LEDs and color mixing
BobS
case to produce any color. I brought a Supper bright Blue (3.5V, 20mA,
475nm) LED, it's in a clear case but turned on it looks more white then the
deep violet blue I was expecting.
Q: What wave lengths should I use to correctly mix RGB colors together. e.g.
The reddest red, bluest blue and so on.
BobS
Jonathan Kirwan
wrote:
>I'm trying to mix Red, Green and Blue LEDs through a defuse white plastic
>case to produce any color.
You may wish to try paraffin wax. It's a much better scatterer, I
believe, and it does so without much absorption.
>I brought a Supper bright Blue (3.5V, 20mA,
>475nm) LED, it's in a clear case but turned on it looks more white then the
>deep violet blue I was expecting.
>
>Q: What wave lengths should I use to correctly mix RGB colors together. e.g.
>The reddest red, bluest blue and so on.
LEDs at 466nm, 525nm and 626nm make a reasonable mix, giving good
control over CIE chromaticity.
Jon
BobS
mixing results?
BobS
"Jonathan Kirwan" <jki...@easystreet.com> wrote in message
news:p8cumu8c4dtgrsofd...@4ax.com...
Jonathan Kirwan
wrote:
>Thanks for that... another question.. does equal mcd luminosity give equal
>mixing results?
Well, I'm not exactly sure what you are asking, here. But no matter
what it is, the answer is probably no.
First, there are two vision systems; scotopic and photopic. One is
pretty much swamped out in normal lighting, that's the scotopic (rods,
or non-color) vision. It only comes into play in low lighting
situations. The other is the photopic or color vision. That's what
you will be referring to, I imagine.
Photopic brightness is a subjective phenomenon. You and I can talk
about things which are about as bright as each other (if we use
special flicker comparators, for example, or disappearing filaments)
or if one is brighter than another. But it's hard for us to correctly
talk about one thing being "twice as bright" as another. Quantitative
comparisons just aren't a part of subjective senses. This is typical
of many human sensations, such as the loudness of a sound or the
hotness of water.
mcd is, in fact, is related to lumens which are half physical, half
subjective. They can be used to compare brightness in the sense of
equal, greater than, or less than. But that's about it. Twice the
mcd is not twice as bright. Our visual system adapts readily to
sunlight at thousands of ft-cd and to a room at just a few ft-cd, as
well. It couldn't do this if our sensation response was linear.
Color is yet another thing. If you choose to use the paraffin wax to
mix the emissions as I suggested, computing the perceived color is
handled by computing the X filter, Y filter, and Z filter values of
the emissions and then normalizing these to develop an x and y which
is then plotted on the CIE chromaticity diagram and then related to a
hypothetical "white point." A line is drawn from the white point
through the plotted (x,y) point and mapped to the CIE curve to figure
out the hue. People's ability to discern hue varies, and what one
decides to call blue or green differs on training and sensation, but
in general if two different (x,y) points yield the same hue and the
white point was correctly chosen, then the two colors will be decided
by the same individual as being similar.
If you want to be able to compute perceived color from milliamps of
drive on each LED, it's going to take some calibration work to get
there.
What are you intending to do?
Jon
BobS
pixel, for color matching.. It works as a device you plug into your PC, like
a light box, driven by software on the PC. So I want to reproduce colors on
my computer desktop basically.
BobS
"Jonathan Kirwan" <jki...@easystreet.com> wrote in message
news:fbbvmuc0tof682e0i...@4ax.com...
Bob Wilson
says...
