Power Designs 4050 supply problems

[zike]

Next patient is a Power Designs 4050 40V/5A supply. It's a beast, 15kg and scary. This is one of the types that has a Variac autotransformer ganged with the voltage control pot. It varies the raw AC presented to the rectifier, tracking the output demand to give a constant voltage drop across the output transistors. Serial #010018 was assembled early 1980 and came to me neglected and dirty, but “sort of working.”



Indeed, pilot light came on, about the right output voltage came out the bananas, and the panel V meter followed; but only as long as there was no load.

Drawing any load whatsoever, even tens of mA, the voltage shot up *higher*. Same happened if one activated current limit, by dialing the control to zero. If the supply had been at 40V with no load, the output shot up to 58V when you drew current. And here, unfortunately, the overvoltage crowbar also proved to be inoperative. I suspect circuits inside the supply could have been fried; at one point my DC electronic load briefly claimed to be sinking 420 watts, but I must have caught it in time as it seems to be OK.   

Scope quickly revealed that the power supply was in fact going unstable and screaming like a banshee, with chaotic 30Vpp power oscillations at 20-50kHz.

Time to confront the schematic. Fortunately Chris Grossman posted this excellent manual scan.  Here is the diagram:



While legible, just try and follow it; it's unintelligible! I'm 99% sure it was deliberately obfuscated, nobody could make such a mess without trying. They did label the controls, wires and test points, and even included test voltages, a step up from earlier PD schematics; so it seems they're letting you test individual components, but unwilling to reveal what they do or how.

I think someone on this forum quoted Bob Pease as saying nobody could understand a PD schematic unless they first redrew it; I have to agree. The very similar (and much earlier) 3650 schematic isn’t great, but has a vaguely logical signal flow. This gave me some hints. Here’s my version of the 4050. Yes, I reverse-engineered a schematic, from the OEM’s schematic (corrections appreciated).



First hint of progress was finding C12 was dried out, only showing 100 uF (diagram says 680 uF, but my unit came with 750 uF as-built). This is a shunt cap directly across the supply output. Adding 1,000 uF in parallel greatly diminished the oscillations, but did not eliminate them; the frequency moved up to several hundred kHz. I'd normally expect adding reactance would further destabilize a loop with poor phase margin, so I don’t really understand this yet.

Anyway at this point I noticed there’s actually a pot, R39, labeled “OSCILLATION SUPPRESSOR”. Someone had a sense of humor. Indeed, twiddling it fixed the oscillation, although  the action seemed irregular and unrepeatable. Eventually I found the best setting was about where it started (marked at the factory with paint). My best guess is that R39 had corroded and the wiper went open, defeating a key lead/lag compensation network. Twiddling may have reestablished contact. (BTW in many places, as here, PD left open the unused ends of pots used as variable resistors, rather than shorting them to the wiper; this strikes me as a bad practice.)

So now the supply regulates well 0-40 volts, and quite easily puts out the rated current. Indeed, quite a bit more, if you ask; unfortunately, the current control is now completely inoperative. As is the overcurrent fault circuit. I’m afraid a lot of stuff may have gotten fried.

I will have to dig into those, and the useless crowbar, next. At least I can say the bias supply, reference, voltage regulator, and output pass bank are still good.   

Taking a break, if anyone has any ideas or suggestions I’m  keen to hear. I will attach some internal photos below to entertain the curious.

Cheers,

Z

[zike]

A smoking gun (or SCR), and a cry for help...

Don't know how I missed this before, the remains of crowbar SCR CR13 speak of a violent past.  Yes, a dead short every way. But then, how did the supply make output?

Pictured below is the SCR and its 18 AWG teflon-insulated anode wire, which fused and parted company at the connection (the bitter end's still wrapped through the solder lug).


This "fuse" evidently stopped further carnage.  Teflon insulation doesn't curdle and peel like that till it's cooked over 250C for a long while (DAMHIKT). I reckon this SCR saw at least 20A, and for some time.

Unclear how the power oscillations could have done this, or if so, why the current limiter, the "power dissipation limiter" circuit, the 7.5A DC fuse, or the output transistors failed to stop it before this point. BTW that fuse is fine; correct rating, and looks to be original.  I'm puzzled, and a little scared what else I may find.

EDIT: After thinking about it just a bit more, I'm pretty sure a previous owner tried to charge a car battery (wiithout a blocking diode) and turned down the crowbar threshold. That'd do it. 


Poking further, I found something unfamiliar and learned something;  hoping someone out there can further illuminate.

There's a small "trigger" SCR, CR4, that's supposed to fire CR13. It may also be toast. It seems to be a strange part. Here's the symbol on the schematic (I drew in CR13 to show the whole circuit):


The symbol has its gate leg on the anode side, unlike a regular SCR. Like everything in the parts list, PD gives it a secret proprietary part number, calling it a "complementary SCR."  From context, one surmises it triggers when the gate (biased toward DC-) is pulled less positive than the anode (vs. normal SCR, like CR13, which triggers when gate becomes more positive than the cathode).

