. The present simulation method (Figure 1.) turned out to be usable to optimize the target parameters of the PC-LED (luminous efficacy, white tone quality and color quality of the illuminated colored objects). A successful simulation allows scientists and engineers to save time when building optimized PC-LED light sources for research and development.
This work is supported by Technische Universität Darmstadt and Merck KGaA.
- Zukauskas A, Shur M, Gaska R: Introduction to solid state lighting. New York: John Wiley & Sons; 2002.Google Scholar
- Schubert EF, Kim JK, Luo H, Xi JQ: Solid-state lighting—a benevolent technology. Rep Prog Phys 2006, 12: 69.Google Scholar
- Chang MH, Das D, Varde PV, Pecht M: Light emitting diodes reliability review. Microelectron Reliab 2012, 5: 52.Google Scholar
- Sheu JK, Chang SJ, Kuo CH, Su YK, Wu L, Lin YC: White-light emission from near UV InGaN-GaN LED chip precoated with blue/green/red phosphors. Photonics Technol Let, IEEE 2003, 1: 15.Google Scholar
- Ye S, Xiao F, Pan YX, Ma YY, Zhang QY: Phosphors in phosphor-converted white light-emitting diodes: recent advances in materials, techniques and properties. Mat Sci Eng: R: Rep 2010, 1: 71.Google Scholar
- Chen L, Chu CI, Liu RS: Improvement of emission efficiency and color rendering of high-power LED by controlling size of phosphor particles and utilization of different phosphors. Microelectron Reliab 2012, 5: 52.Google Scholar
- You JP, Shi FG: Effect of phosphor particle size on luminous efficacy of phosphor-converted white LED. J Lightwave Technol 2009, 22: 27.Google Scholar
- Huang SC, Wu JK, Hsu WJ, Chang HH, Hung HY, Lin CL, Liu RS: Particle size effect on the packaging performance of YAG: Ce phosphors in white LEDs. Int J Appl Ceram Technol 2009, 4: 6.Google Scholar
- Bois C, Bodrogi P, Khanh TQ, Winkler H: White LED light characteristics as a function of phosphor particle size. ECS J Solid State Sci Technol 2012, 5: 1.Google Scholar
- Bodrogi PZ: Semantic interpretation of the colour binning of white and coloured LEDs for automotive lighting products. In 10th International Symposium on Automotive Lighting, September 2013. Edited by: Herbert U. Darmstadt: Verlag; 2013:679.Google Scholar
- Davis W, Ohno Y: Color quality scale. Opt Eng 2010, 49: 3.View ArticleGoogle Scholar
- Chen L, Lin CC, Yeh CW, Liu RS: Light converting inorganic phosphors for white light-emitting diodes. Mater 2010, 3: 3.Google Scholar
- Setlur AA, Srivastava AM: On the relationship between emission color and Ce 3+ concentration in garnet phosphors. Opt Mater 2007, 12: 29.Google Scholar
- Lin CC, Liu RS: Advances in phosphors for light-emitting diodes. J Phys Chem Lett 2011, 11: 2.Google Scholar
- Khanh TQ, Bodrogi P, Vinh TQ, Brückner S: Farbwiedergabe von konventionellen und Halbleiter-Lichtquellen - Theorie. München: Bewertung, Praxis, Pflaum Verlag; 2013.Google Scholar
- Smet KAG, Schanda J, Whitehead L, Luo MR: CRI2012: a proposal for updating the CIE color rendering index. Lighting Res Technol 2013, 45: 689–709. 10.1177/1477153513481375View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.