First thing I wanted to do was see what I needed to produce good color. My very first look was to use a philips 120v LED PAR38 spot to get an idea of color and number of LEDs/lumens I would ultimately need. This test showed that a cool white alone washed out the color of my fish (white cloud mountain minnows) but provided good general light and excellent shimmer.
After looking at various options, I decided to use the LedEngin white. I liked their glass primary optic, spectrum, factory-soldered star (reliable thermal data), and long lumen maint. projection.
After that, I ordered a selection of colors to span the entire visible spectrum.
I built a couple simple test rigs. The first was for a single star-mounted LED powered by C-cell batteries. The second was for 4 stars powered by a PC power supply and allowed dimming for each of the four channels so i could check color mixing. For the single, I used a large northbridge heatsink. For the 4-channel rig, I used a large CPU heatsink.
I tested the following colors:
460nm (royal blue)
490nm (baby blue)
5500k (cool white)
4100k (neutral white)
660nm (deep red)
735nm (far red)
Both of the ultraviolets produced objectional flourescence of the particulates and/or organisms floating in the water column so I did some research to see how important that part of the spectrum was to both plants and fish. It seems that UV-A is not terribly important and may even be harmful, so I elected to discard that portion of the spectrum.
The royal blue, surprisingly to me, induced a fair amount of flourescence in my fish, but it was a redunant part of the spectrum. All the white LEDs have a spike in the 450nm range and adding this to a white was just way too blue.
The 490nm blue proved a bit too blue to properly fill the response valley of the LedEngin whites. The 505nm works good color-wise, but the viewing angle is darn near 180 degrees. A wide-angle (90°) optic is required.
Based on spectrum data comparisons I've done, I suspect that the 490nm would pair up nicely with a Cree white LED since they appear to be a bit more biased toward the deeper blues.
The 660nm red is a true "fire engine red", not one of the more pale red-oranges commonly found and labeled as "red". It is also centered on a photosynthetic response peak and really adds a lot of punch both to PAR readings and my fish's red colors.
The 735nm red is pushing the edge of infa-red as you can feel the heat on your skin and the light produced is very dark red and quite dim. since this color is present in sunlight I decided to add it even though most articles I read indicated that it was responsible for stem elongation. I may reduce or remove this piece of the spectrum at a later time depending on plant response.
Comparing a Cree 4000 kelvin white to a LedEngin neutral white:
(the purple and red vertical lines indicate the practical limits of photosynthetically usable wavelengths. The light gray trace is a response curve of overcast natural daylight (6500°k))
Estimated effect of adding colors to achieve more even resonse: