LIGHTING DESIGN & APPLICATION Spectral vision of a colour scene is determined by the product of the spectral power distribution of the light source, the spectral radiance (reflection) factor of the scene components and the photopic spectral visibility or V() curve. A luminance model for evaluating colour rendering qualities of light sources With the introduction of LED lighting, the lighting industry is faced with an infinite number of colour sources of light with numerous SPDs. The summation of SPDs can create even more diverse sources of light through combinations of these LED sources. The colour appearances and colour rendering capabilities of these sources are currently an important topic of debate for scientists, lighting engineers, vision scientists and architects. The need for improved but simple metrics to define colour appearance and colour rendering capabilities of light sources is tremendous. The current metrics mostly used in the lighting fraternity is the CIE definitions of correlated colour temperature (CCT) and general colour rendering index (Ra). The focus of this research is on colour rendering metrics for light sources and not colour appearance. There is a need for a simple, single-number rating value for basic colour rendering, which shows good correlation with viewer agreement. We investigated a method for evaluating colour rendering by determining (reflected) luminance from 12 CIE colour samples. The incandescent/halogen light source was used as reference and an off-the-shelf LED lamp of the same CCT was used as test source. The main deficiency of the incandescent/halogen source as reference relates to the specific SPD of this source with very low output in the blue range of wavelengths and very strong radiation in the red wavelengths. The photopic vision curve or V() curve by Prof. FW Leuschner and JGJ van der Westhuizen, University of Pretoria adds to the very low reflected luminance from a blue colour sample (CIE colour sample 12 – “Strong Blue”) when illuminated by an incandescent lamp. This is of course accentuated during scotopic or mesopic vision at low light levels. The reflected luminance from the two lamp samples was compared and the results produced a strong confirmation of some of the deficiencies of the general colour rendering index (CRI). Introduction At present, all lamps are characterised by a number of electrical, photometric, colorimetric, life expectancy, dimensional and cost (capital and operating cost) characteristics. Colour appearance and colour rendering are the most important factors in evaluating the colorimetric quality of a lamp for acceptance and comparison with other lamps, especially in the case of retrofitting existing light installations for electrical energy saving or general upgrading. The colour qualities are mainly the colour appearance and the colour rendering metrics, usually specified by the CIE correlated colour temperature (CCT) and the general CRI (Ra). Neither of these two quantities are very realistic and they are definitely not accurate metrics for all light sources. Unless the chromaticity coordinates of two light sources are identical, they will not appear the same and, unless the lamp under consideration’s chromaticity June 2013 - Vector - Page 24 coordinates are on the Planckian curve, the CCT can be very misleading. The CIE Ra value is limited to eight nonsaturated CIE colour samples (R1 to R8) and the reflection from these samples is compared to that obtained from a reference source, which is a Planckian radiator of the same CCT (approximated by an incandescent or halogen lamp). Clearly, these eight samples cannot represent most or all of the colours in visual environments. Neither can a Planckian radiator (incandescent) light source be viewed as a good reference source as the light produced is very low in the short (blue) wavelengths and very high/dominant in the long (red) wavelengths. There are more than ten different proposals for evaluating colour rendering. These concentrate on colour fidelity, colour preference, etc. It is felt that a single number cannot fully cover all these components of colour rendering. It is, however, also accepted that the mathematical calculations cannot dominate the visual appraisal by samples of people from different groups, as the psychophysical variables are too diverse to capture all colour “rendering” attributes in one or more mathematical calculations. Comparison of spectral luminous flux for halogen, LED PAR16 GU10 We started our experiments by comparing two lamps (one incandescent/halogen and one LED lamp) of the same CCT and the same reference luminance on a “white standard reflector”, but different Ra values of 99 and 83 respectively. Incandescent/halogen lamp: 240 V 50 Hz 50 W PAR16 lamp (GU10 