1. Introduction

An important task of colorimetry is to properly describe our perception of light sources. Today, most applied colorimetry employs the 1931 color-matching functions (CMFs) [1]. Despite their historical importance, these are known to be poorly predictive of perception. Further, they pertain to an idealized observer, whereas perception can vary strongly from person to person [2]. From a physiological standpoint, this inter-observer variability is underpinned by two effects. The first is the known variation between individuals of the three cone responses [3,4], and the second is the age-dependent decrease of the transmission of the ocular medium, which is most pronounced at short wavelengths [5–7].

The implications of this variability have been the focus of recent interest in color science. In 2006, the Commission Internationale de l’Eclairage (CIE) released new cone fundamentals that account for the effect of observer age and viewing conditions [8], which in turn spurred multiple studies. [9] provided a theoretical analysis of the impact of these CMFs and predicted significant inter-user differences. [10,11] reported large metamerism breakdown effects in narrow-band displays which led to a proposal to generate individual-specific CMFs for such applications [12]. The impact of viewer variability in lighting color rendition metrics was explored in [13], here again with large effects.

These compelling academic investigations have led to limited real-world implementation so far. When designing commercial light sources, the practical implications of this variability were ignored for a long time, in part because conventional light sources, such as filament lamps, have predetermined smooth spectra which mitigate its impact. However, this needs to be revisited with the spread of LED-based emitters having ever-sharper spectral features –especially at short wavelengths– and offering more opportunities for spectral engineering. Namely, one may wonder whether a transition to more accurate colorimetry tools, suited for such emitters, is warranted.

In this article, we investigate practical implications of user variability – in particular, those related to short-wavelength effects. We first consider light emitters having a structured spectral power distribution (SPD), which reduces their circadian stimulation. We present experimental results of human perception and show that the signal is driven by age and SPD shape, in excellent agreement with a suitable model. We then discuss broader implications for LED emitters, both in lighting and in displays, where we predict large effects. We conclude that the use of modern, age-dependent CMFs is desirable to improve chromatic accuracy.

2. Perceptual chromaticity of blue-free spectra

In this section, we discuss the perceived chromaticity of a series of ‘blue-free’ (BF) SPDs produced by a violet-pump LED exciting a green and a red phosphor. These emitters have unique beneficial properties in lighting applications seeking to affect the circadian cycle. The rationale for their development is discussed in Appendix A.

Variations between various CMF sets, and the effect of age on CMFs, are both most pronounced at short wavelength. BF SPDs have no blue radiation and only rely on violet emission to achieve a white chromaticity. Therefore, by exacerbating short-wavelength effects, they provide an ideal testbed for experimental verification of chromaticity models

2.1 Experiment

We seek to study the differing predictions offered by various CMFs. We use the following notations: $31-2^{\circ }$ indicates the well-known 1931 CIE $2^{\circ }$ CMFs, which still forms the basis of most applied colorimetry; $06-2^{\circ }$ and $06-10^{\circ }$ indicate CMFs derived from the physiologically-relevant ‘CIEPO06’ $LMS$ cone fundamentals recently released by the CIE with $2^{\circ }$ and $10^{\circ }$ fields, respectively. These take into account the observer age, and are derived as follows. The cone fundamentals are obtained according to [8], which accounts for the age-dependent absorption of the ocular medium. They are then transformed into $XYZ$ CMFs following the procedure developed in CIE Technical Committee 1-97 [14], which provides an LMS-to-XYZ transformation by generalizing the original procedure for deriving the XYZ CMFs (as discussed in Ref. [15]). Note that this improved, age-dependent $LMS$-to-$XYZ$ transform avoids artifacts such as negative values of $x$, which would otherwise occur if an age-independent transform was applied, as has been done in other works. Finally, the $XYZ$ coordinates are transformed to ($u'v'$) values, since this chromaticity diagram offers improved color uniformity.

Small color differences are often measured in terms of MacAdam ellipses; a one-step MacAdam ellipse has a center-to-edge dimension corresponding to the standard deviation for just-noticeable color difference perception in MacAdam’s original experiment, and is roughly indicative of a just-noticeable color difference between two stimuli. [16] MacAdam ellipses near the Planckian locus are well approximated by circles in $(u'v')$[17] (with a radius of 0.0011 corresponding to one MacAdam step). Therefore, in the following, we use such circles as a proxy for MacAdam ellipses, and refer to them as MacAdam circles for simplicity.

