In a recent article published in Architectural SSL on the topic of blue light content of LEDs, I attempted to present the discussion of blue light from the perspective of those raising concerns about blue light hazards against known and practical objective knowledge on the topic. The article covered the gambit of concerns, from retinal damage concerns to melatonin levels in occupants, from both sides of the argument, as there are those who dismiss this as a non-issue out of hand. The article also forwarded two conclusive suggestions. First: The research on this specific topic, as it relates specifically to LED light sources, is a little thin. Second: For those concerned about blue light content, selecting LEDs of a lower CCT and higher CRI delivered the lowest blue light content. Whether or not this is the best choice for visual acuity was not the subject of the article, nor was it suggested as the best solution overall. There is a great deal of research supporting the concept of high CCT light for enhancing human visual performance. Much of this was completed under light sources other than LEDs, so there is no caveat included that states anywhere that blue light content of LEDs is at acceptable levels, or of no concern.
On the topic of blue light, the conversation will invariably include a mention of the fact that human kind evolved under daylight, which contains blue light, thus, there is no issue with blue light to be concerned with. That argument is nonsense of course. We also grew up under some amount of UV light as well, yet we know for a fact that this is dangerous to both our skin and eyes. When we added sunglasses to the formula to cut light to the retina, which then causes the pupil to dilate to accept more light, we soon discovered that our low grade sun shades were not filtering UV light to the same degree as visible light, which created an explosion in retina cancers and other UV related physical damages we now know to control. This is why virtually all sunglasses sold today now include specific UV filters to control damaging rays, and why we humans, evolved under the sun, wear sunblock to protect ourselves. We also live under a lot if infrared radiation, but are hardly going to be sticking our faces in the fire claiming our eyes are safe, since that IR radiation has been around since we emerged as a species. The reality is, artificial light is just that, artificial. It is not natural light, no matter how we market it. This artificial creation of light is similar to our fatty foods. The balance between what is a natural blend of fats, proteins and carbohydrates is manipulated to achieve some result that is not the same as cutting a raw slab of meat from a living animal and eating it, or simply consuming a raw unprepared vegetable. Just as we have taken what naturally nourished us in raw foods that now cause us disease, artificial light can and does contain similar liabilities. This has been the case long before LEDs, and is still with us in the solid-state age.
The IES has issued an interesting document on the topic of higher CCT light sources as a more effective approach to efficient visual performance. This is TM-24-13 ‘Adjusting the Recommended Illuminance for Visually Demanding Teasks Within IES Illuminance Categories P through Y Based on Light Source Spectrum’ (a ridiculous document name BTW), is frequently quoted by those supporting a very high CCT light source be used to enhance visual performance. However, this is not the core data point TM-24 uses to establish light source choice at all. The metric used is the S/P ratio, or Scotopic/Photopic ratio, which indicates how the spectral power of a light source supports the full range of Mesopic vision. In this document, typical sources are cited as examples, based on common lamps and light sources, from HPS to Daylight, with a relative S/P ratio for each. None of these sources were LEDs, nor was the discussion of blue light content included in the EVE (Equivalent Visual Effect) summary. There is no suggestion, positive or negative, that LEDs of any CCT support the relative EVE or S/P ratios based solely on the CCT they produce. Further, there is nothing in the document suggesting that the characteristics of LEDs will enhance or detract from the evaluations, because, simply, LED light sources were not addressed at all. The only solid conclusion one can make from this document is that a high S/P ratio is a significant factor in optimizing visual acuity. While with today’s available light sources, and using daylight as the supposed “perfect model”, it might be suggested that high CCT sources alone are the solution, this would be an overly broad, overly generalized and simplistic interpretation of the content of TM-24.
Now, that set out, let’s look into another topic I believe is being slipped by without proper scrutiny. This is the idea of modifying human biophysical responses to light by purposely introducing a light spectral power distribution to attain a desired result. Specifically I am talking about the known effect of blue light in the range of between 430 and 460nm that imparts a melatonin suppression response, with a circadian efficiency peak of 485nm+/-. It is suggested that if we purposefully manipulate and add (or allow to exist) blue light at these wavelengths, we will help people feel less drowsy and more energetic. I believe this broaches an area of ethics we as a community need to fully resolve before we accept this as a desired effect and forge ahead with high CCT light sources, believing we are reaping the benefits of better visual performance with a pep-pill bonus. We all know that adding stimulants to drinking water could reduce drowsiness in working environments. This is, of course, unethical (not to mention illegal), so we don’t do it. So, is purposefully manipulating circadian rhythms of offices and manufacturing facility employees, using light that is specifically and purposefully, in full knowledge and intent, doped to reduce melatonin levels to increase alertness? Do we know for a fact, or even beyond a reasonable doubt, that this has no long term health effects? I suggest that this is not an ethical approach to lighting, and that any attempt to purposefully amp up alertness by doping the lighting system with increased blue light without the consent and knowledge of the occupants, raises serious questions that are well beyond the realm of marketers and proponents of high CCT/high Blue light products.
This leads to the oft-cited suggestion that LEDs produce no more blue light than any other light source, thus are of no concern. The DOE has even supported this position in statements, effectively dismissing any discussion of the topic as irrelevant. What’s missing, and of critical importance, is that this generality is simply incorrect. While there are some LED sources that deliver no more blue light than other conventional sources, does this mean that the sources held up for comparison are acceptable as a baseline? How do they align with the vaunted perfect light source, daylight, and is there a way to attain what we want from light sources without the increased blue light content? In the sweeping, overly broad statements made, none of these questions are addressed, only avoided.
At various times I have suggested that high CRI LEDs of any CCT, produce less blue light as a portion of total light energy, than low CRI LEDs. This is based on LED light sources founded on the effect of phosphor conversion of either a 405nm violet, or a 456nm blue die light source. Since we are talking about blue light, as defined by melatonin suppression (430 and 460nm) the more of the core blue light of the LED that escapes, the greater the blue light presence will be. To increase CRI of an LED, more blue light needs to be down converted to the yellow-green and red wavelengths, leaving less escaping as raw blue light. Thus, the higher the CRI, and a subsequent lower blue light content as a portion of total light energy. As you will see here, this theory has its limits.
So, with this all in mind, I did a few tests of light sources to see just where we are with the blue light content issue between sources. I captured spectral data in micro-Watts per square centimeter using a Mightex CCD Spectrometer, then normalized all sources to an identical total output in watts of irradiance between 400 and 700nm, where we do most of our seeing. Withing that total output, I isolated the energy produced between 430 and 460 (>430 to <460) to see how much of the total energy produced by the source landed in the blue light zone being discussed. Here are the results:
While most spectral data comparing products shows relative distribution, its not until actual irradiance is compared that the differences between sources is revealed. In this case, all of the sources are generating the same exact total irradiance in microwatts per square centimeter, so the differences in balance between produced wavelength energies is made apparent. Note the red spike for the 2800K LED, which generates its high CRI rating, comes from an added red LED die in the array, not phosphor conversion.
- 5800 K Daylight (cloudy sky) = 98.9 CRI – 9.6% of total light energy falls between 430 and 460nm
- 5000K Daylight (direct sun) = 99 CRI – 8.4% falls between 430 and 460nm
- 5800K LED @ 65 CRI – 20.8% falls between 430 and 460nm
- 5000K LED @ 90CRI – 13.1% falls between 430 and 460nm
- 4000K LED @ 80CRI – 11.2% falls between 430 and 460nm
- 4000K D841 T8 Fluorescent Lamp – 12.2% falls between 430 and 460nm
- 2800K LED @92% – 6.8% falls between 430 and 460nm
- 2500K Halogen @99.7 CRI – 2.2% falls between 430 and 460nm
While the graphic looks like a jumble, looking at it carefully reveals the ramp from lower left to upper right of halogen, the distinctive blue pump hump of the LEDs, the flatter total coverage of daylight, and the strange tri-phosphor spikes of the fluorescent source. Note: In these graphics, I have reverted to relative spectral data.
The products tested ranged over and under the black body line, which is apparent when comparing them side-by-side as some yellow green, some magenta. To make matters even more interesting, the poor CRI 5800K and 4000K T8 lamp came closest to sitting on the black body line, even beating daylight, while not matching its CRI (or CQS). This indicates flaws in our modeling, including and beyond CRI itself.
So much for the LED not producing more blue light myth, and the disconnect between CRI and blue light discussion. In LEDs, not only is color quality a measure of a sources visual appearance, it can be corollary to the amount of blue light one can expect to be emitted from the light sources as well – although, as I will demonstrate, this is not always true either. In other words, we must stop assuming that all CCTs generate the same amount of blue light, or that CCT alone is a meaningful single metric in determining blue light content.
Please note that all white light will deliver blue light above the described 460nm through the peak circadian efficiency peak of 485nm. This is necessary to realize any full white appearance. However, that fact does not diminish the impact of addition blue light presence below this peak, as the circadian response curve is quite broad, ranging from <380nm through 625nm.
If we use a 9% of total energy as a baseline of acceptable blue light balance in the spectral power distribution of light sources, the following conclusions can be drawn:
- A low CRI 5800K LED produces 2.3 times the blue light as daylight
- A 90CRI 5000K LED produces 45% more blue light than daylight, but 48% less than the lower CRI/Higher CCT LED
- The high CRI 5000K LED generates no more blue light than an 80CRI 4000K LED or T8 fluorescent
- The 2800K 92 CRI LED generates 25% less blue light than daylight
- The 2500K 99CRI halogen lamp generates 75% less light than daylight
This is, by no means an endorsement of reverting to halogen light sources or call to abandon the pursuit of LEDs or high CCT light sources. What I am suggesting is that we be realistic and recognize what we are working with, and if it is possible to realize both high CCT s and high color performance, while controlling blue light content, perhaps we should. The 5000K LED at 90CRI is one example of this, we gain the higher 5000K CCT over the 4000K, with greater color accuracy, while suffering no gain in blue light. The next question is whether this is corollary to the important S/P ratio. For that, I re-measured each of the sources using the Asensetek Spectrometer to create a side-by-side comparison as it would be applied using the principles described in TM-24-13. Here are those results:
- 5800K Daylight – S/P Ratio 2.41
- 5800K / 65CRI LED – S/P ratio 1.83 (does not measure within TM-24 table showing 2.0+)
- 5000K /90CRI LED – S/P ratio 1.61 (does not measure within TM-24 table showing 1.85-2.05)
- 4000K / 80CRI LED – S/P ratio 1.36 (does not measure within TM-24 table showing 1.6-1.8)
- 4000K / 80 CRI T8 – S/P ratio 1.59 (at bottom of TM-24 table showing 1.6-1.8)
- 2800K / 92CRI LED – S/P ratio 1.36 (within TM-24 table)
- 2500K Halogen – S/P ratio 1.24 (within TM-24 table)
Not what I expected, and more than a little dissapointing. It appears that LEDs are not delivering the same S/P ratio performance as the fluorescent sources cited in the TM-24 recommendations, or the daylight and fluorescent equivalents of the same or similar CCT. Apparently more work is required here. All of the LED sources, regardless of CRI or CCT, are falling short, which indicates they will not perform better than their fluorescent counterparts. Even when the 4000K LED and T8 lamp are compared in my little test, the LED fell short. Additionally, while I personally hoped to see a connection between the CRI and the S/P ratio, this was not to be. Quite the opposite. The high CCT, low CRI 5800K LED topped the charts in regard to the S/P ratio (of the sources tested thus far anyway) while also delivering poor color performance and the greatest amount of blue light. This indicates to me there is a serious flaw in the way all of this is being evaluated that needs to be resolved. Otherwise, TM-24 S/P recommendations become antithetic to color quality, which is not an attractive prospect.
In an effort to understand this a bit more, I decided to make my own LED recipe using some remote phosphor plates and a 456nm blue array I had lying around – to see if I could mix up a very high CCT source with good CRI, yet avoid the blue light content. Here is the result:
This the CIE diagram of the two farbicated LED samples. Note how far south of the blackbody line they fall, looking a little magenta. Also, illustrates the issues with CRI, as these two did well there. CQS calculated to just 80. However, even it has no data to show how far these were off being white.
7000 and 8700 CCT LED Fabricated from Blue pump and phosphor discs
- Fabricated 7000K LED delivered 91CRI – 22.3% falls between 430 and 460nm – S/P ratio of 2.3
- Fabricated 8700K LED delivered 90CRI – 24.3% falls between 430 and 460nm – S/P ratio of 2.5 (spot on to the sky reference and 8000K fluorescent reference in TM-24
I was encouraged that a 7000 and 8700 CCT LED system could be made to generate 90+ CRI. Unfortunately all this did was demonstrate how poor CRI calculations are in this region, as the CQS values for these were a solid 10 points lower, while their position well below the black body line reflects their somewhat purplish presence. As a minimum, we should be given a Du’v’ value for all color ratings – but that is another topic altogether. Although, while the color looked off, it did a fair job of looking white on reflected surfaces and did render colors on my color checker nicely enough. The S/P ratio certainly looked good, nearing daylight and better than any of the other LEDs I tested before. So, at the least we can conclude that a high color rendering accuracy does not harm attaining a high S/P ratio.
The blue light evaluation produced decidedly disappointing results, delivering the highest of all sources tested so far. Obviously, a high CCT/high CRI (CQS) with high S/P ratio can be achieved while still deliver very high blue light content. So my high CRI to low blue light content argument falls apart as the CCT increases.
