high voltage LDO

The effective dropout voltage is 1.6, not reaslly LDO.

Sure! LTC usually costs 2x-4x what others want. We use LTC when the performance justifies the cost. Or when they decide to be sorta competitive, like with octal dacs.

The LM2941 feedback is referenced to ground, and the chip has a ground pin. So any zener would need to be in the high side of the feedback path.

But I breadboarded an LM2941 reg, and it seems fine with up to 30 volts in and 29.xx out. It does shut down at just over 30 volts in, which isn't a problem since I'll be running from a 24 volt wart.

Nice chip, but a little weird.

Connectors are worse.

Enclosures are worse still.

I'm being harassed to use a Cuk converter to make my negative supply; I need -6 from +24. Cuk's are quiet and we have the parts. Not cheap and a fair number of parts, but being done has its own virtues. I can just swipe it from an existing design.

That LTC part costs us $2.55, which isn't real bad.

Care to explain how anything pinned to GND cannot be considered referenced to GND? There is no such requirement on placing the zener on the high side.

I wanted to suggest a switchmode solution but you probably hate them even more...

If the desired 22V is also +/-5%, why not just make a 2V dropper? ...Jim Thompson

Seems overkill to switch 24 to 22. A linear is 92% efficient doing that.

This board will be full of switchers!

The reason to use 22, and not 24, is that I want regulation.

Rather than engage in a philosophical argument, you might read the data sheet:

I see it now:

Please view in a fixed-width font such as Courier.

OUT -----+-------------+--> | | | | . | --- . | + --- . \-\ VZ | . ^ -+- . | - /// . +------- . | | . [R2] | . | | . | | . ADJ-----+ | . | [RB] . | | . [R1] | . | | . | | . | | . --- --- . /// /// . . . R2 . stability region 5< VREF X (1 + -- ) < 20 . R1 . . . R2 . VOUT= VREF X (1 + -- ) + VZ . R1 . . R2 . 5< VREF X (1 + -- ) =VOUT -VZ

It is not about efficiency, but because a suitable switcher design does not care that the input and output voltages are so close.

That could make it even simpler, depending on how they are designed.

The LT3066 has 270 to 380mV dropout at 400mA, two in parallel would work.

Actually, I suggest an opamp driving a p-channel MOSFET in a roll-your-own regulator.

There you go, reason returns. ...Jim Thompson

-- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | STV, Queen Creek, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at

Actually, I could forego current and thermal limiting in this application, so, given a handy reference from somewhere, a homebrew regulator would only take about 6 parts.

Great! Switch them all together or, even better, alternating phases.

Well, there's the old Maxim AN103 app note. But watch out to avoid the inverting opamp connection, since the p-channel MOSFET is playing that role. The appnote's drawing mistakenly marks pin 3 as the - input, whereas of course it's the + input, as you want. I'd also avoid copying other silly things they have, like pin 2 feedback. **

Google "p-channel mosfet ldo" for maybe 50 different bloke's p-channel circuits. But looking, I didn't see one to recommend.

Note: there are two amplifying stages in the feedback loop, and the p-channel stage will have an output capacitor as its load, making a serious pole. Furthermore, the op-amp is driving the MOSFET's Ciss capacitance, ouch. So loop compensation can be a messy issue. Add a zero with a cap on the feedback divider resistor. The output electrolytic's esr is your friend, adding another zero. Although piles of 0.1 bypasses can wreck that.

A good idea: slow the loop's bandwidth. You can use a cap from the op-amp's output to invert input, see AoE III, Fig 5.87.A ** Or a cap from the FET's drain to gate. Hah, whatever, add another zero with a resistor in series with that cap. :-)

SPICE modeling of MOSFET linear operation in the subthreshold region is a mess, but the errors it makes will probably model the FET's gm too high in your operating region, making SPICE a conservative test.

Let us know if you pursue this approach.

'T'is a shame that discrete manufacturers don't provide better models. The foundries I work with provide excellent models of subthreshold behavior. I did a medical application a few years ago where my whole circuit ran on 10nA (when not asleep ;-) ...Jim Thompson

I've done this before, and loop stability is no big deal. The pfet is a current source, so add a giant output cap - you need one anyhow - and the power thing becomes an integrator. Then just limit the gbw of the opamp with an RC network.

I guess I'll use the LM2941 and cheat the specs a little. It is interesting that it cuts off at +30 in, while the abs max is 60. And that the output is restricted to 5 to 20 volts. I wonder why.