While surfing for more LED information, I found this rather nifty little circuit on Electronic Design's website. It's a little Joule Thief-like circuit, but with enhanced efficiency (~80%) and it can drive white LEDs with rather large forward voltage drops. Archived for later....
When I was still in grade school, I (and this will be a shock to my readers) spent a lot of time in libraries. Our library used to have a free bin, where they would toss things that they no longer wanted in their collection. One day, I came by and found a pile of more than two decades of Scientific American magazines. Being a bit of a science nut, I carted them home, and spent many a happy evening reading Martin Gardner's Mathematical Games and C.L. Stong's Amateur Scientist column. Stong's column was really cool, and had great articles about building plasma jets, X-ray machines, diffraction ruling machines, and all sorts of other good stuff.
It was then with some suprise that I read Nyle Steiner's article on making FET transistors from cadmium sulfide photocells, because he made reference to a 1970 article by Roger Baker that was about making home made FET transistors. I didn't remember any such article. Luckily, I bought a CDROM which contains all the Amateur Scientist columns, and found it. And, indeed: it's the second half of a two part column, and talks about the deposition of thin films. I guess I had never read it closely enough to see that it included the manufacture of a thin film transistor.
The article also makes reference to THIN FILM MICROELECTRONICS: THE PREPARATION AND PROPERTIES OF COMPONENTS AND CIRCUIT ARRAYS which was available in preview mode on Google Books, and it seemed enticing enough that I decided to track down a copy via online search (I managed to find a copy for only $6 + $3.50 shipping). It seems like the level of this book might be within the reach of the dedicated amateur.
Okay, so here's the schematic for the LED transmitter circuit as I assembled this evening. I tried to write up an exposition of how it works, but frankly, it pales in comparison to the clarity and completeness of KA7OEI's page. But here's the basic idea: imagine that you supplied 1V to the input of the op amp. The op amp will drive the voltage at the base of the transistor so the voltage coming off the Rsense is the same (1V). When that happens, the current through Rsense (and through D1) is just 1V/100 ohms (in this case) or 10ma. And, the current is linear in the input voltage: when the voltage rises to 2V, the current is 20ma, and when the voltage drops to zero, the current is zero. So all you really need to do is take a 1V peak-to-peak audio signal, offset it by 1V (which I didn't show on the schematic, it's just inside the audio source node), and it will generate the appropriate current.
It's very cool, and requires very little math to actually design. And it doesn't matter much what the transistor is either: KA7OEI's page uses a power MOSFET, because he is driving a much more powerful LED, but for my purposes, using an ordinary 2N3904 worked just fine.