Perceptually, when two SPDs with differing chromaticities perpendicular to the Planckian locus are compared, the one with higher $v'$ is often described by observers as ‘greener/yellower’ and the one with lower $v'$ as ‘pinker’. Likewise, when two SPDs with differing chromaticities along the Planckian are compared, the one with higher $u'$ is often described as ‘warmer’ while the one with lower $u'$ is termed ‘cooler’ – in accordance with their respective correlated color temperatures (CCTs). Figure 1(a) summarizes these trends.

 

Fig. 1. Perception experiment. (a) $(u' v')$ diagram. Black line: Planckian locus, with dots showing color temperatures of 2,500 to 3,000 K and black circles showing 3-step and 6-step MacAdam circles. Dots: SPD chromaticities, as labeled. All chromaticities are according to $06-10^{\circ}$ CMFs (at age 30). (b) Reference and BF SPDs used in the experiment. (c) Digital camera picture of the four BF spots projected on a white wall. Here, the camera selected BF3 to define the white balance of the image, therefore showing the other spots with greener or pinker tint. This rendition is not necessarily representative of an observer's perception. (d) Layout of the spots during the experiment. The four BF SPDs are projected in turn.

Download Full Size | PDF

For the perception study, we designed four BF SPDs (labeled BF1-4) with a CCT of $\sim$2,700 K and with chromaticities varying from pinker to greener (i.e. from farther below to farther above the blackbody locus). We further designed three additional ‘full-spectrum’ SPDs [18,19] to be used as comparison points for evaluating the BF SPDs. These reference SPDs (labeled R0, R+, R-) are targeted on, above, and below Planckian, respectively, with $u'v'$ distances from the Planckian of about $\pm 0.007$ – therefore, their chromaticities are respectively at the center and edges of a 6-MacAdam circle. The SPDs are shown in Fig. 1(b), and their chromaticities according to various CMFs are shown in Table 1 and in Fig. 1(a). Within each family of SPDs, changes in chromaticity are achieved by varying the phosphor composition (but keeping the pump LED wavelength and phosphor types constant); for the BF SPDs, this translates into a varying magnitude of the violet peak. Note that, although the BF SPDs were intended to have the same CCT and vary only along the pink-green axis (perpendicular to the Planckian locus), they also show slight variations in CCT due to experimental inaccuracy in these research prototypes.

Table 1. Properties of the SPDs used in the Perception Experiment. The $06-10^{\circ }$ Values are for Age 30.

View Table | View all tables in this article

For each SPD, we fabricated a directional lamp with an MR-16 form factor, an input power of 7.5 W, and a beam angle of $25^{\circ }$. A picture of the four BF spots is shown in Fig. 1(c).

The reference SPDs are smooth and contain both violet and blue radiation, whereas the BF SPDs only rely on violet radiation for balancing chromaticity. Therefore, we expect that any short-wavelength effect will be much more pronounced when observing the BF SPDs than when observing the reference SPDs. We thus make the simplifying hypothesis that the perception of the reference SPDs is approximately the same for all observers, whereas the perception of a BF SPD relative to the reference SPD may vary markedly between observers. Thus, the reference SPDs constitute comparison points (respectively green, centered, and pink) which form a scale against which the chromaticity of the BF SPDs can be evaluated.

The experimental protocol was as follows: a wall of a dark room was covered with a high-reflectivity white diffuser sheet (spectrally flat in the range 400-700 nm), on which lamp spots were projected, each with a diameter of about 50 cm. Participants were introduced in the room in groups of 2-4. When they entered the room, the three reference lamps were visible side-by-side (from left to right: R+, R0, R-). Participants were instructed as follows:

«You can see three reference spots, each with a different tint of light. The one on the left may be called greener, the center one may be called neutral, and the one on the right may be called pinker. You will be presented with several other spots and asked to rank their colors relative to these reference spots. The reference spots provide a scale for ranking. The left/greener reference has a score of minus 4, the center/neutral reference has a score of zero, and the right/pinker reference has a score of plus 4. You will be able to rank the other spots on a scale from minus 6 to plus 6. For instance, if a spot appears exactly as green as the left reference, give it a score of minus 4. If a spot has a tint midway between those of the left and center references, give it a score of minus 2. If a spot has a tint between those of the left and center references but is closer to the left reference, give it a score of minus 3. If a spot has a tint somewhat greener than that of the left reference, give it a score of minus 5. If a spot has a tint much greener than that of the left reference, give it a score of minus 6.»

Participants were standing in front of the illuminated wall, at a distance of about 1-2 meters. They were free to move around, but could not discuss with other participants. The viewing conditions corresponded to a large field of view (about $10^{\circ }$, depending on the participant’s position). The four BF SPDs were shown in turn, each for about 30 seconds. The layout of the spots is shown in Fig. 1(d).

