Light Flicker from LED Lighting Systems - an Urgent Problem to Solve

TECHNOLOGIESCATEGORY LIGHT FLICKER & DRIVERS 50 Light Flicker from LED Lighting Systems - An Urgent Problem to Solve

Flicker is the modulation of a lamp’s light output caused by fluctuations of the mains voltage supply. Recent research has shown that fluctuations of short wavelength emissions are perceived to a higher extent and light flicker may have a huge influence on the well-being of end users. Prof. Georges Zissis, Head of Light & Matter Research Group at the LAPLACE Université de Toulouse presents and discusses the influence of driver topologies, research results, metrics and standards.

Light flicker is one of the so- Recent investigations show Light Flicker and its Impact called Temporal Light Artefacts that some LED lighting products on Health and Well-Being (TLA) defined as undesired may exhibit anomaly high flicker Light flicker has always been changes in visual perception rates, especially under dimming an issue for lamps with possible induced by a light stimulus conditions. While today there important health impacts. There are whose luminance or spectral are no mandatory regulations, more conditions that are supposed distribution fluctuates with time there are some stringent to be sensitive to flicker but for an observer in a certain recommendations. This is regardless of how much they environment. The second TLA an important item for both correlate to this, everyone would is the stroboscopic effect. consumer satisfaction and like to see it minimized. It isn’t Light flicker combined with consumer acceptance of uncommon to hear people in offices rotating, moving parts or spatial Solid State Lighting products. complain about headaches and patterns may be responsible dizziness brought on by fluorescent for stroboscopic effects. lamps with magnetic ballasts [1, 2]. Stroboscopic effects might In fact, research has shown that induce hazards to workers in fluctuations of short wavelength proximity to rotating machines emissions are perceived to a and tools. The typical frequency higher extent [3, 4]. range in which the stroboscopic effect is perceived is from It is known that exposure to light 80 Hz up to 2000 Hz. flicker (in particular at frequencies between 3 Hz and 55 Hz) can cause With the ever-increasing photosensitive epileptic seizures in penetration of light emitting various forms, depending on the diode lamps in more individual and his visual pathology, applications and considering the contrast, the wavelength and that their mode of operation the viewing angle or distance. differs from the legacy Approximately 1 in 4,000 humans technologies, one might suffer from photosensitive epilepsy. expect that light flicker Women and elder people are more would disappear. sensitive to flicker than men and younger people.

Furthermore, it is known that people who suffer from migraines are more likely to be sensitive to flicker.

Figure 1: In addition, flicker may have serious Light flicker definition consequences for people suffering from specific medical conditions.

On the opposite side, pulsed lights may also have some positive effects: it has been reported that the pulsed operation of lamps could offer opportunities for energy savings according to the Broca-Sulzer effect [5, 6] due to enhanced perceived brightness. Thus, it is argued that energy savings can be achieved by using pulsed LEDs at a very high frequency. In that case it is Figure 2: absolutely necessary to understand Experimental c the influence of flicker on humans in orrelation between order to avoid any deleterious Flicker Modulation effects appearing with products and Flicker Index (R2 is the least using that type of pulsed light. square determination coefficient) Definitions, Measurement and Origin of Light Flicker Light flicker refers to quick, periodic changes in the lamp’s light output based on the “luminance contrast modulation”. Light flicker is due to fluctuations of the absorbed electrical power mainly due to fluctuation or modulation of the mains voltage. As the mains frequency in Europe is 50 Hz the light flicker is expected SZ1 at 100 Hz (120 Hz, if the mains Fl = ______(2) For the moment there is not any frequency is 60 Hz). SZ1 + SZ2 formal proof that, for periodic light variations, one or the other metrics Referring to figure 1, there are is considered as more accurate for two widely accepted methods Both approaches are used in describing the flicker perception for defining light flicker [7]: lighting. In fact Flicker Modulation by the end user. Furthermore, metrics is accurate only in the case Kitsinelis et al [9] showed that, Flicker Modulation (or percent of periodic signals and doesn’t take in the case of sinusoidal-like signals flicker): it takes the maximum the waveform of the light variations found in the large majority of and minimum luminance values into account. On the contrary, electrical lamps there is a strong

