3 ms·

Not all filament designs have horrendous flicker—we are selling one that we designed with 8% worst case, usually more like 2-3% [0][1]. That's better than incan

Not all filament designs have horrendous flicker—we are selling one that we designed with 8% worst case, usually more like 2-3% [0][1]. That's better than incandescent. Even the normal non-filament LED bulbs can have high levels of flicker. I found one product claiming to be "healthy" that had around 55% flicker. Most of the bulbs in The Home Depot are going to flicker a lot, especially the filament LEDs. [0] Bedtime Bulb: Low-blue, low-flicker light bulb https://bedtimebulb.com/ https://bedtimebulb.com/ [1] When I say percent flicker, I am referring to the modulation percentage. A good primer is this article: https://www.dial.de/en/blog/article/ieee-1789-a-new-standard-for-evaluating-flickering-leds/ https://www.dial.de/en/blog/article/ieee-1789-a-new-standard...

nkurz 7y ago
Great article. What are the design decisions that affect the amount of flicker? Is it solely based on electronically smoothing out the input voltage, or are there other factors? For example, are there phosphors that are better at smoothing the white light even if the blue is modulated?
- yeutterg 7y ago
  Adding capacitance to a "linear" driver design is one way to reduce modulation, with some tradeoffs. There are other power supply designs, such as switch-mode and several variations of "constant current" drivers, which could reduce modulation at a higher cost. Adjusting the frequency also plays a role in mitigating perceived flicker. In higher-end drivers, you'll often see a hybrid approach, such as constant current between 5 and 100% and PWM below 5%. Drivers with deep dimming down to 0.1% are just using 10-bit PWM. This would be difficult to do in a bulb form factor. I haven't seen any meaningful research into the persistence of various phosphors. It's an interesting thought. My guess is that there could be some effect, but usually we would just try to improve the power electronics.
  - jacobush 7y ago
    I heard IKEA uses the phosphor
    - yeutterg 7y ago
      Their non-dimmable filament bulbs (silver base) tend to be pretty low flicker, but I found the dimmable ones (gold base) have upwards of 30% flicker, even at full output. Without having done a teardown, I suspect the flicker is due to the power electronics.
waterhouse 7y ago
Regarding the Bedtime Bulb, I see the graph of wavelength vs ... the vertical axis isn't labeled, but I assume it's roughly "amount of light at that wavelength": https://static1.squarespace.com/static/59c9469ae5dd5bc276433d46/t/5c452608032be40b01274537/1548035614376/Bedtime+Bulb+Traditional+Source+Comparison.png https://static1.squarespace.com/static/59c9469ae5dd5bc276433... Two things come to mind, looking at it: 1. The total area under each different bulb looks wildly different. I think normal user behavior would be to get enough of the bulbs to light up a room to roughly a certain brightness; therefore, it's the ratio of blue/green light to other light that matters, and a proper comparison would normalize the total brightness of each bulb. (Otherwise, if bulb A was the same as bulb B except half the brightness, the line on the graph would show bulb A producing impressively less blue light than B.) Is there a reason this normalization wasn't done? 2. The levels of green and blue light, while lower than the comparison 2700K LED, are still far higher than zero. Why not make them zero or near-zero? Is this difficult? Is there research showing that "somewhat less blue" is significantly better? My impression is that humans' perception of brightness is mostly logarithmic, which means 0.5x is a lot closer to x than to, say, 0.1x or 0.01x; I don't know if this carries over to the effect on melatonin production, but my prior hypothesis would lean in that direction.
- Sharlin 7y ago
  The unlabeled vertical axis makes it difficult to tell how these are perceived, given the very nonuniform spectral sensitivity of the human eye. As a rule of thumb, it takes five times the power for predominantly blue light, and twice the power for red light, to appear perceptually as bright as green light.
  - yeutterg 7y ago
    My understanding is that it's still up for debate how linear or nonlinear melanopic sensitivity is, which is the main thing we are attempting to show in the plot.
- yeutterg 7y ago
  The main purpose of this graph is to compare "melanopic" light, which is light in the blue-green area between about 400 and 580 nm (the gray curve). This is the sleep-disturbing blue light everyone is talking about, which also includes green, as you remarked. The vertical scale is just relative power from 0-1, for the relative amount of light at each wavelength, as you said. Why not normalize? Because those other sources are twice as bright. We're trying to compare to a light source people would typically buy. Yes, you could use a ~40 watt incandescent at about the same lumen output, but it would still have a higher melanopic effect. The melanopic ratio for incandescent is usually around 0.50-0.54, which ours is 0.39. You'll also notice that we did not include a daylight LED in our comparison, which is something our competitors do, but we think it's just a ridiculous exaggeration of the effect. In other words, if you pointed a spectrophotometer at the 4 light sources at a consistent distance and did not normalize, that is roughly the plot you would get. If we take reading as a use case, there are aspects of our design which enable you to use a source that is half as bright and still read just as well. For one, the CRI and R9 (red color rendering) are very high, basically at incandescent levels. This makes the printed page look better. The beam angle is much wider than a normal LED (most LED bulbs mounted upright will shoot all the light at the ceiling, not where it's needed). And most people simply don't need a 60 Watt bulb (650-800 lumens) to read with, even if they don't yet realize it. Why not eliminate all blue and green? It's certainly possible, and there are bulbs that do that. The problem is that blue/green-free light is very orange/yellow, and in our testing, most people found the light undesirable and impractical. Our theory is that people would go back to their old higher-blue lighting after some time. In other words, people like white light. However, we are working on a mostly melanopic-depleted source with a slightly different use case. The point is that it comes down to both spectrum and illuminance (brightness). Everyone else in the market is only considering spectrum. There is an explicit tradeoff between color rendering (CRI, R9, Rf, Rg) and the amount of blue/green reduced. If you can balance those two, and then give someone between 100-200 lux, as opposed to 300+ lux for typical reading light, the melanopic input is much lower and the light is more comfortable to use for most people. Our light will give you about 100-200 lux in a shaded or unshaded lamp at a typical reading distance. But it's hard to explain that from a marketing perspective.