The sixty-four subjects were employees of Soraa Inc., who volunteered to participate. They had general knowledge of lighting products but no specific knowledge of colorimetry, and were unaware of the objective of the experiment. Their answers were recorded on an anonymous questionnaire, as were their age, ethnicity, and gender. Participants were generally able to assess light tint and fill in the questionnaire without difficulty given the instructions. The participants’ demographic details are included in Appendix B.

2.2 Results and interpretation

The details of the participants’ responses are shown in Appendix B. These data show systematic general trends together with significant individual variations. Participants generally agree on the relative green/pink ranking of the four BF SPDs, in agreement with qualitative expectations given the SPDs’ chromaticities. On the other hand, significant variations appear in the absolute scores, with a standard deviation of about three points for each SPD – a large value in view of the total score range of −6 to +6. Note that, given the difference in chromaticity between the reference lamps, a score difference of 1 point corresponds to approximately 1.5 MacAdam unit.

Further inspection of the data reveals that age is strongly correlated to this variation. This is shown in Fig. 2, where the statistics of lamp score are shown versus age group (binned in ten-year spans). A systematic trend is observed, with the perceived chromaticity becoming greener with age. This trend is very pronounced, with some lamp scores shifting by 9-10 points (i.e. nearly the full scale) from the youngest to the oldest age group. Note that some cases saturate the ranking – for instance, many younger participants rank BF1 at +6 and many older participants rank BF4 at −6, suggesting they may have given more extreme scores had a wider scale been available. Further, there is still individual variation in each group, with the two-quartile width of the box plots typically spanning 2-4 points.

 

Fig. 2. Experiment results and model. (a) Chromaticity score versus viewer age for the four BF SPDs. Boxes: experimental data. Line: model prediction. (b-c) Model details, for ages 20 and 40 respectively. Dots: SPDs as labeled. Red arrow: orthogonal projection of BF3's chromaticity along the axis defined by the reference SPDs (shown as a dashed line), from which the modeled score is derived. Black line: Planckian locus. Black circles: 3- and 6-step MacAdam circles.

Download Full Size | PDF

To interpret these data, we analyze the lamps’ SPDs with the $06-10^{\circ }$ CMFs to derive a model for their scores. First, we compute the ($u'v'$) coordinates for each SPD and age range (using the average age in each range). Second, we convert this two-dimensional information to a one-dimensional value corresponding to the green/pink component defined by the reference lamps. To do so, we compute the orthogonal projection of the BF SPD on the line formed by the reference SPDs, as illustrated in Figs. 2(b)–2(c). Finally, we rescale this numerical value by a factor of 570 (the scaling factor between distances in ($u'v'$) and our arbitrary perceptual scale, found by converting a perceptual score difference of 4 to the corresponding $(u'v')$ distance between the reference SPDs R0 and R-). This yields a predicted score for each lamp, which is shown in Fig. 2(a).

The quantitative agreement between this model and the experimental scores is remarkable. For each SPD and age group, the model is within the two-quartile box, often very close to the median value. The model thus predicts the variation with age as well as the difference between SPDs for a given age group.

This indicates that the first-order effect underlying the data is the interplay between age-dependent ocular transmission and short-wavelength SPD variations. Older participants generally have a lower sensitivity to short-wavelength radiation, causing them to perceive less of the violet peak needed for chromaticity balancing and therefore to report greener tints. This effect is accurately described by the CIEPO06 cone fundamentals, providing strong support for their validity in accounting for short-wavelength effects. Within an age group, scatter around the average trend might be caused by a combination of experimental noise and of actual variation in inter-observer perception. The CIEPO06 model cannot account for the latter effect, although models have been proposed to account for individual variations [12]. Nonetheless, CIEPO06 gives accurate guidance for the average perceived chromaticity. Given the large span of perceived chromaticity (up to $\pm$ 9 MacAdam units), our findings highlight the critical nature of this model when designing such structured SPDs.

3. Implications for LED light sources

The spectra investigated in the previous section represent an unusual case of structured SPDs with an unconventionally short pump LED wavelength, which causes a pronounced sensitivity to ocular transmission. In this section, we model and discuss the broader implications of short-wavelength effects, in the case of more conventional SPDs with blue-pump LEDs. Based on the previous experimental results, we make the assumption that chromaticities given by CIEPO06 are predictive of average perception trends.

3.1 Lighting

We seek to quantify the joint impact of observer age and spectral variations of LEDs on the perceived chromaticity of white light sources.