(respectively Lmin, Lmax) of the Flicker Index accounts for obvious correlation between Flicker fluctuation and performs the differences in waveform shape and Modulation and Flicker Index; following calculation: can be used in fully aperiodic Figure 2 illustrates this finding. signals. Flicker Index is higher when Assuming that Flicker Index is zero

Lmax - Lmin rising and/or falling fronts are steep, when Flicker Modulation is null we F [%] = 100 ∙ ______= m Flicker Modulation can’t show this can obtain the following relation: Lmax + Lmin effect. As Michael Poplawski [8] said Fm “percent flicker is extremely simple _____ ≈ 3.3 (3) | ΔLmin | 100 ∙ ______(1) to determine - requiring only the Fl L + L max min measurement of maximum and

minimum values with respect In the above relation Fm is expressed Flicker Index: it requires the to a reference and simple math. in decimal value not in percentile. calculation of the areas of “zone 1” Flicker Index, on the other hand, and “zone 2” situated above (SZ1) requires the accurate measurement As has been mentioned in the above and below (SZ2) the average of waveform shape with respect paragraphs the light flicker definition luminance as defined by its to a reference and much more is based on luminance contrast RMS value (Lav). complex integral math”. modulation (or variation in general).

Figure 3: Experimental set- up for light flicker measurement in LAPLACE laboratory

However, luminance measurements To compare the light flicker “driver”) that is interfacing the LEDs are rather difficult to carry out. behaviour of different technologies inside the bulb to the mains plug. For this reason Zissis [10] proposed (especially LED omni-directional Figure 5 shows simple schematics to measure light flicker based on lamps and LED spots) a 100 W of LED ballast generic architecture. luminous flux time variation instead A-shape incandescent lamp Light flicker is then linked to two of luminance contrast modulation. has been used as reference. facts: (1) residual ripple after the This is a good approximation The reference lamp exhibited a 10% rectifier-filter stage and (2) final because it smooth’s the spatial Flicker Modulation value at 100 Hz, stage converter architecture. variation of the light emitted by the the light signal was very close to lamp. The measurement can then sin2(2πƒt) shape. This is perfectly Residual ripple is responsible be carried out using an Ulbricht logical. Light fluctuation in an for light flicker at a frequency of integrating sphere equipped by a incandescent lamp is, in fine, twice the mains, whereas, time resolved light detector. due to the variation of filament converter architecture usually temperature within a full AC-period; leads to high frequency flicker Figure 3 shows the original set-up the temperature depends on the (several tens of kilohertz). at LAPLACE laboratory in Toulouse. Joule heating which is proportional Low quality, low price, ballasts This is a very robust way to evaluate to sin2(2πƒt) and light emission using passive filters are usually light flicker. In this primitive version follows Stefan’s law proportional to responsible for high light flicker. of the set-up an Ø80 cm integrating the 4th power of the filament As US-DoE mentioned [11], a low sphere was equipped with a temperature (thus a small variation cost requirement for a small integral selenium cell as a photo-detector within an AC-period is responsible LED lamp may force a fundamental coupled to a Tektronix 2002B for measurable luminous flux trade-off between flicker and the oscilloscope for detecting and variations at twice the mains power factor. Add to that, that very recording the light waveforms and frequency). This amount of light often, passive elements used as a true-RMS voltmeter for measuring flicker at 100 Hz is not perceivable filters (for instance chemical the average light output (in millivolts) by the human eye. capacitors) have a low lifespan and independently. The selenium cell a huge impact on the environment has a rather slow answer and thus LED flicker characteristics are [11]. All in all, a large amount light only low frequency flicker can be primarily a function of the LED flicker at 100 Hz constitutes a detected (up to 1 kHz). Since then driver. However, LEDs have to be serious indication that ballast the photo-detector was replaced by driven in DC-current; so, in theory quality and reliability are probably a rapid photodiode (SLD-70BG2) light flicker should be missing. unacceptably low. coupled with a logarithmic amplifier However, a LED lamp uses a driver in order to resolve light flicker at (CIE official vocabulary imposes Dimming a LED source can increase higher frequencies (up to 100 kHz). the use of term “ballast” instead of or induce flicker, most notably when phase-cut controls are used and/or Figure 4: Schematics of pulse-width modulation (PWM) is usual LED ballast employed within the driver to reduce architecture the average light output from the LED source.

