new power supply architecture

Yes, it is a silly article.

Anyways, spikes on the output without RC divider can damage the chip. Spikes occur when output capacitors age or the load does things.

And their 'design' is not suitable for high voltage outputs.

It's an Analog Devices sales pitch - essentially an ad disguised as a application note. Using a precision current source to set up a flexible reference voltage does have advantages for relatively low output voltages. It's not silly.

Do tell us how the capacitors across a low voltage output age in a way that lets them product current spikes. High voltage capacitors can fail slowly as corona discharges eat away the metalisation, but that doesn't produce current spikes.

Obviously. It isn't claimed to be.

Especially since the current is probably derived from a bandgap reference.

Bypassing the lower resistor of a 317-style regulator improves the noise, and bypassing the upper resistor with ~10nF improves stability with low-ESR output caps. (I learned that trick from you.)

Doing both requires more thought.

Cheers

Phil Hobbs

You should buy a scope and measure stuff in switchmode regulators. I have had / repaired quite a few wallwarts that needed new electrolitics, both on the primary side and the secondary side You needs some hands on, repair too, experience Babble about papers is no good.

And FYI repairing stuff, from many manufacturer, gives you a lot more experience than reading papers.

No reg is going to be lower noise than its reference, and at least this idea allows the reference to be bypassed, although the values in the example are not impressive.

LP2985 has a reference bypass pin but its LF base noise is about 1 uV/rthz, about a hundred times what's reasonable.

A really low noise power supply needs a deliberate design, with references (plural?) and opamps and stuff. Doing this trick in a switcher seems silly to me.

We quoted a +-10 volt 20-bit programmable PPM-accurate supply last year but didn't get the job. But it was interesting and I learned a lot that will be useful some day.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Bill doesn't need an oscillosope. Whatever I say, he just takes the opposite position. Beats thinking or soldering or dangerous things like that.

Breadboarding with new parts, and mesauring things, is educational too. Data sheets seem to keep getting worse.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

I don’t do a lot of sub-3V supplies, so 1117s and so on are usually fine. I was just redoing the regulators on our best photoreceiver, the QL01, to run on +-15V instead of a 24V wart.

Like most single-ended circuits, the first stage doesn’t have much supply rejection, so I need the ripple to stay below 20 nV in 1 Hz. I don’t have much control over the input supply, so I want an honest >120 dB PSR in the regulators.

The first stage negative rail is -5, so that’s easy, but it sets the overall bias, so it needs to be accurate

Accordingly I used an LM337L followed by a three-pole, two-transistor cap multiplier supervised by an LM4041C-ADJ shunt reference. That part behaves like a PNP—the reference voltage appears between FB and cathode. That’s upside down from the usual TVL431ish things, which is convenient for negative regulators.

Because of the huge phase shift in the filter, you have to use the split-feedback trick, where the AC feedback is local and the DC comes from the output. Of course that makes a nasty sneak path, so you have to split the FB resistor in three sections, each with a bypass cap.

The positive rail needs to be at least +13 to meet the output swing spec in the worst case (14.5V in, 10V out into 100 ohms).

I wound up with an AP2205 putting out 13.9V and an unregulated one-transistor cap multiplier.

A fair few parts, but oh well.

Cheers

Phil Hobbs

We recently did a proto PCB to, among other things, characterize the Efinix T20 FPGA for prop delay and jitter and internal crosstalk and supply currents. It looks worthwhile to invest in a super-stable low-noise 1.2v power supply. Prop delay is about inverse on core voltage so microvolts matter.

Next time the temperature creeps above 0F, pull up a lawn chair and a gin-and-tonic and read the AD5791 data sheet.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

the 317 should be much more stable, they must be quite different, the

Different vendors’ parts have different stability properties, and it depends on the output voltage. Twenty microfarads with 5 mΩ ESR can easily make a 317 sing.

JL’s trick is to put 10nF across the top resistor, which gives enough phase lead to stabilize the loop even with big ceramic caps.

Another good approach is to use a huge wet-Al output cap as a lead-lag network.

Cheers

Phil Hobbs

That won't work when I want 1.2 volts out!

A tantalum output cap works too, 33 uF or some such; the ESR is about right, even with a mess of ceramics downstream. Just derate 2x or 3x on voltage so it won't detonate.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

What's "new" about this control loop architecture, anyway? The MC1466 does the same and dates from maybe 1978 or so.

John Larkin does like to think that world revolves around him. He doesn't do much thinking, because he isn't good at it. I've spent most of my life soldering and looking at oscilloscopes and thinking about the results, and I'd take it up again if anybody was willing to pay me to do it.

What counts is knowledge rather than undigested experience.

Or John Lakin's capacity to extract information from data sheets - never great - is declining as senile dementia kicks in.

If you don't mind the extra components, you can reduce the L.F. noise of the reference with a large brute-force R and C; then apply it to the base of a low-noise transistor which acts as an emitter follower to give the 'quiet' supply. I did this when I found that a voltage regulator was contributing more LF noise to the audio than the microphone capsules it was supplying.

In extreme cases, one can make a power supply out of opamps and voltage references and transistors and such, and get nanovolt noise. It's hard for an IC to do all the required stuff on one chip with one process, at a price that will sell many pieces.

Phil's favorite C-multiplier is great, especially if you don't mind the junction drop and its tempco.

Audio is easier in that people don't hear low Hz 1/f stuff.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

I realized just how noisy the LM317 was when I tried to use it to power an LNA with a target input-referred noise level below 300pV/sqrt(Hz).

It's specified as 0.003% of Vo in a 10Hz to 10kHz band. Assuming that's white noise, that would be 1.5uV/sqrt(Hz) for a 5V output! That's horrendous! A cascaded capacitance multiplier saved the day.

Jeroen Belleman

I have once measured the noise of some regulators. Noise density of my preamp = 220 pV/rtHz, Agilent 89441A spectrum analyzer + lots of remote control software. (source avail.)

There are also plots of Zeners / LEDs used as refs etc.

BTW on the leftmost photo is a peaceful valley, where the descendants of Mr. Gauss (the math Gauss) run a hotel. I used to live there about

Cheers, Gerhard.

I.E. a one-pole cap multiplier running open-loop.

Cheers

Phil Hobbs

That’s where the shunt regulator comes in. Also, if you have a higher rail handy, you can bias up the base to keep the drop down to 300 mV or so.

A resistor from base to emitter makes a useful beta-independent current sink for controlling the distribution of voltage drops in the base resistor string.

Cheers

Phil Hobbs