The prevailing architecture for warm-white LEDs is a combination of a blue-pump LED ($\lambda \sim 450$ nm), a garnet yellow phosphor, and a red nitride phosphor, as shown in Fig. 3(a). This yields a white spectrum with color rendering values deemed acceptable for general lighting ($R_a=80$, $R_9=0$). The spectral properties of the phosphors are generally very repeatable. On the other hand, blue LEDs have a wavelength distribution caused by the non-uniformity of epitaxial wafers. Manufacturers often seek to maximize their use of this distribution to improve manufacturing yield. Therefore, it is common to observe peak wavelengths in the range 440-460 nm across various products.

 

Fig. 3. Modeled chromaticity of white LEDs. (a) SPDs for general-lighting: 3,000 K LED with varying pump wavelengths ($\lambda$), targeted according to $31-2^{\circ}$ CMFs. Blue/red/green: $\lambda=440/450/460$ nm, respectively. (b) Locus of $06-10^{\circ}$ chromaticities for these SPDs and various viewer ages. ($du',dv'$) are the distances to a 3,000 K Planckian emitter. Circles/squares/diamonds: ages 20/45/70, respectively. Black line: Planckian locus. Black circles: 3- and 6-step MacAdam circles. (c) Same as previous, but with SPDs targeted according to $06-10^{\circ}$ CMFs at age 45. (d-f) Same as (a-c) for 6,500 K display LEDs.

Download Full Size | PDF

This variation is of course accounted-for in the phosphor recipe to maintain a final white chromaticity (apart from the inherent variation in the phosphor dispense process, which causes a scatter of chromaticities around the target point). However, this chromaticity is conventionally targeted using the $31-2^{\circ }$ CMFs. Since these are poorly correlated to perception, two SPDs with the same nominal chromaticity (i.e. nominally metameric) may in fact appear different to observers – an effect which can be exacerbated by age. Such effects can be especially noticeable in a mixed-illumination setting where two sources with different SPDs are observed simultaneously.

To illustrate this, we generate numerical SPDs with pump peak wavelengths of 440, 450, and 460 nm, all targeted at 3,000 K with $31-2^{\circ }$ CMFs. For each SPD, we compute the $06-10^{\circ }$ chromaticity for observer ages 20, 45, and 70. We also compute the age-dependent chromaticity of a 3,000 K Planckian radiator. For each age, we then compute the chromatic difference ($du', dv'$) between the LED and the Planckian; this enables us to subtract the overall effect of chromaticity shift with age (given by the Planckian’s shift), an effect which is unavoidable and unrelated to LEDs’ spectral variations. The difference ($du', dv'$) therefore describes the difference in perceived chromaticity at a given age, for instance in a mixed-illumination situation where halogen and LED sources are viewed side-by-side.

Figure 3(b) shows the calculated chromatic differences. Ideally, all chromaticities would be equal (i.e. $du'=dv'=0$), irrespective of age and pump wavelength. Instead, the differences can be of significant magnitude (as much as 6 ($u'v'$) units from the origin). They depend both on age and on the pump LED wavelength, in a non-trivial way. Overall, the LED sources are never metameric with each other or with the blackbody radiator. Depending on age, one LED may appear greener/pinker or warmer/cooler than another. Therefore, in mixed-illumination situations, these sources may have significantly different perceived chromaticities. These results underlie a challenge often encountered in lighting, wherein LEDs with the same nominal CCT do not match and observers cannot always agree on the type of mismatch.

This situation can be improved if the LED spectra are targeted using $06-10^{\circ }$ CMFs. For instance, we generate SPDs targeted for a 45-year-old observer. Figure 3(c) shows the resulting chromatic differences. The relative movement of each SPD with age is the same as in Fig. 3(b); however, the absolute discrepancy between chromaticities is mitigated, with most points fitting in a 3 MacAdam circle. The absolute discrepancy is zero for 45-year-old observers by design. This indicates that the use of CIEPO06 CMFs with a properly-chosen average observer age may be a suitable approach toward reducing the metameric breakdown caused by age and SPD variations.

3.2 Displays

In this section, we discuss the implication of short-wavelength effects on the color rendition of displays. We consider SPDs typical of modern LCD displays (Fig. 3(d)), obtained by combining a blue-pump LED ($\lambda \sim 450$ nm), a $\beta$-SiAlON phosphor, and a Mn-doped red phosphor, color-targeted at the D65 color point. Such SPDs enable a wide color gamut, e.g. meeting the DCI-P3 standard [20]. In the following we discuss two possible effects of the interaction between SPD and observer age, related respectively to the display’s white point and to its blue primary.

