I am considering building my own driver for a ~1.5A LED to power a microscope for my research.

For this application, efficiency isn't critical, but because exposure times to the camera can be short (~1ms), stability/lack of ripple is important. Thus buck/boost or other switching regulators, and PWM, are usually avoided here. (Perhaps suitable output filtering would sort this out, but <500ns [edit: <500 microsec] turn-on/off time is also highly desirable.)

What's the best current regulator for this set-up? (I'll be using a decent switching DC voltage source wall-wart to provide input power to whichever constant-current source I choose.)

One option would be to use an adjustable voltage regulator like a LM317 (but capable of handling more current, so maybe the LT1764 which also has a useful shutdown pin), set up in the standard constant-current mode where current is determined by a resistor between the Vout and Vadj pins. (The voltage between these pins is maintained at ~1.21V for the LT1764, so for 1A you'd want a 1.2-ohm resistor, and for 1.5A a 0.8-ohm resistor.)

(source: diyaudioprojects.com)

The other option might be using a new "one-resistor" voltage regulator like the LT3083, where the voltage is controlled by the resistance-to-ground from a single pin (through which 50 microamps is pushed by varying the voltage: so e.g. a 20kohm resistor induces a 1V output, and a 1-ohm resistor on that output will therefore cause 1 amp to flow).

Here there are two "suggested" constant-current circuits from the datasheet: A constant-current source:

Or a lower-dropout LED driver

The nice thing about the LT3083 is that a 20-kohm potentiometer can be used to adjust current, unlike the LT1674, where you'd need a 20-ohm pot (rather harder to source) to adjust between ~.1A and 1.5A. But I'm not sure which of the two suggested driver circuits for the LT3083 would be more resistant to ripple from the input voltage source, if either. And I don't know if this is a bad idea for other reasons compared to the more standard regulators. (e.g. if the adj pin is left floating, it looks like the LT3083 will probably drive the voltage way up and fry the LED. So if the pot fails or a connection breaks, this could be bad news.)

So, anyone have suggestions for the best option?

  • \$\begingroup\$ If using a pot ensure that wiper bounce will not give fatal output bounce. \$\endgroup\$
    – Russell McMahon
    Commented Sep 24, 2012 at 16:16
  • \$\begingroup\$ Aah yes, oh dear, wiper bounce could be bad. (Seems especially worrisome in the LT3083 case where that would cause a voltage spike, I think?) Would this best be addressed via capacitors across the output? Or is there a better way to "debounce" a pot? \$\endgroup\$
    – zachrahan
    Commented Sep 24, 2012 at 16:22
  • \$\begingroup\$ Connect the wiper to the unused end of the pot => worst is that the current ramps up to the (presumably allowed) maximum. \$\endgroup\$ Commented Sep 24, 2012 at 17:02
  • \$\begingroup\$ If efficiency isn't critical, and ripple is, why use LED? The old bulb has thermal inertia and is just perfect for you, especially if fed with a DC voltage. Keep it simple \$\endgroup\$
    – Roland
    Commented Aug 27, 2022 at 11:40

1 Answer 1


You mention that you want fast switching. None of the devices will be able to do that. Any the feedback circuits internally are designed to be slow so that they don't to much incorporate noise. And in fact some of the feedback is thermal feedback. Put another way, the dominant pole is low, 100's of KHz is typically the the fastest response time. This is true whether you are using a disable pin or connecting/opening the load to the source.

The best way to fix this is that you build an external circuit that current steers between two limbs. On one limb will be your LED chain and on the other a dummy load that draws the same amount of current (and hopefully also has similar load characteristics). An example of such a circuit is a differential pair. You should be able to get 10's of ns switching time then.

The challenge with the two limbs will be that the load characteristic differences may disturb the feedback of the LM317 etc. which then reacts slowly. If you can't match the two limbs then I'd suggest building a current mirror that decouples the current output the LM317 etc. so that it only sees a constant load the mirror transistor sees the variability.

You should be able to simulate all of this in LTSPice or similar before building.

The other factors you mention above not show stoppers, so pick what you want.

  • \$\begingroup\$ Where does the OP mention he wants fast switching? \$\endgroup\$
    – Oli Glaser
    Commented Sep 24, 2012 at 15:55
  • \$\begingroup\$ 2nd paragraph see 500 ns comment. \$\endgroup\$ Commented Sep 24, 2012 at 15:57
  • \$\begingroup\$ Ah yes, the switch on/off time (I was looking for continuous switching) \$\endgroup\$
    – Oli Glaser
    Commented Sep 24, 2012 at 16:01
  • \$\begingroup\$ Oh dear, I intended to write <500 microsecond switching, not <500 nanosecond! (But what's a few orders of magnitude between friends...) Sorry -- that makes a big difference, I bet. So it seems any of the above configurations will easily provide that speed, it seems? Or would current-steering / still be better for switching? And any other caveats to keep in mind? (I'm a biologist, not an EE...) \$\endgroup\$
    – zachrahan
    Commented Sep 24, 2012 at 16:09
  • \$\begingroup\$ Nope, they should be fine at the lower speed. What you probably need is reproducibility which if you heat sink the device will be fine. So just use the enable. \$\endgroup\$ Commented Sep 24, 2012 at 16:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.