Hopefully this will help. Attached is the data sheet for one brand of lm-317. Digikey is a gold mine for data sheets, if you use the search and get all the way to the actual parts listings, then click on the one you want a sheet for, the page that takes you to will usually have a link for the data sheet for that part, and that manufacturer. That's very likely where this one came from, as it's a National Semi part originally, but this one is from another brand.
The circuit I am using is their figure 3 "basic adjustable regulator", with a 200 ohm R1 and a variable R for R2. In my case, I used a 1k pot, but that will limit maximum output voltage to
1.25vref + 5 x (vref). In other words, this chip makes 1.25v reference between the adj pin and the output pin. The adj pin is an input, draws almost no current itself, and the output pin will always be 1.25v above this if there's enough input voltage available to make that possible. So, we will draw 1.25/200 amps current through R1, or 6.25 ma. This same current will also go through R2, which in turn sets the voltage at the adj pin off ground. So, with say 500 ohms there, we would have .00625 * 500 = 3.125 volts on the adj pin, and add 1.25 volt to that for a total of 4.375 volts on the output pin. The lowest voltage you can get out of one of these with the adj pin grounded is 1.25v. The highest is set by the spec on input->output drop, and is not an absolute number, one can use these in a 300v regulated supply if you are careful.
One trick for a quieter output is to add a capacitor to ground on the adj pin, but I didn't do that here, and putting a big one there needs added protection diodes (see data sheet). I did use a cap on the input, and one on the output 47uF for both in this case -- this isn't critical, I just grabbed the first two caps that came to hand out of the junk box. The input one is required due to the expected long leads to the real power source, and the output one smooths out the peaks of current draw from the CCFL portion of the circuit to help keep noise down. Ripple is not otherwise a worry here, as there's plenty of drop from the 13v input to where we are running -- 3.xx volts in this particular case, adjusted to set the output voltage from the supply as a whole.
Some LM-317 tricks:
- The smaller packages can be used to get a lower current limit, and a lower power limit as they also have a thermal shutdown. This does stress the chip, but they cost about the same as a fuse anyway...and are much more effective than a mere fuse.
- You can use an LM-350 if you need more current. It's otherwise the same thing.
- There is an LM-337 for negative regulators, and all the same tricks apply to it.
- All the other examples on the data sheet work fine, as in constant current, programming voltage outputs and output impedances and so on.
- You can even use them as an NPN audio output transistor, with the bias built in and in the other direction -- I've done it and it works fine. If you use a darlington PNP and emitter resistors, you can make a bipolar current booster with one, and the bias is free.
- You can drive the adj pin with the output of an opamp to close feedback loops. That wasn't required this time.
- LM-317's have a minimum output current you have to draw, or the output will rise. They're designed so all the internal drive currents come out this pin. In our case, the set resistor loads take care of that, but in a case where you're skimping on power and/or the load will be above minimum, you can use bigger R values for R1 and R2 to cut the parasitic losses to just the point where the output starts to rise from internal currents, usually lots less than the worst case data sheet spec.
For the CCFL part of things above, the main issues were things like noise, and how much power and output "stiffness" is required. Noise is always with us, and in this case I used the fact that the little board has two ground pins and used one for the input ground and the other for the output ground. The way I wired the board, there's also a grounded wire that tends to shield the adj pin on the lm-317 from the 50khz noise the inverter generates which is considerable -- that's one big square wave with fast edges.
I determined that we just did not need a lot of output stiffness or maximum current. After all, we are drawing sub-microamps during a pulse with the tube, and that's it, I didn't use a bleeder at all anywhere (would be OK to do that, but beware the extra power draw). So rather than substituting their tiny 27 pf current limiting caps with something bigger, I took one out instead, and used that spot for the diode to ground part of the standard voltage doubler circuit. I did use a pretty good high quality ceramic .01uf output capacitor for the output filter. Since I'm doing both parts of this, I also know there's another one of those .01uf caps in the preamp to help filtering. Since I know that, I put a series resistor in the output, so we have a second stage RC low pass filter as part of the design. This resistor is also a bit of current limit, but likely it will burn out if the output is shorted, a 100k one surely would, and arc over in the bargain at high voltage CCFL drives. Here again, it's cheaper than a fuse, so no particular worries.
Something perhaps counterintuitive that I did here was insert a resistor in the output ground as well, in this case about 120 ohms. This is because the bulk 13v input supply is grounded at one power strip, and the counters and audio amp grounded at another spot via a power strip. The result is about a 6 foot loop antenna, and a fusor can make some real EMI, which means ground over here is not the same as ground over there when there's noise present. There can be enough current (it's all fat wire) across this shorted turn to drive a signal effectively into the counters or audio amp. So we break that loop with a resistor, and have a low resistance cable to the audio amp, which grounds the preamp directly to that rather than some random spot on the "one turn inductor" that is our system ground(s). This prevents things like arcs from causing false counts and clicks. In this case, it was better to put these on the output side, because there the current is very low, so there's no appreciable drop across these resistors, and basically no other currents flowing there. Had we done this on the input ground, we'd have let in a bunch of 50khz current noise created by the CCFL input load, not good. One could improve on this by adding some turns through ferrite beads on both the input and output wire pairs, but in tests here (fairly brutal ones) they weren't needed.
Here's the data sheet for the CCFL I use, not too useful perhaps, but for completeness and the mechanical specs.
Note that I would do a couple of things differently depending on the expected load. For a phototube, which has some real current drain due to the divider R chain, I'd use a larger, maybe even .01uf series cap in the doubler instead of their 27 pf ones, for starters. I'd have to lose the series resistor in the output most likely as well. These changes in turn require more care to not blow out the fuse on the CCFL -- like using the smaller lm-317 with lower peak current limit. Phototubes have a gain curve that is very sensitive to supply voltage, so if I wanted to make a gamma spectrometer, I'd go full boat and use an opamp, a divider off the HV output, and a reference diode to close the loop and have truly well regulated output voltage. That's another post, though. I did make one up like that, and use it as a bench supply for detector testing and it's very handy. Since there's no size issue with that one, I went whole hog and used .47 uf output filter caps, and it will run a geiger tube for minutes with the input power switched off just from the charge those retain (or give you a darn painful shock/burn, there's no free!).
While virtually all counter type tubes want a positive output, which this is wired for, sometimes you need a negative one too, as with some very nice phototubes I have that have the output ground referred -- very handy. In that case, you simply turn all the diodes around. If you need both polarities (like in my bench supply version of this) you simply add the other two each diodes and capacitors and you'll have a nice split supply where both outputs track -- this is what I did for my ion source extraction supply in fact. What's super nice about the negative polarity powered phototubes is that the resistor and dynode currents don't pass through the signal output like they usually do when the other polarity lashup is used. You have to have one more wire, but then not needing any way to remove the HV DC off the signal (transformer or capacitor) is a pretty large benefit, and less noise from resistor current is always nice. Most of the phototube/socket combos out there don't have this feature, but can be rewired in most cases if you want to. If they are sealed, like the little NaI heads we stripped off a PET scanner, then you're out of luck and have to deal with that. You can't just put the load in the ground wire, because in that case, it's also the case/shield ground and will collect noise like a magnet.
Note, I've also floated the transformer secondary off ground as a test, with not very interesting results. I have no idea if that side will stand off the full rated output voltage....but it will float some. You can do this by cutting a track on the PCB on the secondary side.
Posting as just me, not as the forum owner. Everything I say is "in my opinion" and YMMV -- which should go for everyone without saying.