It also appears that there is a missing piece to this puzzle – since I can nearly duplicate daylight performance in both CCT, CRI, and S/P ratio, but deliver nearly 3 times the amount of blue light in balance compared to natural light. That means these artificial light sources have 3 times the impact on melatonin suppression than natural daylight. I do not find that a desirable result, nor should anyone else. I am not even considering or suggesting that there is any health impact other than potential sleep/wake cycle effects. While there is some research indicating blue light near the 400nm through 450nm range can cause retinal damage, there are at least as many papers indicating that retinal illuminance levels are unlikely to be high enough to be a concern. So, we’ll leave that topic for others or another time. My concern is that application of LEDs everywhere, coupled with the advantages of increased CCTs in visual performance to reduce illumination levels (to trim energy use) means we are going to be exposing ourselves to a significant increases in spectral energies that effect our melatonin levels. Dismissing this out of hand seems reckless. As an insomniac, the prospects are not attractive at all.
The heart of the blue light issue as it relates to melatonin, is the use of the 456nm blue LED. Move to a 405nm violet LED, and the majority of the energy from the core LED falls outside the range of concern. However, being closer to UV light energy brings with it its own baggage that will need to be addressed, while most 405nm LEDs fall short of the lofty efficacy we see with 456nm products. Further, my experience with conventionally made 400nm to 405nm LEDs is they are significantly more sensitive to temperature, aggravating getting the most of them. Perhaps there is a solution in GaN on GaN architecture that resolves that issue.
Overall, if we agree that the goal is to achieve optimal visual performance, as shown in TM-24-13 – through increased CCTs to some degree, while seeking a high S/P ratio, we need to discover a way to maintain good color quality and accuracy (which begs a new standard, as the CRI is obviously not helping us here), while also managing the blue light content. It seems to me, after looking at these results, that all that blue light escaping is lost efficiency in delivering high color performance and creating even better S/P ratios for visual acuity. However, I was unable to achieve the desired result on my admittedly make-shift work bench.
So, to the origins of this discussion. While I have suggested that for those most concerned about blue light, the only real approach using LEDs available on the market today is to select lower CCTs with high CRI/CQS scores. These products use up more of the blue light to create the yellow-green and red colors necessary to achieve the higher color accuracy results. This was consistent with the test results presented here. When you get beyond 5000K, all bets are off, no matter what CRI/CQS rating the source achieves. This is especially true for spaces where high precision lighting is not the task in hand, and where relaxation before heading to bed is in order, the 2800 93CRI LED tested generates less blue, melatonin suppressing light than daylight. However, it does generate 3 times that of halogen, so there may be a case made for bedside lighting remaining conventional, or consider a light source like Soraa’s LEDs, which use 405nm LEDs as their blue pump, which is below the threshold of concern.
For high precision task lighting, the blue light issue is harder to avoid. To get the high S/P ratio means higher CCTs, often much higher than what many of us are accustomed to. While this can be done with high CRI (which does reduce blue content to some degree), it is not necessarily corollary, while most high CCT LEDs are poor in color accuracy (and very high blue content). Couple this to the fact that these applications also include use of higher illuminance levels, compounding exposure as a whole. There is also a failure in the S/P and CRI/CQS metrics when we get beyond 4000K, indicating a more careful evaluation is necessary to make solid decisions. Assuming a high CCT and high S/P ratio is the whole story is obviously inaccurate. Further, with blue light energies – as a component of total light energy – exceeding what we experience under daylight by as much as a factor of 3 (much more if you consider low CRI LEDs with much higher blue content), being cautious and conscious of the effect this might have on sleep patterns is a prudent measure. We might keep in mind that workers in construction, precision shooters, and others exposed to daylight on a daily basis have found blue light filtering glasses to be an aide to both visual acuity and comfort. Under the same CCT of LED light, there is an even greater component of blue to be considered. Whether or not this is going to be an issue, we do not yet fully understand. Whether the benefits of the higher CCT S/P ratio are negated by the high blue light content of LEDs is something else that will need to be studied and resolved.
I suggest that as time passes and LED technology develops, much of the blue light issue will fade as a combination of growth in understanding and increased conversion efficiency where that energy is used to generate improved color qualities – rather than simply allowed to escape, assuming some attention is paid to the issues raised here. In the meantime, avoiding simplistic assumptions, either for or against blue light, CCT levels, CRI/CQS accuracy, and S/P ratios and their meaning, is the first step in breaking through to greater results. Moving toward lowering blue light, raising CCT, establishing a new color accuracy metric, and establishing a metric for clearly indicating and defining the effect of spectral power on human visual and non-visual response also appear to be in order in the long term.