It should be noted that Kitsinelis et al [13] proposed a very effective way to detect light flicker using the camera of a cellular phone.

Figure 5: Flicker detection using the built in camera of a cell phone obtained by LAPLACE. Flicker leads to the appearance of dark fringes on the image

Figure 5 shows the obtained results of the Flicker Modulation. Generally This easy and quick flicker test with by this method. The simplicity of speaking, when fringes are visible such a widely available tool as a the method allows its use by the Flicker Modulation is higher than mobile phone may prove useful to general consumer: just use a smart 20%. Large fluctuations (i.e. large consumers seeking to avoid the phone with an integrated camera, flicker) are clearly recorded as undesirable effects of flicker even target the lamp and look at the striped images on the smartphone at such non visible frequencies. phone screen. If striations appear display. The stripes tend to fade with around the lamp (dark fringes), lower Flicker Modulation but they then flicker is present. The spatial can still be detected by fine-tuning Light Flicker frequency of the fringes is related the camera settings such as the Measurement Results to the flicker frequency and the exposure. Lamps that show no Globally, 26 lamps (LED, but also frame rate of the camera. flicker or very small fluctuations will CFLs) have been tested with this The contrast between the fringes be seen with no fringes on the device under real operating is a straightforward estimation display and the recorded images. conditions (230 V 50 Hz pure

Figure 6: Luminous flux variation waveforms for CFL (1st line) and halogen lamps (2nd line) and LED lamps (3rd and 4th lines)

sinusoidal stabilized voltage feeding). some tested lamps at nominal percentage of 10% and good Figure 6 shows a sample of operating conditions. quality CFLs may reach a per cent the results. flicker of 20%. In both cases Very recently, Kukačka [14] also end-users don’t notice any flicker, While all CFL and incandescent observed that flicker behaviour thus we can consider that up to a lamps exhibit smooth flicker could change with the lamp Flicker Modulation of 20%, flicker is behaviour as expected, the tests operating time (e.g. lamp warming). acceptable (the next paragraph of show that LED lamps have different Figure 8 shows our measurements this paper contains a more detailed behaviours ranging from zero flicker, (photodiode current and its FFT) discussion of flicker perception). to high flicker content (up to 100%). for a LED lamp immediately after Thus, looking at figure 7 we Moreover, the waveforms of the switching on and 20 minutes later conclude that many products sold light output fluctuations are not as (the lamp was operating at fix today in occidental markets are smooth and symmetrical as the ambient temperature 25±0.5 °C). subject to high flicker rates at ones for CFL and halogen lamps. We suspect here that some ballast low frequencies. This last is a clear indication that components are sensitive to various LED lamps use different temperature increases; this is not The tests also show that for LED converter architectures. a good sign for ballast quality. lamps exhibiting flicker, a voltage decrease (dimming), induces, Figure 7 shows the experimental It should be noted that a 100 W in many cases, more flicker. Flicker Modulation values for incandescent lamp has a flicker Figure 9 gives the results for some

Figure 7: tested LED lamps. As can be seen, Some experimental in infrequent cases Flicker Modulation percentage flicker stays constant when voltage drops values of various (lamps n°10 and 17), then the lamp lamp technologies as obtained in LAPLACE switches off at an approximate laboratory voltage lower than 100 V.

Flicker Perception Following DoE [11], when discussing the potential human impacts of flicker, it is important to understand the difference between sensation and perception. Sensation is the physiological detection of external conditions that can lead to a nervous system response, while perception is the process by which the brain interprets sensory information.