3.2.1 White point.

A first concern is the impact of age and SPD on the perception of the display’s white point. This is similar to the effect described above in the case of illumination, albeit with a different color temperature and SPD. We therefore proceed with the same analysis, generating SPDs with blue pump LEDs emitting at 440, 450, and 460 nm, all targeted to D65 with $31-2^{\circ }$ CMFs. The resulting age-dependent chromaticities according to $06-10^{\circ }$ CMFs are shown in Fig. 3(e). We still consider a $10^{\circ }$ field of view, since screens displaying extended white regions are common in various applications.

Here again, the joint effect of variation in pump LED and observer age causes large, non-systematic variations in perceived chromaticity, with some configurations being more than 0.01 ($u'v'$) units away from the target D65 chromaticity. In fact, in some cases, the relative green/pink ranking of two SPDs is inverted depending on age. This indicates that two displays calibrated for a D65 white point may appear substantially different. Again, these predictions are in line with common experience of side-by-side displays having different-looking color points.

As is the case for lighting, such discrepancies can be mitigated a priori by targeting the default chromaticity with $06-10^{\circ }$ CMFs. Figure 3(f) shows an example where SPDs are targeted for a viewer of age 45: most chromaticities are within a 6-step MacAdam circle regardless of SPD and age.

Additionally, a more advanced approach would consist of calibrating the screen given the viewer’s age. One might consider a software implementation wherein the viewer’s age is an input parameter and the display’s SPD is provided, and the white point is customized to achieve a consistent perceptual chromaticity.

3.2.2 Blue primary.

A second concern arises for displays: the perception of the short-wavelength blue primary. To quantify this, we consider the same display SPDs as in Section 3.2.1 and generate idealized RGB primary filters for each. The filters have a perfect transmission plateau over a wavelength region, then the transmission decreases linearly to zero over a 25 nm width. The plateaus are optimized such that the corresponding color display of the gamut exactly matches the DCI-P3 standard. While such transmission functions are clearly simplistic, they result in a behavior which is similar to realistic filters.

For each SPD and corresponding blue filer, we compute the $06-2^{\circ }$ chromaticity of the blue primary (using a $2^{\circ }$ field, as saturated color patches are often displayed in small areas). The results are shown in Fig. 4(a). Although these primaries are nominally metameric according to the $31-2^{\circ }$ CMFs, they vary widely, depending again in a non-trivial way on the observer age and on the pump LED wavelength. The locus of these chromaticities spans a substantial distance of about 0.1 ($u'v'$) unit. The corresponding dominant wavelength of the blue primary is shown in Fig. 4(b): depending on age and pump wavelength, it can vary by up to 10 nm.

 

Fig. 4. Modeled chromaticity of a display's blue primary. (a) Locus of chromaticity as a function of viewer age and LED pump wavelength ($\lambda$), according to $06-2^{\circ}$ CMFs. Blue diamonds/red circles/green squares: $\lambda=440/450/460$ nm, respectively. Ages span the range 20-70. The inset shows a magnification of the region of interest. Black square: D65 white point. Black line: edge of the chromaticity diagram (calculated for age 35). (b) Resulting dominant wavelength versus age

Download Full Size | PDF

These predictions may be imperfectly accurate. Indeed, the uniformity of the ($u' v'$) diagram is limited near the space boundaries, and this diagram was initially derived based on the $31-2^{\circ }$ CMFs rather than the CIEPO06 CMFs used here. Nonetheless, such a large span of chromaticities for nominally metameric primaries strongly suggests that the actual perception of saturated blue colors would be widely dependent on the actual display SPD and observer age, thus degrading the naturalness of such colors for some viewers.

As in the case of the white point, an individual calibration based on the viewer’s age would be desirable to correct for such effects. These conclusions are in line with those of [10,11]. Additional human factor research would be warranted to assess the accuracy of color spaces when they are derived from CIEPO06 CMFs – including the ($u'v'$) diagram, as well as more sophisticated object-type spaces such as CIELAB, CIECAM [21], and display-specific spaces [22].

4. Discussion

Today, most LED products are designed using conventional colorimetry based on the $31-2^{\circ }$ CMFs. This leads to user complaints about varying chromaticity in some situations – e.g. when several beams are projected on a white or gray surface, as side-by-side comparison can exacerbate tint differences. Such inconsistent chromaticity can significantly affect the acceptability of LED lighting. In part, it is simply caused by the natural dispersion in chromaticities during the fabrication process, as most LED products are binned to a 3-MacAdam ellipse. However, it is likely that inter-user variations also contribute to such occurrences.