nkurz 7y ago

Great article. What are the design decisions that affect the amount of flicker? Is it solely based on electronically smoothing out the input voltage, or are there other factors? For example, are there phosphors that are better at smoothing the white light even if the blue is modulated?

yeutterg 7y ago
Adding capacitance to a "linear" driver design is one way to reduce modulation, with some tradeoffs. There are other power supply designs, such as switch-mode and several variations of "constant current" drivers, which could reduce modulation at a higher cost. Adjusting the frequency also plays a role in mitigating perceived flicker. In higher-end drivers, you'll often see a hybrid approach, such as constant current between 5 and 100% and PWM below 5%. Drivers with deep dimming down to 0.1% are just using 10-bit PWM. This would be difficult to do in a bulb form factor. I haven't seen any meaningful research into the persistence of various phosphors. It's an interesting thought. My guess is that there could be some effect, but usually we would just try to improve the power electronics.
- jacobush 7y ago
  I heard IKEA uses the phosphor
  - yeutterg 7y ago
    Their non-dimmable filament bulbs (silver base) tend to be pretty low flicker, but I found the dimmable ones (gold base) have upwards of 30% flicker, even at full output. Without having done a teardown, I suspect the flicker is due to the power electronics.

waterhouse 7y ago

Regarding the Bedtime Bulb, I see the graph of wavelength vs ... the vertical axis isn't labeled, but I assume it's roughly "amount of light at that wavelength": https://static1.squarespace.com/static/59c9469ae5dd5bc276433d46/t/5c452608032be40b01274537/1548035614376/Bedtime+Bulb+Traditional+Source+Comparison.png https://static1.squarespace.com/static/59c9469ae5dd5bc276433... Two things come to mind, looking at it: 1. The total area under each different bulb looks wildly different. I think normal user behavior would be to get enough of the bulbs to light up a room to roughly a certain brightness; therefore, it's the ratio of blue/green light to other light that matters, and a proper comparison would normalize the total brightness of each bulb. (Otherwise, if bulb A was the same as bulb B except half the brightness, the line on the graph would show bulb A producing impressively less blue light than B.) Is there a reason this normalization wasn't done? 2. The levels of green and blue light, while lower than the comparison 2700K LED, are still far higher than zero. Why not make them zero or near-zero? Is this difficult? Is there research showing that "somewhat less blue" is significantly better? My impression is that humans' perception of brightness is mostly logarithmic, which means 0.5x is a lot closer to x than to, say, 0.1x or 0.01x; I don't know if this carries over to the effect on melatonin production, but my prior hypothesis would lean in that direction.

Sharlin 7y ago
The unlabeled vertical axis makes it difficult to tell how these are perceived, given the very nonuniform spectral sensitivity of the human eye. As a rule of thumb, it takes five times the power for predominantly blue light, and twice the power for red light, to appear perceptually as bright as green light.
- yeutterg 7y ago
  My understanding is that it's still up for debate how linear or nonlinear melanopic sensitivity is, which is the main thing we are attempting to show in the plot.
yeutterg 7y ago
The main purpose of this graph is to compare "melanopic" light, which is light in the blue-green area between about 400 and 580 nm (the gray curve). This is the sleep-disturbing blue light everyone is talking about, which also includes green, as you remarked. The vertical scale is just relative power from 0-1, for the relative amount of light at each wavelength, as you said. Why not normalize? Because those other sources are twice as bright. We're trying to compare to a light source people would typically buy. Yes, you could use a ~40 watt incandescent at about the same lumen output, but it would still have a higher melanopic effect. The melanopic ratio for incandescent is usually around 0.50-0.54, which ours is 0.39. You'll also notice that we did not include a daylight LED in our comparison, which is something our competitors do, but we think it's just a ridiculous exaggeration of the effect. In other words, if you pointed a spectrophotometer at the 4 light sources at a consistent distance and did not normalize, that is roughly the plot you would get. If we take reading as a use case, there are aspects of our design which enable you to use a source that is half as bright and still read just as well. For one, the CRI and R9 (red color rendering) are very high, basically at incandescent levels. This makes the printed page look better. The beam angle is much wider than a normal LED (most LED bulbs mounted upright will shoot all the light at the ceiling, not where it's needed). And most people simply don't need a 60 Watt bulb (650-800 lumens) to read with, even if they don't yet realize it. Why not eliminate all blue and green? It's certainly possible, and there are bulbs that do that. The problem is that blue/green-free light is very orange/yellow, and in our testing, most people found the light undesirable and impractical. Our theory is that people would go back to their old higher-blue lighting after some time. In other words, people like white light. However, we are working on a mostly melanopic-depleted source with a slightly different use case. The point is that it comes down to both spectrum and illuminance (brightness). Everyone else in the market is only considering spectrum. There is an explicit tradeoff between color rendering (CRI, R9, Rf, Rg) and the amount of blue/green reduced. If you can balance those two, and then give someone between 100-200 lux, as opposed to 300+ lux for typical reading light, the melanopic input is much lower and the light is more comfortable to use for most people. Our light will give you about 100-200 lux in a shaded or unshaded lamp at a typical reading distance. But it's hard to explain that from a marketing perspective.