For many years it was generally accepted that people cannot notice light flicker if its frequency is higher than the eye fusion frequency (e.g. 70 Hz). However, recent research has shown that Figure 8: fluctuations of short wavelength Light Flicker from a emissions are perceived to a LED lamp measured higher extent. For example, by the photo-induced people suffering from migraines current in the detecting photodiode are more likely to be sensitive (upper plot) and its to flicker at high frequencies [15]. FFT decomposition The main issue here is how people (lower plot) just after switching on the lamp perceive flicker and if they accept (blue line) and after it or not. This question is difficult 20 minutes of operation to answer because flicker perception (green line) is a real multi-parametric effect.

Bullough et al [16] studied the flicker perception based on two parameters: the flicker frequency and the flicker depth at constant 50% duty cycle. The visual task Figure 9: used to assess flicker perception Flicker modulation for was waving a light-coloured rod LED lamps (numbered against a dark background and from 10 to 20) under represents close to a worst-case dimmed voltage input scenario for detection of stroboscopic effects.

For rectangular waveforms operated so that the maximum light output is produced 50% of the time and the minimum light output is produced 50% of the time, the percent likelihood of detection can be estimated as follows:

Figure 10: 25 Fm + 140 LAPLACE experiment d [%] = 100 ∙ ______(4) for studying flicker f + 25 Fm + 140 perception

where ƒ is the flicker frequency (in Hz)

and Fm the Flicker Modulation (in %). This equation is applicable to frequencies from 100 Hz to 10 kHz and for Flicker Modulation values from 5% to 100%.

To assess acceptability of flicker producing noticeable stroboscopic effects, a five-point scale was used. The results suggest that even when stroboscopic effects from flicker were readily detected, they were pulses generates 540 lm and an and lower frequencies the not always judged as unacceptable. average illuminance of 167 lx on flicker was more easily detected. The acceptability rate, a, can be the background observed surface. As for the different backgrounds, estimated using the following Background surface colour can we see that with a white empirical relation: be either black or white; background the volunteer the experiment has been carried- detected flicker for a larger 4 out with both colors. During the range of frequencies. a [%] = 2 - ______(5) experiment the frequency was The fluctuations are more easily 1 + (f/fb) varied between 50 to 70 Hz and perceived with a white background the duty cycles from 20% to 90%, due to increased luminance where ƒ is the flicker frequency (Hz) in all experiments the average contrasts. Because of that we

and fb the borderline frequency (Hz) illuminance was kept constant. noted that the visual fatigue of between acceptability and the participants is also higher unacceptability, that can be The observers enter the room and with the white background than given as follows: sit 1m from the edge of the light with the black background. box. They could not directly f [Hz] = 130log(F - 73) (6) b m perceive the light source or the Furthermore, we wanted to know diffuser. We asked the observers how the volunteers perceive the Arexis et al [17] in LAPLACE to look directly at the background most extreme LED flicker cases laboratory designed a new of the box in front of their visual that we found in our tests experiment in order to understand axis and to not move their heads. (100 Hz, more than 90% flicker). the influence of the frequency We then performed 40 lighting So we ran two lamps side by combined to the duty cycle (Flicker scenarios with duration of side in two similar light boxes. Modulation depth is also studied). 5 seconds each. After each one One of them was operated in we asked the observers to push DC mode with no fluctuations Figure 11: A black box (Figure 10) was a button if they perceived or felt and the other was programmed Flicker perception (number of positive placed in a dark room and a High light flicker. No other action to run with pulses at 100 Hz answers) with black Brightness warm white LED (3000 K) was required. and a duty cycle of 50% thus background (left) white module placed behind a frosted giving us a 100% flicker. background (center) glass plate and never seen directly The number of positive answers Both black and white background and comparison between the two by the observer. The LED module, for perception of flicker is shown were used again. The average colors (right) supplied with rectangular current in figure 11. For lower duty cycles power was kept at 2 W.

Figure 12: The results were the following: Pareto optimal solutions obtained by • With both black and white Arexis. The numbers background 16 out of 16 indicate distinct “islands” in the volunteers noted that the [frequency, duty cycle] pulsed LED was brighter domain based on the 4 (Broca - Sulzer effect) zones assumption (0%, 5%, 20%, 50%) • 15 out of 16 volunteers did not perceive a light fluctuation with any background colour. More especially, 8 out of 8 did not perceive a fluctuation with the white background and 7 out of 8 volunteers did not perceive a light fluctuation with the black background (the one that Using the obtained answers a dimmable lamps within Energy Star perceived it was correct in Pareto optimality procedure requirements [20]. This is not an identifying the pulsed lamp) was used in order to define the acceptable situation that scientists • When both lamps were set to [frequency, duty cycle] areas and policy makers tend to remediate. pulse (100 Hz, 50% duty cycle) that flicker is not perceivable. with different backgrounds then Pareto optimality is a state of In the absence of flicker metrics all our volunteers agreed that the allocation of resources in which and waveforms, specifiers can white background gives a higher it is impossible to make any one pursue qualitative means for brightness impression, and two individual better off without making evaluating flicker. Specifiers should noted a fluctuation with the black at least one individual worse off. consider how the risk of flicker- background To achieve this classification Arexis related problems is heightened or used the work of Villa et al [19] reduced by a given light source, There are two important points here. based on NSGA-II genetic the type of space, its occupants, The first one is that the pulsed algorithm. Figure 12 shows an and the tasks being performed. operation of LED lamps can indeed example of results obtained by In fact, flicker effects can be give us energy savings if people this method. As can be seen, different following the executed perceive them as brighter for the the Pareto solutions are organized task by the end-user. same power input. The second in 7 distinct islands: For static tasks or tasks with point here is that although in our Island 7 corresponds to situation moderate eye motion limiting FI at visual perception experiment the 1 (no flicker perception), Islands 5 values lower than 0.5 should fine. white background made it easier for and 6 to situation 2 (almost In the case of more stringent tasks flicker to be detected, in the 100 Hz imperceptible flicker), Island 4 to implying rapid movement of objects test this was no longer relevant and the situation 3 (perceptible flicker) or visual fixation locations, FI should in fact the mobile phone camera and Islands 1 and 3 correspond be lower than 0.1. The case of and three volunteers detected it with to situation 4 (noticeable flicker). confined environments has to be the black background. This type of classification allows considered apart. choosing power supply In her PhD thesis, Arexis [18] tried characteristics in such a way in A first attempt to provide a generally to identify the combinations of order to avoid flicker and achieve accepted standard was to limit [frequency, duty cycle] that flicker energy savings. flicker passing through a maximum is not perceivable by end-users acceptable Flicker Index as a (the sample size was in the order function of frequency within the of 500 people). Flicker and Standards 100 Hz to 800 Hz range following As has seen in the above a linear relation: To achieve that study four paragraphs flicker is a very max (Fl) = 0.001 ∙ f (7) situations were defined as follows: important issue that could implead SSL technology adoption from the where ƒ is the flicker frequency • Situation 1: 0% of end-users general public. However, for the expressed in hertz. This criterion perceive flicker moment, no regulation of light leads to FI<1,2 for USA at 120 Hz • Situation 2: 5% of maximum flicker is really implemented in and FI<1,0 at 100 Hz for Europe. end-users perceive flicker various world-region regulations. The rationale used by the US • Situation 3: 20% of maximum Flicker requirements could Environmental Protection Agency end-users perceive flicker significantly improve commercial (EPA) to establish that limit appears • Situation 4: at least 50% of lamp evaluation for flicker, but in to originate from experiments on end-users perceive flicker Energy Star® only apply to high-quality conventional light

Figure 13: Flicker Modulation The proposed criteria will be depth as a function adapted to the “tier concept” of flicker frequency. proposed in the annex. Flicker modulation is These recommendations will be given for no effect level (green) and low risk published sometime in 2016 in the level (yellow) next version of Performance Criteria. More information can be found at ssl.iea-4e.org web page.