In particular, the short-wavelength effects studied in this work appear especially relevant. Indeed, LED SPDs show most variation in the short-wavelength region, especially across varying products from different manufacturers. As demonstrated here, the interplay of these variations with age-dependent sensitivity leads to a metameric breakdown. This breakdown can be significant (more than 6 MacAdam units) and non-systematic – with viewers of different ages sometimes having opposite perceptions of compared chromaticities. Such adverse consequences are only likely to become more prevalent with the spread of human-centric LED systems, whose short-wavelength spectral content is intentionally varied – either by removing blue radiation, as in the present study, or conversely by maximizing it.

The use of modern colorimetry appears as a sound approach to deal with this increase in spectral variability. At a minimum, color-targeting products with modern CMFs and considering a mean user age would reduce metameric breakdown significantly in static light emitters. Tunable emitters (i.e. RGB-based white lighting and displays) offer the more advanced opportunity to tailor the spectrum to the sensitivity of a specific user. We note that a small amount of commercial products already use the CIEPO06 framework for binning; however, market adoption has remained scarce so far. From a manufacturer’s standpoint, the benefits of upgraded color science have to be weighed against their cost of implementation and the effort of educating users, which is compounded by the lack of suitable standards to describe such products. Therefore, an evolution of standards and specification tools towards modern colorimetry would be highly desirable.

In addition, separate from the choice of CMFs, Fig. 3 reveals that the degree of age-induced metamerism depends on the SPD: the span of chromaticity with age can vary by a factor of 2-3 depending on the wavelength of the pump LED. This brings up the more general question of quantifying the potential for a given SPD to induce ‘chromaticity metamerism’ between different users – i.e. how likely is it for different users to agree on the chromaticity of the SPD (a distinct effect from color rendition metamerism [23]). The present study focuses on short-wavelength effects; however, variations in the green and red regions are also likely to induce such metamerism, especially in highly-structured SPDs. This calls for a future metric to quantify an SPD’s chromaticity metamerism.

5. Conclusions

The perception of white LEDs, which plays a central role in their acceptability as a technology, can be significantly affected by the interplay between variability in their short-wavelength spectrum and the viewer’s age. We demonstrate this with an experiment on low-circadian-stimulation emitters having a blue-depleted spectrum: their perceived chromaticity is strongly dependent on age and SPD. The results are in excellent agreement with a model based on the modern CIEPO06 color-matching functions. Further, these effects are not restricted to highly-structured spectra: viewer age and spectral variations also have a substantial effect on the perception of conventional LED emitters, both in general lighting and display applications. These results confirm the inadequacy of conventional colorimetry and provide support for the future use of modern color-matching functions in designing LED spectra, to improve their perceptual consistency.

Appendix A – Blue-free emitters

The impact of blue radiation on our circadian cycle is currently a topic of great interest [24–26], spurred by the discovery of specific blue-sensitive retinal cells with circadian-regulating properties [27] and by the study of their wavelength sensitivity [28].

Importantly, conventional amounts of artificial light are sufficient to influence the circadian cycle, whether from indoor lighting [29] or electronic displays [30]. Engineering the spectrum of light has been confirmed to impact circadian entrainment [31,32]. In addition, simple metrics have been introduced to quantify the circadian entrainment of light sources, such as the melanopic lux [33]; these constitute tools to guide spectral design. Accordingly, the ability to engineer LED spectra opens new perspectives in harnessing this effect and providing light that is better-suited to human biological needs.

This opportunity is twofold. During the day, blue-enriched SPDs may increase wakefulness, especially for users suffering from insufficient daylight exposure; such SPDs are easily achieved by selecting suitable pump LEDs and phosphors, and by employing a high CCT. Conversely, in the few hours before bedtime, one should reduce the blue content and the intensity of light sources. However, blue suppression is challenging from a technology standpoint: nearly-all current LEDs are predicated on the use of blue-emitting pump chips, meaning blue radiation is unavoidable in the final emission. Reduction in blue content is thus necessarily tied to CCT; this is problematic as very low CCTs (below 2,500 K) have a pronounced yellow tint and offer poor color discrimination. Nonetheless, this low-CCT approach is found in a majority of ‘sleep-friendly’ lighting products and in display-tuning software, for lack of a better solution. Figure 5(a) illustrates the relationship between CCT and melanopic lux for conventional emitters.

Recently, we proposed the use of the ‘blue-free’ LED architecture to break this trade-off between chromaticity and blue content [34]. The combination of efficient violet-pump LEDs [35] with appropriate green and red phosphors yields an SPD with a wide spectral gap in the blue range (less than 1% power in the range 440-500 nm). By properly tuning the relative magnitude of the violet, green, and red emissions, any desired chromaticity can be achieved – including cool- and warm-white. An example is shown in Fig. 5(b). This SPD has a CCT of 2,700 K, high color fidelity metrics ($R_a$=80, $R_9$=90), and a relative melanopic lux below 65% of conventional emitters. As seen in Fig. 5(a), this SPD therefore provides the circadian entrainment of a sleep LED combined with the chromaticity of a conventional LED, resulting in a practical solution for a low-circadian-stimulation light source with sufficient light quality.