The last company I worked for is making LED luminaires used to light up the
sides of buildings using Lumiled's Luxeon LEDs. Colour mixing with LEDs is a
big problem, because of the poor characterization of their exact output
wavelength. In short, you need the deepest (longest wavelength) red, and the
shortest wavelength blue possible. Blue LEDs are not violet, but should be a
deep intense hard blue. Whereas red LEDs can be had that are a real deep
red, blue ones are more troublesome. Even the slightest hint of "cyan" in
the colour dramatically reduces the abiltiy to produce proper magenta, for
example. One of the biggest problems is that blue LEDs in particular are not
tightly toleranced in their output wavelength. From one LED to the next,
they vary quite significantly in output wavelength. We had to actually get
Lumileds to colour bin the LEDs in order to get any mixing consistency at
all.
Bob.
Jonathan Kirwan
(Bob Wilson) wrote:
><snip>
>tightly toleranced in their output wavelength. From one LED to the next,
>they vary quite significantly in output wavelength. We had to actually get
>Lumileds to colour bin the LEDs in order to get any mixing consistency at
>all.
I know. I actually develop systems to do color and intensity binning
for LED parts and systems. :)
Jon
Jonathan Kirwan
wrote:
>I'm trying to make a reasonable approximation of a single computer screen
>pixel, for color matching.. It works as a device you plug into your PC, like
>a light box, driven by software on the PC. So I want to reproduce colors on
>my computer desktop basically.
Color matching? How does the operator match colors up, here? Are you
driving the LEDs and asking the operator to do something with the
screen? Or driving the screen and asking the operator to do something
with the LEDs? Or??
Jon
Don Klipstein
Bluest blue: Dominant wavelength in the 450's to maybe as high as
around 460 nm, with peak wavelength in the 440's to low 450's of nm.
Avoid ones with dominant wavelength 465 nm or more or either wavelength
near or under 430 nm.
Greenest green: Preferably dominant wavelength in the 530's of nm, but
525 and 527 are common. Peak wavelength should be around 520-530 nm or
maybe as much as low 530's (closer to 520 is more common)
Reddest red: Dominant wavelength near or in the 636-670 nm range, peak
wavelength mid-640's to 680. Longer is better, but expect uselessly low
brightness and/or really large bandwidth extending into (or through and
even past) orange if the peak wavelength is in the 690's or more or the
dominant wavelength exceeds 650 nm.
- Don Klipstein (d...@misty.com, http://www.misty.com/~don/ledx.html)
BobS
screen is reproduced by the device.
BobS
"Jonathan Kirwan" <jki...@easystreet.com> wrote in message
news:d7f0nuk9prdujmbfm...@4ax.com...
Don Klipstein
>Thanks for that... another question.. does equal mcd luminosity give
>equal mixing results?
NO!
To get closer, have the blue photometric output equal that of green
times the blue's ratio of Y to Z CIE chromaticity coordinates of blue
(ratio is likely 1/10 or less). And red should have photometric output
about that of green times the red's ratio of X to Y CIE chromaticity
coordinates - about 1/2 that of green or less. ALL OF THIS IS
ROUGHLY-ROUGHLY!
- Don Klipstein (d...@misty.com)
Jonathan Kirwan
The peak wavelength is probably obvious, being the wavelength at the
peak of the radiated spectrum.
But the dominant depends not only on the LED spectral emission as a
whole, but also on the chosen illumination assumed as an additive mix.
In LEDs, the manufacturers I've worked with before often use what is
called the D65 white point to represent the ambient lighting source,
roughly like an overcast daylight sky (if I recall) in establishing
the dominant. LEDs have very good color purity and therefore plot
very close to the CIE color curve perimeter, so the effect that the
assumed illumination has on the dominant isn't so strong. But it does
have some. So it's possible that two different manufacturers may
assume different illumination and thus a different resulting dominant,
even when the LEDs would appear much closer than the numbers indicate,
had the illumination been assumed similarly by them.
The dominant takes into account the entire LED spectrum, so it's
probably a better choice when selecting colors. The peak is usually
used when looking at transmission through an optical filter, for
example, because it gives an easy way to estimate aggregate losses by
looking at the filter's response curve at the peak.
Regarding the better dominants to use, that really does depend on the
application.