More recently, Bullough and Marcus [23] conducted a study in which different temporal flicker waveforms at several frequencies, suggested that Flicker Index was a superior metric for characterizing stroboscopic effects than percent flicker. This is going to the opposite way of the IEEE standard.

In addition, NEMA in its position sources, including CFLs, performed because it leads to the rejection paper [24] asserts that current by the US Department of Energy of “European flicker” at 100 Hz TLA standardization is hampered by (DoE); in this work, the highest FI but not “US flicker” at 120 Hz. lack of adequate TLA assessment value is 0.11 for magnetically This recommendation leads to metrics and that new flicker metrics ballasted CFLs. For the moment a Flicker Modulation up to 8.0% and associated measurement this requirement is still not included at 100 Hz and 9.6% at 120 Hz. methods for lighting are required. in the last version Energy Star It is also underlined that this could requirements [20]. It should be underlined that IEEE add unnecessary cost to the 1789 is a recommendation and in electronics gears that may be The Institute of Electrical and any case can’t be seen as a flicker not really justified. Electronics Engineers (IEEE) P1789 measurement standard. In fact, committee who has been working we don’t know how to handle Today, as per DoE advice [20], on light flicker for several years has complex waveforms with several “…in the absence of flicker metrics recently published recommendations Fourier components at different LED systems should always be based Flicker Modulation metrics [21]. frequencies. IEEE 1789 suggests visually evaluated, ideally with Figure 13 shows the proposed limits using spectral analysis based on flicker-sensitive clients. Waving a by Lehman and Wilkins [22]. Fourier transform but the “integrated finger or pencil rapidly under the It is more appropriate to refer to criterion” is still an open area of LED source, or spinning a flicker flicker based on fd as the dominant research. IEEE 1789 suggests an wheel, can expose the presence of flicker frequency (the term algebraic sum but also mentions flicker through the stroboscopic “dominant frequency” is used at other integration methods effect, even for those who are not two instances in IEEE 1789 (Minkowski norm for instance). naturally sensitive…” We should recommended practice 2 [25] but express the wish that a formal not defined in the text). Furthermore, International Energy Agency 4E-SSL world acceptable standard will be the IEEE recommendation fixes annex is now working also to adopted before “toxic”, low quality 90 Hz as cut-off frequency. establish flicker criteria inspired by products and unscrupulous sellers This may be seen as a bad choice the above IEEE recommendation. poison the market.

New Cree® XLamp® XQ-E High Intensity Color LEDs Offer Highest Candela Performance XLamp® XQ-E High Intensity LEDs are the industry’s first family of color LEDs optimized to deliver the highest candela performance. The XQ-E High Intensity LED is the smallest building block for color LED designs, allowing lighting manufacturers to boost performance and reduce size for directional applications. For additional information, please visit www.cree.com/xlamp/xqehi