 

Fig. 5. (a) Relationship between CCT and relative circadian lux (normalized to unity at 2,700 K) for various sources. Blue line: Planckian emitters. Gray dots: conventional blue-pump LEDs. Orange dot: blue-pump sleep LED. Purple dots: Blue-free LEDs. (b) SPD of a blue-free LED with good color rendition properties.

Download Full Size | PDF

Appendix B – Perception study results

Sixty-four participants took part in the perception experiment. Their demographic details and chromaticity perception scores are shown in Table 2.

Table 2. Results of the Perception Study

View Table | View all tables in this article

Acknowledgments

We would like to thank Prof. Jan Henrik Wold for access to the upcoming CIE TC1-97 color-matching functions.

References

1. T. Smith and J. Guild, “The CIE colorimetric standards and their use,” Trans. Opt. Soc. 33(3), 73–134 (1931). [CrossRef]  

2. W. A. Thornton, “Toward a more accurate and extensible colorimetry. Parts I-III,” Color Res. Appl. 17(2), 79–122 (1992). [CrossRef]  

3. L. T. Sharpe, A. Stockman, H. Jagle, H. Knau, G. Klausen, A. Reitner, and J. Nathans, “Red, green, and red-green hybrid pigments in the human retina: correlations between deduced protein sequences and psychophysically measured spectral sensitivities,” J. Neurosci. 18(23), 10053–10069 (1998). [CrossRef]  

4. M. Neitz and J. Neitz, “Molecular genetics of color vision and color vision defects,” Arch. Ophthalmol. 118(5), 691–700 (2000). [CrossRef]  

5. J. Pokorny, V. C. Smith, and M. Lutze, “Aging of the human lens,” Appl. Opt. 26(8), 1437–1440 (1987). [CrossRef]  

6. G. Abraham, H. Korosi, J. Schanda, A. G. Shapiro, and K. Wenzel, “Anomalies in additive colour matches,” Color Res. Appl. 20(4), 235–244 (1995). [CrossRef]  

7. J. Xu, J. Pokorny, and V. C. Smith, “Optical density of the human lens,” J. Opt. Soc. Am. A 14(5), 953–960 (1997). [CrossRef]  

8. “Fundamental chromaticity diagram with physiological axes – Part I,” CIE Technical Report 170-1:2006 (2006).

9. A. Sarkar, F. Autrusseau, F. Vienot, P. Le Callet, and L. Blonde, “From CIE 2006 physiological model to improved age-dependent and average colorimetric observers,” J. Opt. Soc. Am. A 28(10), 2033–2048 (2011). [CrossRef]  

10. D. L. Long and M. D. Fairchild, “Modeling observer variability and metamerism failure in electronic color displays,” Proceedings of the Color and Imaging Conference, Society for Imaging Science and Technology (2014).

11. Y. Asano, M. D. Fairchild, L. Blonde, and P. Morvan, “Color matching experiment for highlighting interobserver variability,” Color Res. Appl. 41(5), 530–539 (2016). [CrossRef]  

12. Y. Asano, M. D. Fairchild, and L. Blonde, “Individual colorimetric observer model,” PLoS One 11(2), e0145671 (2016). [CrossRef]  

13. M. J. Murdoch and M. D. Fairchild, “Modelling the effects of inter-observer variation on colour rendition,” Light. Res. Technol. 51(1), 37–54 (2019). [CrossRef]  

14. Online repository for CIE TC 1-97 colorimetric functions, courtesy of Prof. Jan Henrik Wold: https://github.com/ifarup/ciefunctions.

15. “Fundamental chromaticity diagram with physiological axes – Part II: spectral luminous efficiency functions and chromaticity diagrams,” CIE Technical Report 170-2:2015 (2015).

16. D. L. MacAdam, “Visual sensitivities to colour differences in daylight,” J. Opt. Soc. Am. 32(5), 247–274 (1942). [CrossRef]  

17. “Chromaticity difference specification for light source,” CIE Technical Note 001:2014 (2014).

18. A. David, M. R. Krames, and K. W. Houser, “Whiteness metric for light sources of arbitrary color temperatures: proposal and application to light-emitting-diodes,” Opt. Express 21(14), 16702–16715 (2013). [CrossRef]  

19. M. Wei, K. W. Houser, A. David, and M. R. Krames, “Perceptual responses to LED illumination with colour rendering indices of 85 and 97,” Light. Res. Technol. 47(7), 810–827 (2015). [CrossRef]  

20. “Cinema quality – reference projector and environment,” SMPTE Recommended Practice RP 431-2 (2011).