Without knowing the application at all, one might argue for maximum
coverage of the interior of the CIE chromaticity area. That would
suggest selecting three points to make a triangle enclosing the most
area. For that, the red should be anything longer than about 625nm or
so, as your eyes have poor wavelength discrimination past that and all
the dominant hues from 630 to 700 (past which you don't see, anyway)
are bunched up close. Anything more than about 650nm won't matter.
The green should be optimally around 520nm or so, if interior area is
the prime goal, but a little longer is okay say to 530nm. On the blue
end, 470nm isn't bad at all (and is close to the phosphors used in
TV.) There's not much shorter to worry about, but if you push it
shorter, 450nm will be as much as matters. (Even 380nm isn't far away
on the curve, from there.)
But knowing that the application relates to computer screens suggests
that there are other optimal choices. NTSC recommended an R of
(.67,.33), a G of (.21,.71), and a B of (.14,.08) back in 1953. But
what is used now is an R of (.68,.32) using europium yttrium vanadate,
a G of (.28,.60) using zinc cadmium sulfide, and a B of (.15,.07)
using zinc sulfide. These cover only about 1/2 the area within the
perimeter of the CIE diagram.
The R phosphor produces a rather pure color which is very close to a
dominant of about 610nm. This might be particularly surprising, if
you looked at the emission spectrum, which shows a peak around 670nm.
But the broad emissions and the fact that human red response has its
peak wavelength shorter than 600nm explain this result. So selecting
a red LED at 626nm is certainly just fine. Anything longer will end
up wasting energy in an area where one's eyes just don't have much
response. It's pointless.
The G phosphor produces an impure color, which is arguably close to
545nm (depends on where you imagine the white point to be.)
Certainly, a 545nm LED would enclose the monitor's phosphor triangle
for this purpose. Anything from about 525nm to about 545nm is just
fine, really. Closer to 545nm would set up the triangle just a bit
closer to the monitor. 545nm is also about the peak of our eye's
green response.
The B phosphor is close to pure and roughly dominant at 460nm. It
would seem to me that selecting LED dominants at about 466nm, as I
mentioned, strikes the best balance between catching the triangle side
to red and to green at the same time. 470nm is okay. Anything much
shorter than about 450nm or so and it starts cutting off the B-G side
of the phosphor triangle a bit (depending on the G LED used.) Not
noticeably, but there's no point to doing that.
Super red (standard ones) LEDs have dominants in the area of 622nm to
634nm, depending on batch. Super-reds are usually just called 628nm.
Those are just fine. So are the hyper-reds, with a spec'd dominant
usually around 633nm. Longer is just a waste, as I said. InGaN true
greens (not pure greens) are given at a dominant of 528nm and are just
fine. The pure green types are more like 560nm and just "green" are
570nm. Those are too long and won't be good for RGB work. Both the
GaN blues (465nm) and InGaN blues (470nm) are both perfect for RGB, I
believe, though the GaN types might be just a bit better color-wise.
If that helps any.
Jon
Jonathan Kirwan
wrote:
>Never mind the details. The only important thing is that a color on the
>screen is reproduced by the device.
Then I've already answered as best I can, elsewhere.
Jon
Jonathan Kirwan
<jki...@easystreet.com> wrote:
>If that helps any.
Oh, forgot to add... it turns out that my original recommendation of
466nm, 525nm and 626nm from recollection were about on target for the
application space -- I just don't make guesses any luckier than that.
Jon
BobS
"Jonathan Kirwan" <jki...@easystreet.com> wrote in message
news:4i11nu8hrnjolac6a...@4ax.com...
N. Thornton
> I'm trying to make a reasonable approximation of a single computer screen
> pixel, for color matching.. It works as a device you plug into your PC, like
> a light box, driven by software on the PC. So I want to reproduce colors on
> my computer desktop basically.
You may need to take into account the gamma of the monitor too.
Monitors vary from one to another as well. Ballpark CRT monitor gammas
are 1.5 to 2.2, very roughly.
Regards, NT
Don Klipstein
wrote in part:
> Both the GaN blues (465nm) and InGaN blues (470nm) are both perfect for
> RGB, I believe, though the GaN types might be just a bit better
> color-wise.
GaN types are whiter than InGaN types.
InGaN comes in a variety of wavelengths, although 470 nm is the most
common. Lumileds makes InGaN blues with nominal dominant wavelengths of
either 470 nm or 455 nm. Cree makes InGaN chips used by various LED
manufacturers in nominal dominant wavelengths of 470 and 460 nm.
GaN is usually a Cree chip with a peak wavelength around 428-430 nm and
a dominant wavelength supposedly 465-466 nm, although it looks more like
460 even to me. There is an obsolete Nichia GaN chip with a peak
wavelength of 450 nm and a dominant wavelength close to 470 nm. Both are
significantly whitish.
- Don Klipstein (d...@misty.com)
Jonathan Kirwan
Klipstein) wrote:
>In article <c0p0nu077p5n8rtvc...@4ax.com>, Jonathan Kirwan
>wrote in part:
>
>> Both the GaN blues (465nm) and InGaN blues (470nm) are both perfect for
>> RGB, I believe, though the GaN types might be just a bit better
>> color-wise.
>
> GaN types are whiter than InGaN types.
>
> InGaN comes in a variety of wavelengths, although 470 nm is the most
>common. Lumileds makes InGaN blues with nominal dominant wavelengths of
>either 470 nm or 455 nm. Cree makes InGaN chips used by various LED
>manufacturers in nominal dominant wavelengths of 470 and 460 nm.
Thanks. That's interesting.
> GaN is usually a Cree chip with a peak wavelength around 428-430 nm and
>a dominant wavelength supposedly 465-466 nm, although it looks more like
>460 even to me. There is an obsolete Nichia GaN chip with a peak
>wavelength of 450 nm and a dominant wavelength close to 470 nm. Both are
>significantly whitish.
Hmm. Any idea what's causing the whitish nature? Are they
intentionally including some fluorescent materials? (Sounds like
these aren't close to the CIE perimeter, then.)
Jon
Don Klipstein
Both have much wider rated spectral bandwidth around 70 nm than the
usually 25-30 or so nm of InGaN blue. The obsolete Nichia ones just
had a wider emission band. The Cree GaN chips have a distinct 428-430 nm
peak superimposed upon a very wide band that spreads out over most of the
visible spectrum.
No, there are no fluorescent materials in these - they just have some
wide emission bands.
Another broadband LED chemistry is red GaP (GaP doped with ZnO - GaP is
usually yellowish green). The nominal peak wwavelength is usually 690 or
697 nm, but the emission band extends so much towards orange that the
dominant wavelength from just this main band is usually closer to 640 nm.
And there is a secondary band in the yellowish green, more apparant at the
higher currents where the wide red band experiences impaired efficiency.
This chemistry is often referred to as "low current red" - the color is
more red at lower currents, and overall luminous efficacy is maximized at
usually .5-1 mA and significantly decreases as current increases past 3-5
mA. The dominant wavelength can be as low as in the upper 590's at 30 mA,
which these LEDs can usually survive long-term.
Jonathan Kirwan
Klipstein) wrote:
>Both are
>significantly whitish.
By the way, the only blues I've actually spent time testing were from
OSRAM and their dominant was around 470nm (x=.13,y=.06) They were
rather pure in hue, though. Close to the perimeter. Not whitish at
all, looking at them. An almost stunningly sharp blue, in fact.
(I have a spectrophotometer set up with merc-argon to calibrate the
lambdas and calibration-standard tungsten lamps to get the pixel
amplitude responses calibrated, as well. Also wrote the software to
process the CIE filtering, normalization, and so on. Nice set up,
actually.)
Jon