Google and Gemini didn't find much on this. The symbol is absent from my 1967 GE SCR Bible; but my 1972 Fifth Edition does show the symbol on page 25, first entry under Thyristors,  calling it a "P.U.T.," for "Programmable Unijunction Transistor."  On page 105 it says this device conducts once the anode voltage exceeds the gate voltage by one diode drop. So that sounds about right. 

The GE bible has two such parts, 2N6027 and 2N6028. Miraculously, they're still made and in stock at Digikey ($4 each).

However, these only appear rated for 40 Volts absolute maximum, which is the nominal working voltage in this circuit. The fault trip range should include 50V.

I can't find any other parts of this description, so I'll get a couple to play with anyway.  But I think a PNP transitor or PFET could serve as well here; there seems no need for the trigger itself to latch,  CR13 will latch on its own.

Comments or suggestions very welcome!

Z

[zike]

TLDR:
Unit is working great now, 100% recovery.

Details:
The melted crowbar wiring looked more ominous than it actually was. After removing  CR13 from the chassis (BTW it is a 2N3897, 200V/20A stud mount SCR) I found it wasn't blown after all. I misled myself trying to check it in-circuit. Amazing how tough those things are; the anode wire melted, but the silicon just shrugged.

I put it back, with a new piece of wire. CR4, the trigger “Programmable Unijunction Transistor” aka “Complementary SCR” (which MIGHT be a 2N6028- no markings!) turned out to be just fine, too.

The crowbar worked a treat. Once, that is. It immediately blew DC fuse F2.

Of course; we still have no current control (D'OH!). OK, excellent test of the protection circuit! Yeah, I meant to do that!

With a new fuse and no oscillations, it was quick work to isolate the next problem. CR9, a so-called “Stabistor” which sets a -2.6V (w.r.t DC +) reference threshold for both the current limit and the passbank "dissipation monitor”, was not regulating. Its voltage drop varied from 4V to more than 12V, depending on how you touch the leads.

Stabistors are extinct, can’t even find a datasheet. This one is marked “S1250A” and is the shape and size of a 1N4001. The symbol looks like a bunch of diodes in series, which I understand is what’s inside; so I initially just planned on stringing together 4 1N4148’s. But then I found a nice 1N5221 2.4V Zener in my antique HP spares. A bit low, but it stabilized the current/power reference so I could keep going.

The supply still didn’t regulate current. This was quickly traced to NPN transistor Q9 which had an open emitter. Labeled a “TI22720/U” in the parts list, this turns out to be just a 2N2219A, easy swap. The failed stabistor probably did it in by feeding -12V into its base. 

Et voilà, everything works! Almost everything. Alignment procedure specifies 5.5A headroom to guarantee regulation up to 5.0A in all conditions, but I could only get to 5.3A, due to the diminished CR9. Not a big deal, I'd have just lived with it. But then it occurred to me, this isn't the 20th century anymore. A 7-cent TL431 ref gets our supply over 5.5A,  presumably with more current stability than it ever dreamed of. TL431 to the rescue! Warms my heart.



At first it wasn’t obvious how much current CR9 handled, so I wasn’t sure a 431 was viable long-term, but I eventually surmised it's in series with the current return path for voltage reference VR3. This is fed by 15 mA current source Q2. Including Q2’s bias, the loop carries about 20 mA, no problem for the TL431. Just to be safe, I checked it with a thermal camera and it’s happy in there, about 10C above ambient.

With everything functioning I went through the simple alignments in the manual. Everything landed in spec without trouble, including the “dissipation monitor” circuit. I had planned to revisit the “Oscillation Overthruster” R39 and see if I could optimize the transient recovery; but the best setting is at the end of the pot’s range (very close to the factory mark). This appears to maximize local negative feedback around Q10. I did not quite certify that banging from 500 mA to 5 A and back settles to .01% in under 50 microseconds, my DC load is too slow. But from what I could see on the scope the edges are clean with no overshoot.

I also verified that the ripple and noise are well below 500 uV, and the load regulation is within 2 mV from zero to 5A (there's a feedforward adjustment labeled “Load Current Compensation,” but it’s in spec so I left it alone )

Here is the finished product warming a load.

[zike]

Stabistors are extinct, can’t even find a datasheet. This one is marked “S1250A” and is the shape and size of a 1N4001.

Postscript: There is Stabistor data in the attached Central Semi product discontinuation notice, dated 2016. Original part could have been something like a "selected" CN5179 (PD parts list says "red dot" next to their internal PN, suggesting a selected part).

Which would be needed, if this ID is correct; initial voltage tolerance is ±10% and load variation from 1 to 100 mA is almost a volt (never mind tempco, presumably that of a standard Si diode). Truly an awful choice for a voltage reference; no wonder they went extinct. PD still using it in the 80's is kind of bizarre. It doesn't appear to be legacy, as the most obvious ancestor, the 3650, used a normal zener for this exact function. Maybe they had some deviously clever load/temperature/lunar phase compensation mechanism in mind? Or a warehouse full of surplus stabistors?

But then it occurred to me, this isn't the 20th century anymore. A 7-cent TL431 ref gets our supply over 5.5A,  presumably with more current stability than it ever dreamed of. TL431 to the rescue!

I stand corrected, the TL431 was in fact announced in 1977.  Making OEM design choice even more puzzling.