base), beam angle ≈ 36°. LED lamp: 230 V 50 Hz 5 W (GU10 base), beam angle ≈ 36°. A KonicaMinolta CS-2000 spectroradiometer was used for all spectral measurements. The following was determined for both lamps: Spectral power distribution from 380 to 780 nm for each lamp. This was used to calculate the following: Colour rendering indices from R 1 to R 12 , general colour rendering index R a , correlated colour temperature (CCT), x and y chromaticity coordinates, dominant wavelength and relative luminance. The results are shown in Table 1. Most colour quantities are similar, except for the colour rendering metrics: Ra is 16% lower for the LED and R9 (“strong red”) 81% lower for the LED. Comparing the spectral luminous flux distributions of the two lamps gives another picture. A comparison of Figs. 1 and 2 clearly indicates a very similar spectral luminous flux distribution at around 575 nm, with the halogen lamp stronger above 650 nm and weaker below 450 nm, but no dramatic differences are evident. Both graphs seem to be shifted to the right of 555 nm and both are stronger in the long wavelengths than in the short wavelengths. Reflected spectral luminous flux, 12 CIE colour samples The authors propose that a simple weighted luminance comparison of all 12 colour samples be used to determine a colour rendering value for any light source. These include R 1 to R 8 for calculating the Ra value, plus R9 to R12 for bright “strong” red, yellow, green and blue respectively. One can include R 13 “Caucasian skin” and R 14 “leaf green”. R15 is a non-CIE colour sample for “Asian skin” colour and a possible “R16” for “African skin” colour can be added. The following equation was used to calculate the relative reflected luminous flux of twelve colour samples (R1 to R12) when illuminated by the two light sources under test: ଼ Iோ ൌ ͺ͵ න I ሺOሻUோ ሺOሻܸO ሺOሻ݀O ଷ଼ Where: Ri is the reflected luminous flux from the specific colour sample i =1 to 14. Fig. 1: Spectral luminous flux distribution for 50 W halogen lamp. Fig. 2: Spectral luminous flux distribution for 5 W LED lamp. Quantity 50 W Halogen lamp 5 W LED lamp Relative Luminance Lv 63,11 62,53 Correlated colour temperature CCT 2652 2651 Ra 99 83 R1 99 81 R2 99 90 R3 99 98 R4 99 80 R5 99 81 R6 99 88 R7 99 85 R8 99 63 R9 98 19 R10 98 78 R11 99 78 R12 97 74 x 0,4664 0,4663 y 0,4148 0,4145 Table 1: Colorimetric characteristics of halogen and LED lamps. V() is the photopic eye response curve. R() is the spectral reflectance of the specific colour sample. ej() is the spectral radiant flux from the specific light source for j = 1 or 2. The top section of Table 2 compares the reflected luminous flux from all eight CIE colour samples making up the general colour rendering index Ra. The lower section of the table compares the four highly saturated (“strong”) CIE colour samples 9 to 12 for red, yellow, green and blue respectively. Eight colour samples R1 to R8 The luminous flux reflected from the June 2013 - Vector - Page 25 eight colour samples (R1 to R8) can be assumed to be identical as the average deviation is 2% and the highest is 4%, considering the variation in source luminance of 1%. So why should the colour rendering of the LED be 16% lower than that of the incandescent lamp? It is also important to note that the values of reflected luminous flux are all in a band from 16,4 to 21,6 (a.u.) – low variation and all these reflected colours will appear to be almost equally “bright”. Four colour samples R9 to R12 Note that these four colour samples do not form part of the general colour rendering index Ra. Colour sample 50 W Incandscent 5 W LED % Difference Colour sample appearance TCSO1 20,9 20,8 0 Light greyish red TCSO2 19,4 19,4 0 Dark greyish yellow TCS03 19,2 19,2 0 Strong yellow green TCS04 16,8 16,5 2 Moderate yellowish gren TCS05 17,5 17,2 2 Light bluish green TCSO6 17,0 16,4 3 Light blue TCS07 18,9 18,4 3 Light voilet TCS08 21,6 20,8 4 Light reddish purple Average 18,9 18,6 2 Ra 99,0 83,0 16 TCS09 11,0 9,9 10 Strong red TSC010 40,6 40,7 0 Strong yellow TSCO11 10,9 10,4 4 Strong green TSC012 2,7 2,4 11 Strong blue Average 30,5 27,5 10 Strong colours Table 2: Comparing reflected luminous flux (p.u.) from twelve CIE colour samples when illuminated by two lamps. The average luminous flux reflected from the four colour samples (R9 to R12) is 10% higher for the halogen lamp than for the LED lamp, with only the “strong yellow” colour sample showing a slightly higher value for the LED than for the halogen lamp. The deviations are, however, still not significant with red and blue 10% or more. From these numbers one can conclude that the saturated colour sample representation is better for the halogen lamp than for the LED lamp by approximately 10%. It is important that the values of