References: [1] Berman S.M., D.S. Greenhouse, I.L. Bailey, R. Clear and T.W Raasch, Human electroretinogram responses to video displays, fluorescent lighting and other high frequency sources. Optometry and Vision Science 68(8), 645-662, 1990 [2] Wilkins A.J., I.M. Nimmo-Smith, A. Slater and L. Bedocs, Fluorescent lighting, headaches and eye-strain, Lighting Research and Technology, 21(1), 11-18, 1989 [3] Wilkins A.J., C. Clark, Modulation from fluorescent lamps, Lighting Research and Technology 22(2), 103-109, 1990 [4] Wilkins A.J., P. Wilkinson, A tint to reduce eye-strain from fluorescent lighting, Preliminary observations, Ophthalmic and Physiological Optics, 11, 172-175, 1991 [5] Hart WM Jr., Adler’s physiology of the eye. Clinical application. Editors R.A. Moses RA, W.M. Hart, The C. V. Mosby Company St Louis, 1987 [6] Jinno M., K. Morita, Y. Tomita Y. Toda, H. Motomura, Journal of Light and visual environment 32(2), 2008 [7] Rea, M.S. (editor). IESNA Lighting Handbook: Reference and Application, 9th edition. New York: Illuminating Engineering Society, 2000 [8] Poplawski, M., N.J. Miller, Exploring flicker in Solid-State Lighting: What you might find, and how to deal with it, Pacific Northwest National Laboratory, 2010 - http://www.e3tnw.org/Documents/2011%20IES%20flicker%20 paper%20poplawski-miller-FINAL.pdf [9] Kitsinelis S., G. Zissis, L. Arexis, A study of flicker of commercial lamps, Light & Engineering 20(3) pp 25-33, 2012 [10] Zissis G., S. Kitsinelis, L. Arexis, Etude comparative des effets stroboscopiques d’un échantillon de systèmes d’éclairage, Final Rapport, ANSES Convention 2010-CRD-22, 2011 [11] US Department of Energy, Solid-State Lighting Technology Fact Sheet: Flicker, Building Technologies Office, PNNL-SA-94791, March 2013 [12] Tähkämö L., M. Bazzana, P. Ravel, F. Grannec, Ch. Martinsons, G. Zissis Life cycle assessment of light-emitting diode downlight luminaire—a case study, International Journal of Life Cycle Assessment 18(1), DOI 10.1007/ s11367-012-0542-4, 2013 [13] Kitsinelis S., G. Zissis, Proceedings of 13th Int. Symposium on the Science and Technology of Lighting Systems, pp. 363-364, Troy NY (USA), 24-29 July 2012 [14] Kukačka L., P. Dupuis, G. Zissis, M. Kolàĭ, Retrofit LED Lamps - Photometric Flicker Analysis, Proceedings 50th IEEE IAS Annual Meeting, 18-22 October 2015, Dallas TX (USA). [15] Shady S., D.I.A. MacLeod, H.S. Fisher, Proceedings of the National Academy of Sciences (PNAS), 101(14): 5170-5173, 2004, [16] Bullough J. D., K. Sweater Hickcox, T. Klein, A. Lok, N. Narendran, Flicker Parameters for Reducing Stroboscopic Effects from Solid-state Lighting Systems, ASSIST recommends 11(1), May 2012 [17] Arexis-Boisson L., S. Kitsinelis, G. Zissis, J. Salon, S. Martin, N. Ouahhabi, Visual Perception of LED Flickering Controlled in a Colored Environment, Journal Light & Visual Environment 37(4), pp. 183-188, DOI IEIJ130000504, 2014 [18] Arexis, L. Etude et optimisation d’un système d’éclairage efficace énergétiquement et adapté aux besoins de ses utilisateurs (santé, sécurité et qualité de vie), Thèse de l’Université de Toulouse, 19 june 2014. [19] Villa C., Labayrade R., Multiobjective optimisation of lighting installations taking into account user preferences - a pilot study, Lighting Research and Technology, 45(2), pp. 176-196, DOI 10.1177/1477153511435629, 2013 [20] ENERGY STAR® Program Requirements for Lamps (Light Bulbs) V2.0-Draft 3, 4 September 2015 - http://www. energystar.gov/sites/default/files/ENERGY%20STAR%20Lamps%20V1%201_Specification.pdf [21] IEEE, Standard 1789-2015 - IEEE Recommended Practices for Modulating Current in High-Brightness LEDs for Mitigating Health Risks to Viewers, June 2015 [22] Lehman B., A.J. Wilkins, Designing to Mitigate the Effects of Flicker in LED Lighting: Reducing risks to health and safety, IEEE Power Electronics Magazine, pp. 18-26, DOI 10.1109/MPEL.2014.2330442, September 2014 [23] Bullough, J. D., D. Marcus, Influence of flicker characteristics on stroboscopic effects. Lighting Research and Technology, doi: 10.1177/1477153515599566, 2015. [24] NEMA, Temporal Light Artifacts (Flicker and Stroboscopic Effects), A NEMA Lighting Systems Division Position Paper, 15 June 2015 - https://www.nema.org/news/Pages/NEMA-Lighting-Systems-Division-Publishes- Position-Paper-on-Temporal-Light-Artifacts.aspx