How to debug and fix voltage droop in SMPS

I have no idea of your skill level, so if I say someting too obvious don't take it personal.

First, how do you know the LED is only pulling 6 mA ? If it is a typical ab out 2 volt LED that would mean like a 500 ohm resistor in series with it. F or testing actually you could just use resistors.

Another thing, to tighten up regulation you can change the top resistor to a lower value and use a couple of junction drips like across 1N400X diodes. As long as the temperature stays halfway within reason it shouldn't drift too much. If you need high precision on the output voltage though forget it . Thermal comp will be a PITA.

I would be havin' a scope on that choke. See what happens to the duty cycle under different load conditions. Also with varying input voltages.

Also, another dummy one, if you are using batteries, are they maintaining t he input voltage ? Batteries vary vastly in current capability and amp-hour s. Ideally you should be using a bench power supply with a current meter on it.

I tellya, you can't trust batteries. got a batch in and while they work wel l in remote controls, in the digital camera they are only good for two or t hree shots. So if you are not assured of the proper input voltage, monitor that when applying the load, if you haven't. (that's what I mean by don't t ake offense, some people really would not have thought to do that)

Anyway, with the information given, that's all I got.

If you're really using an electrolytic cap -- stop. Surface-mount ceramic 10uF caps are vastly cheap these days, and are almost certainly what's intended for this circuit.

A 'lytic will give you a horrible amount of ripple at the switching frequency, and may well be confusing the regulator into thinking that it's holding the set point just fine.

I haven't used this exact part but I've used ones like it and not had trouble, so I'm not sure why it's giving you fits.

If you don't have one, then even a cheap one will help. For this job you probably want to look for an advertised bandwidth at least four times the switching rate of the chip -- more is always better, or would be if it didn't translate to more money.

I've seen shit like that. Oscillators will not oscillate, but other things will...

I have tried it with electrolytic, 10uF ceramic, and both in parallel. No change in the output voltage sag.

I had a similar problem with another TI part that turned out to be simnply me overloading the chip beyond its regulation ability. That one used a lot lower switching frequency, though.

I am thinking maybe this is a problem of coupling between the switching inductor and the feedback pin. Here's a scope capture I did with the inductor SW pin on Channel 1 and the FB pin on Channel 2.

Does this seem right?

It isn't that, then.

Come to think of it, though, it could be C18 -- if you haven't tried that with a ceramic yet you may want to.

It could be a layout issue. Ideally the ground ends of C18 and C19 would be on top of the chip ground, and the hot ends of C18 and C19 would be close to their respective pins on the chip.

If you can figure out how to monitor the coil current you may find some joy there -- if the chip limits coil current, and if the coil is saturating, that could cause issues like you're seeing. Or if the coil is just plain under-valued.