reflected luminous flux for red and blue colour samples are much smaller than for yellow and green, for both lamps, i.e. around 40 (a.u.) for yellow and around 2 (a.u.) for blue. To understand the reasons for this we have to look at two main parameters: The radiant flux radiated by the source at the peak spectral reflectance values of the specific blue and red colour samples (R9 and R12 respectively). The spectral photopic eye response values at the same wavelengths as mentioned here. The blue colour sample shows a peak at 408 nm and the photopic V(408) value is 9,4 x 10-4. The halogen and LED lamps show steep drops in radiant flux below 420 nm, with the LED peaking at about 450 nm. This is usually suppressed through the phosphor to produce more yellow/red light in the LED, for a higher Ra value. How to improve colour rendering Fig. 3: Spectral reflectance for four strong colours, red, yellow, green and blue. June 2013 - Vector - Page 26 Fig. 3 shows the spectral reflectance characteristics of the four saturated (“strong”) CIE colour samples red, yellow, green and blue. It is clear that there are dramatic differences which must be investigated further. Once we multiply the spectral reflectance of the colour samples with the spectral photopic eye responsivity values, we must understand the reflected spectral luminous flux for different light source SPDs. Fig. 4 shows the effective spectral ranges for the four colours and the dramatic differences in producing reflected luminous flux (proportional to the area under each curve). To match these spectral values to produce the same luminous flux from each colour sample, the different colour sample SPDs must be multiplied by the following factors (using yellow as the reference): optimal colour mixing ratios to establish constant luminous flux from each reflector sample: red = 4,82, yellow = 1,00, green = 3,04 and blue = 9,98. Fig. 5 shows a clear increase in average spectral reflectance factors for the 14 CIE colour samples, from about 0,08 at 360 nm to 0,54 at 830 nm. To counter this tendency we require a source SPD which should at least be the inverse of the graph in Fig. 5 i.e. “high” at the short wavelengths (blue) and “low” at the long wavelengths (red). The incandescent is the complete opposite and even worse than the graph in Fig. 5. This proves that the incandescent/ halogen lamp cannot be rated as 100% CRI for all colour rendering indices R1 to R15 and especially not the general colour rendering Ra to be 100%. Individual colour LEDs, however, can be matched to produce virtually any SPD for combinations of three or more LEDs. The weighting factors listed here can be included as well as compensation for the V() curve at short and long wavelengths. Such weighted reference SPDs can then be accompanied by categories of luminous efficacy of radiation to produce a lamp comparison matrix. Conclusion Luminance level reproduction of the CIE colour samples R1 to R12 were used to confirm one of the main deficiencies of the general colour rendering index Ra which uses an incandescent/halogen source (of the same CCT) as the reference lamp. The combined signal of light source SPD, radiance (reflectance) factor of the colour samples and the Fig. 4. Spectral reflectance for four strong colours, red, yellow, green and blue. Fig. 5: Average spectral reflectance for 14 CIE colour samples. photopic eye responsivity curve show a clear weakness in short wavelength (“blue”) light and even in the long wavelengths (“red”), even for the incandescent/halogen lamp itself. Combinations of LEDs can be used to compensate for these discrepancies. Improved colour rendering will, however, be accompanied by reduced luminous efficacy of radiation. There will always be a trade-off between the two metrics which will be determined for each specific application. Note This article is based on a paper presented at the ninth IESSA Conference & AGM, and is reproduced here with permission. June 2013 - Vector - Page 28 References [1] CIE 13.3: "Method of measuring and specifying colour rendering properties of light sources," 1995 Vienna, Austria: Commission International de I’Éclairage. [2] Illuminating Engineering Society: The lighting handbook: tenth edition: reference and application, 2011, Illuminating Engineering Society of North America, ISBN 978-087995-241-9. [3] W Davis and Y Ohno: “Toward an improved color rendering metric”, Fifth international conference on solid state lighting. [4] V Viliuna, H Vaitkevicius, R Stanikunas, A Svegzda and Z Bliznikas: “LED-based metameric light sources: rendering the colours of objects and other colour quality criteria, Lighting research and technology, 43, Aug. 2011, p. 321 – 330. Contact Prof. FW Leuschner, University of Pretoria, Tel 012 420-2283, leuschner@up.ac.za