Blue Sky + Corrected spectrum White LED fixture

Fixing terrible spectrum (missing cyan&green) of generic white LEDs to get perfect flat spectrum and impression of blue sky in the room.

Similar projects worth following
It is known that White LEDs produce very poor spectrum (=color rendering) because they are effectively blue LEDs with yellow phosphor (missing cyan and half of green wavelengths - right where our eye is the most sensitive). This is probably the last major show stopper for complete LED dominance.

I've fixed poor spectrum of white LEDs by adding 495nm cyan LEDs, 520nm green LEDs, and 460nm blue LEDs in first iteration. In second iteration I've used 470nm blue instead of 460nm ones to get even better and flatter spectrum. Now I believe it has better spectrum than one can get from best ceramic metal-halide high-intensity discharge lamps.

I've placed spectrum correcting LEDs on the top of the fixture, so that they illuminate the ceiling, giving "blue sky" effect. Your limited RGB display (as well as my limited RGB digital camera) just cannot pass you how beautiful this color is - very tender, mild bluish/cyanish. It literally brings smile to my face every time I turn it on ☺

Many years ago I wanted to have really bright light. I went way too far with 100'000 lumen HID fixture. It was so bright that I eventually left only 1 out of 7 lamps on which gives "mere" 14700 lumen. The light was nice and white, but it was far from perfect: it required 5 minutes cool-down to reignite it, 2 minutes warmup and was frying insects... It's spectrum had no gaps, but several distinct peaks (see first update for spectrum).

That is when I started thinking about switching to LED with corrected spectrum.

Our eye is the most sensitive in the green region, that is why having corrected spectrum in this area is essential for comfort of our eye. White LEDs give you alot of blue and orange-red colors, but not cyan-green ones. Advertised CRI index is misleading, as it is calculated by precision of rendering of a small, fixed set of colors. This is probably the last obstacle to LED widespread penetration including areas requiring perfect color rendering. Some people just don't like the hole in the green-cyan.

I've decided to try various combinations of extra LEDs to fill in the gap. Here is how generic White LED spectrum looks like (grey), and how I managed to correct it (blue):

Final correction details are as follows: 40W 4000K white LED, 10x0.7A 495nm cyan, 5x0.5A 470nm blue, 5x0.5A 520nm green.

  • 10 × 3W 495nm Cyan LED Main horse power of spectrum correction. Hence largest power goes here.
  • 5 × 5W RGB LEDs Second revision used green LED from here. Third revision will use red & blue for special effects.
  • 5 × 3W 470nm Cree Blue LEDs To fill in the gap between 450nm peak of white LED and 495nm cyan.
  • 3 × XL6990 step-up modules. One for cyan will require heatsink
  • 1 × 12V 120W Meanwell PSU Meanwell is considered "good" chinese PSU with proper filtering and no corners cut.

View all 7 components

  • New LEDs arrived!

    Mikhail Svarichevsky08/17/2015 at 20:43 0 comments

    With new 490nm and 470nm LEDs I can finally proceed:

    470nm LED came close to specs:

    With 10x0.7A for Cyan LEDs step-up regulator is taking too much heat, so I had to add few extra radiators:

    After fine-tuning current in the LED chains I've got the final spectra:

    Finally there are no missing colors - this is something which could satisfy me. And "blue sky" effect is really nice. As we were mostly adding LEDs near to eye (and camera) maximum sensitivity, light feels much brighter than before (as well as photo exposures).

    Is there a visible difference? Yes, but not looking at spectrum analyzer it is hard to show you on the RGB display with RGB camera. I have cyan shirt for example, and it's color is visibly different under new lighting - but making and comparing a photo won't show much - camera renders it bluish, while in reality it is slightly greenish (regular LED on the left, corrected LED on the right) :

    Personally, I am more than satisfied by light itself and I like it much more than my older 100k lumen HID fixture.