21. N. Moroney, M. D. Fairchild, R. W. G. Hunt, C. Li, M. R. Luo, and T. Newman, “The CIECAM02 color appearance model,” Proceedings of the Color and Imaging Conference, Society for Imaging Science and Technology (2002).

22. M. Safdar, G. Cui, Y. J. Kim, and M. R. Luo, “Perceptually uniform color space for image signals including high dynamic range and wide gamut,” Opt. Express 25(13), 15131–15151 (2017). [CrossRef]  

23. A. David, T. Esposito, K. Houser, M. Royer, K. A. G. Smet, and L. Whitehead, “A vector field color rendition model for characterizing color shifts and metameric mismatch,” LEUKOS (posted 01 February 2019, in press).

24. J. M. Zeitzer, D. J. Dijk, R. Kronauer, E. Brown, and C. Czeisler, “Sensitivity of the human circadian pacemaker to nocturnal light: melatonin phase resetting and suppression,” J. Physiol. 526(3), 695–702 (2000). [CrossRef]  

25. C. Cajochen, J. M. Zeitzer, C. A. Czeisler, and D.-J. Dijk, “Dose-response relationship for light intensity and ocular and electroencephalographic correlates of human alertness,” Behav. Brain Res. 115(1), 75–83 (2000). [CrossRef]  

26. S. W. Lockley, G. C. Brainard, and C. A. Czeisler, “High sensitivity of the human circadian melatonin rhythm to resetting by short wavelength light,” J. Clin. Endocrinol. Metab. 88(9), 4502–4505 (2003). [CrossRef]  

27. S. Hattar, H. W. Liao, M. Takao, D. M. Berson, and K. W. Yau, “Melanopsin-containing retinal ganglion cells: architecture, projections, and intrinsic photosensitivity,” Science 295(5557), 1065–1070 (2002). [CrossRef]  

28. G. C. Brainard, J. P. Hanifin, J. M. Greeson, B. Byrne, G. Glickman, E. Gerner, and M. D. Rollag, “Action spectrum for melatonin regulation in humans: evidence for a novel circadian photoreceptor,” J. Neurosci. 21(16), 6405–6412 (2001). [CrossRef]  

29. J. J. Gooley, K. Chamberlain, K. A. Smith, S. B. Khalsa, S. M. Rajaratnam, E. Van Reen, J. M. Zeitzer, C. A. Czeisler, and S. W. Lockley, “Exposure to room light before bedtime suppresses melatonin onset and shortens melatonin duration in humans,” J. Clin. Endocrinol. Metab. 96(3), E463–E472 (2011). [CrossRef]  

30. A. M. Chang, D. Aeschbach, J. F. Duffy, and C. A. Czeisler, “Evening use of light-emitting eReaders negatively affects sleep, circadian timing, and next-morning alertness,” Proc. Natl. Acad. Sci. U. S. A. 112(4), 1232–1237 (2015). [CrossRef]  

31. V. Gabel, M. Maire, C. F. Reichert, S. L. Chellappa, C. Schmidt, V. Hommes, A. U. Viola, and C. Cajochen, “Effects of artificial dawn and morning blue light on daytime cognitive performance, well-being, cortisol and melatonin levels,” Chronobiol. Int. 30(8), 988–997 (2013). [CrossRef]  

32. S. A. Rahman, M. A. St Hilaire, and S. W. Lockley, “The effects of spectral tuning of evening ambient light on melatonin suppression, alertness and sleep,” Physiol. Behav. 177, 221–229 (2017). [CrossRef]  

33. R. J. Lucas, S. N. Peirson, D. M. Berson, T. M. Brown, H. M. Cooper, C. A. Czeisler, M. G. Figueiro, P. D. Gamlin, S. W. Lockley, J. B. O’Hagan, L. L. A. Price, I. Provencio, D. J. Skene, and G. C. Brainard, “Measuring and using light in the melanopsin age,” Trends Neurosci. 37(1), 1–9 (2014). [CrossRef]  

34. M. R. Krames and A. David, “US patent 9410664 – circadian-friendly LED light source,” (2014).

35. C. A. Hurni, A. David, M. J. Cich, R. I. Aldaz, B. Ellis, K. Huang, A. Tyagi, R. A. DeLille, M. D. Craven, and F. M. Steranka, “Bulk GaN flip-chip violet light-emitting diodes with optimized efficiency for high-power operation,” Appl. Phys. Lett. 106(3), 031101 (2015). [CrossRef]