I'm taking some time off from this project, and mostly wanted a placeholder so I can remember where I left off - I have a bad habit of "taping over" memories from suspended projects. So, this is just a jumble of thoughts I'll update as they come until I start this thing back up.

The Energizer Cylindrical Lithium Primary Battery Applications Manual has some great info about this unique battery chemistry, and discusses the two-stage discharge seen in the graphs, temperature performance, etc. The only thing I don't like about these batteries is the need to use spring contacts, which are notoriously unreliable. I would like to spot-weld tabs on them. A homebrew spot-welder may be in my future. I guess I should ask Energizer if they produce tabbed versions first.

The previous lumens/watt calculations I did for the Luxeon Z LEDs use the datasheet-reported output at 500mA, so they're a valid comparison assuming the three LEDs share a similar efficiency curve, which is doubtful. I need to actually run the tests to compare these LEDs properly.

I should compare the battery model to back-of-envelope calculations. I should also add a model for ultra-capacitors, which would be trivial. Of course, just running with a capacitor (if you have one) would be a quick test, since the energy density is low.

The V1.0 design isn't protected against reverse battery insertion, which is easy to do with coin cells. This paper from Ti discusses adding a MOSFET to protect the circuit with minimal loss. Assuming I end up using a MOSFET for the LED driver, I might be able to find a dual version cheaper than two individual parts. http://www.ti.com/lit/an/slva139/slva139.pdf

I didn't think to compare the efficiencies of different LED colors - it turns out that green LEDs have traditionally been less efficient than other colors, although this information is probably dated. It would be interesting to survey the field and see where the most efficiency (electrical-to-eye-sensitivity) ones are. The luminosity curve probably dominates here, so even inefficient LEDs near the eye sensitivity peak win.

It might be interesting to consider smaller batteries. The CR2032 has decent capacity and is widely available, but is a little bulky compared to the other components (at least until I start using larger LEDs).

I need a way to make some simple absolute intensity measurements. All the data I have been able to collect so far for LED intensity has been relative. If I can measure the absolute output at even one current setting for an LED, I can calibrate the relative intensity measurements. This would allow comparison of the intensity of different LEDs. I think I'm going to build an integrating sphere with a 12" hollow styrofoam sphere and barium sulfate paint. It should be easy enough to get a reading for the total output at high current, which will be enough to calibrate the whole relative curve.

Jim Williams wrote a few applications notes detailing his designs for very efficient fluorescent LCD backlight drivers. The technology is different, but the idea of measuring and optimizing electrical-to-optical efficiency is the same. There are probably some good clues in there.