    Future plans:

    This is not the end of the journey though - I ought to replace step-up converters with ballast resistors with a proper constant current ones and make it RF-controlled (nRF24LE01). I am going for nRF instead of hiped WiFi ESP8266 because I personally cannot tolerate wakeup latency.

    Instead of using standalone constant-current ICs I am going to drive mosfets from uC PWM pins while monitoring current. As there are no fast current changes this should do the job with very simple schematic.

  • MOAR LEDs!

    Mikhail Svarichevsky08/17/2015 at 20:24 0 comments

    So I needed to add more LEDs. Apparently just 5x cyan LEDs cannot cover the hole left by 40W white ones, so I used all 10 cyan I had on first fixture (initially I planned to use 5 on each), and set them to maximum rated current (700mA), tuned 460nm blue die form RGB LEDs to get as flat as possible, and the best I got was this:

    Much better than original white LEDs, but there is still a hole near 470nm, not as terrible as before though. It was clear this is as good as one can get with LEDs I had. If I increase current on any of them - I'll get overshoot. So I needed 470nm blue LED instead of 460nm one to fill in that hole. I ordered LEDs and started to wait...

  • Base LED fixture

    Mikhail Svarichevsky08/17/2015 at 20:13 0 comments

    I really liked simple geometrical forms, so I got square 40W 4000K LED fixtures. They were relatively inexpensive (~40$ each), their large area ensures decent cooling and long life. I disassembled one to figure out if it's possible to stick LEDs inside - but there is just too little space left for high-power LEDs.

    Fixture design is quite clever, like on LED display back light: There is acrylic light-guide with diffuse dots across the area to make it uniformly lit. light-guide is lit from the side.

    After fixing minor issues (like black marker marking on the edge,tighter assembly to improve thermal resistance) I assembled it back. It was clear I can only add LED on the back, facing to the ceiling. That might be interesting effect though. I glued first LEDs using ALSIL-5 glue and generic "heatsink plaster".

    For first revision I decided to use step-up regulators with ballast resistors to control the current over LEDs. I set the current in "hot" state. Surely in the future I will need to replace it with digitally controllable constant current sources.

    First tests shows that with 5x cyan alone spectrum is very far from desired. I needed more LEDs, so I added RGB ones, intending to use green and blue dies.

    Current adjustment near the ceiling feels a bit tricky:

    With green on I've got the following spectra:

    There is definitely a progress, but still very far from desired.

  • First spectrums

    Mikhail Svarichevsky08/17/2015 at 20:04 0 comments

    This all started with a borrowed spectrum analyzer (Thorlabs CCS100, 350-700nm, <1nm resolution). I started measured spectrum of every LED and light source I had around... :

    Daylight spectrum:

    Just blue light from sky (not including Sun) - we can see it is blue-cyan:

    My nuclear light fixture HID lamps were quite funny. Right after turn on - lots of mercury lines:

    Heating up...

    And it's done:

    Cyan LED:

    Blue 460nm (from RGB):

    Green 520nm (from RGB):

    Red (from RGB):

    And finally white LED:

View all 4 project logs

Enjoy this project?



Ruben P wrote 05/27/2018 at 08:18 point

Great project ! I totally want to do something like that. I'm more interesed by an indirect lighting (reflection on ceiling) of all the spectrum. Do you thnik something similar could be achieved with led strips ?

  Are you sure? yes | no

DeepSOIC wrote 08/20/2015 at 18:45 point

I've had a similar idea a while ago, but haven't made anything =) thumbs up!

  Are you sure? yes | no

zakqwy wrote 08/18/2015 at 17:55 point

Very cool--this has been one of my holdups on fully adopting LED lighting as well. Think someone else could use your settings and get a decent result without having access to a spectrometer (which I have just realized I desperately need)?

  Are you sure? yes | no

Mikhail Svarichevsky wrote 08/19/2015 at 04:25 point

Yeah, the idea is that this should be reproducible without spectrum analyzer, as long as same color temperature is used (4000K) and LED currents/quantity are scaled according to white LED power. 

  Are you sure? yes | no

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates