power supply subtelties

There's a chapter(*) in one of Jim Willams' books about a guy who built big SMUish things using a '1/2 pole' rolloff--a bunch of lead-lags that approximated a 10 dB/decade, 45-degree phase shift network. At that point it didn't matter what the load capacitance was, the loop was always stable. It's probably possible to make a digital version of that.

Cheers

Phil Hobbs

(*) Phil Perkins, "My approach to feedback loop design", Ch 22 of Jim Williams, _Analog Circuit Design: Art, Science, and Personalities_

On Fri, 24 Jun 2022 12:45:41 -0400, Phil Hobbs snipped-for-privacy@electrooptical.net wrote:

Here's a possible filter.

The ESR could be native to some electrolytic caps, but probably added. They will get warm from the 250 KHz ripple current from our half-bridge switcher, which encourages a big inductor.

I have the Williams books; I'll look that up.

Version 4 SHEET 1 1348 680 WIRE -112 96 -160 96 WIRE -48 96 -112 96 WIRE 144 96 32 96 WIRE 288 96 224 96 WIRE 432 96 288 96 WIRE 528 96 432 96 WIRE 672 96 608 96 WIRE 720 96 672 96 WIRE 880 96 800 96 WIRE 1040 96 880 96 WIRE 1088 96 1040 96 WIRE 1248 96 1088 96 WIRE 1344 96 1248 96 WIRE 288 144 288 96 WIRE -160 176 -160 96 WIRE 672 192 672 96 WIRE 736 192 672 192 WIRE 880 192 880 96 WIRE 880 192 800 192 WIRE 432 208 432 96 WIRE 1088 208 1088 96 WIRE 1248 224 1248 96 WIRE 288 256 288 208 WIRE 880 256 880 192 WIRE -160 384 -160 256 WIRE 288 384 288 336 WIRE 432 384 432 272 WIRE 880 384 880 320 WIRE 1088 384 1088 288 WIRE 1248 384 1248 288 FLAG 288 384 0 FLAG -160 384 0 FLAG 1088 384 0 FLAG 1248 384 0 FLAG 432 384 0 FLAG 1040 96 OUT FLAG -112 96 GEN FLAG 880 384 0 SYMBOL ind 128 112 R270 WINDOW 0 -33 52 VTop 2 WINDOW 3 -38 53 VBottom 2 SYMATTR InstName L1 SYMATTR Value 50µ SYMBOL cap 272 144 R0 WINDOW 0 51 21 Left 2 WINDOW 3 48 50 Left 2 SYMATTR InstName C1 SYMATTR Value 1m SYMBOL res 272 240 R0 WINDOW 0 51 44 Left 2 WINDOW 3 46 75 Left 2 SYMATTR InstName Resr SYMATTR Value 250m SYMBOL res -64 112 R270 WINDOW 0 -33 55 VTop 2 WINDOW 3 -41 54 VBottom 2 SYMATTR InstName Rgen SYMATTR Value 100m SYMBOL voltage -160 160 R0 WINDOW 0 32 11 Left 2 WINDOW 3 38 73 Left 2 WINDOW 123 41 42 Left 2 WINDOW 39 0 0 Left 0 SYMATTR InstName V1 SYMATTR Value SINE(0 1 1K) SYMATTR Value2 AC 1 SYMBOL res 1072 192 R0 WINDOW 0 -78 35 Left 2 WINDOW 3 -68 66 Left 2 SYMATTR InstName Rload SYMATTR Value 100 SYMBOL cap 1232 224 R0 WINDOW 0 -81 2 Left 2 WINDOW 3 -68 37 Left 2 SYMATTR InstName Cload SYMATTR Value 1m SYMBOL cap 416 208 R0 WINDOW 0 56 21 Left 2 WINDOW 3 51 49 Left 2 SYMATTR InstName C3 SYMATTR Value 10µ SYMBOL res 512 112 R270 WINDOW 0 -39 65 VTop 2 WINDOW 3 -47 59 VBottom 2 SYMATTR InstName Rshunt SYMATTR Value 25m SYMBOL ind 704 112 R270 WINDOW 0 -30 34 VTop 2 WINDOW 3 -2 95 VBottom 2 SYMATTR InstName L2 SYMATTR Value 10µ SYMBOL cap 800 176 R90 WINDOW 0 67 63 VBottom 2 WINDOW 3 40 -4 VTop 2 SYMATTR InstName C2 SYMATTR Value 40n SYMBOL cap 864 256 R0 WINDOW 0 48 20 Left 2 WINDOW 3 49 49 Left 2 SYMATTR InstName C4 SYMATTR Value 5µ TEXT -16 288 Left 2 !.ac dec 20 100 300k TEXT -24 328 Left 2 ;Power Supply Filter TEXT -16 360 Left 2 ;JL Jun 24 2022 TEXT 728 272 Left 2 ;250 KHz TEXT 752 304 Left 2 ;trap

Interesting. I use notch filters in feedback loops for resonant actuators. They're the bomb for that, because the resonance is usually simple and isolated, so notching it out lets you use a much wider feedback BW.

I've played with them for switchers, but have never used one because they don't work that well with harmonic-rich waveforms (especially highly asymmetric ones). I'm usually happier keeping the extra two poles at high frequency.

Cheers

Phil Hobbs

I didn't go thru the analytics, but the output Z for the buck is proportional to sqrt(L/C) or something. Low impedance at high frequency requires only small C. It's up to the user to do their own decoupling anyway. It's a lost cause to try to do that with a general purpose power supply. I analyzed more than few HP bench tops ( 30 years ago) and don't recall them do anything arcane, manufacturing success correlates strongly with simplicity.

National invented a more than few unconditionally stable circuit topologies for their voltage regulator and power op amp product families. They go back to Widlar's day.

This is my current thinking. I can get my AC feedback from a local node that I can control the dynamics of, and get DC fb from the nasty remote sense. The notch filter really helps kill 250 KHz and above, and its impedance actually helps the control loop a little.

I guess my HF filter could be un-notched too, with a bigger L maybe. I'll try that.

Sure, as far back as (iirc) the 80s I used to use a fair number of LM6361As that were like that. It rather involves putting the compensation cap in the output stage, so that the capacitive loading appears in parallel with it.

Since JL is rolling his own, it might be possible to do that. It's tougher to do with an internally-compensated regulator that somebody else designed.

Cheers

Phil Hobbs

Fred, you confirm my intuition about how responsibility for proper power supply operation ultimately rests with its user. On the other hand, the use of a ground return for current sense remains nonintuitive.

The use of ground symbols in schematic diagrams (actually just a convenience for avoiding more lines in the drawing) lulls us into thinking that they're all at the same potential. That’s the essence of the fantasy ... but far from the truth. Until room-temperature super-conductors become a common reality, "grounds" are connected by wires, PCB traces, or sheets of metal - all of which have both resistance and inductance. So much for the fantasy! - Bill Whitlock

Danke,

HP did often include a big final electrolytic cap that could be jumpered in or out.

We want fast programmable voltage slew rates, clean fast current limiting, and stable remote sense no matter how far away or how stupid the wiring and the load may be. We could expect the load to be some decent fraction of a mile away.

We have to answer the phone when something doesn't work as expected. So we prefer to design a power supply that is maximally tolerant of customer wiring and loads.

I sometimes do the split AC/DC feedback thing wrapped round a cap multiplier. It does need a buffer to break the sneak path from the output reservoir cap to the output via the RC diplexer.

The ESR on the 1000 uF cap is probably on the high side. I'm using some nice 220 uF alpos with 25 mohm ESR.

Cheers

Phil Hobbs

I need that ESR to tame the phase shift at node MID, so we can close a reasonable loop. It will probably be an actual resistor.

Better be a honking big pulse rated job, then. A short could potentially dump

into that poor little resistor in under a millisecond. Toasty!

Cheers

Phil Hobbs

It kills efficiency too. AoE X-chapters has nice data on exploding various resistors.

The other approach is to use an LCLC filter with an effective bandwidth in the KHz region, and let it flail the phase all it wants up there, as long as it doesn't wreck our maybe 350 Hz control loop.

Something roughly like