Moog power supply help needed.

Fixing a Minimoog at the moment with lots of issues, and I have repaired them all except for one nasty problem which I have seen once before...

If you look at the attached schematic, transistor Q2 on the left goes into oscillation and puts sawtooth ripple onto the power throughout the machine. It is very temperature sensitive, it gets better as it gets hotter, worse as it gets colder. Does anyone have any suggestions for a mod to stabilize the circuit? Someone in the past has tried to fix it with a 47 uFarad electro between the emitter and collector, it helps a bit but isn't a good answer.

What the heck, I will just build a new power supply with an LM317 and an LM337.

Not immediately clear what Q2 is supposed to do anyway! It doesn't appear to be associated with the major regulatory loop.

I'm sure whoever (Bob himself, or??) put it in there had his/her reasons.

> Not immediately clear what Q2 is supposed to do anyway!

It's the starter, isn't it? Actually the starter cut-out.

Everything else is powered through Q1. So at power-on, Q1 is off, everything is off, Q1 stays off forever.

But R5 R12 leak raw supply onto Q5 base, which sucks Q1 on.

R5 R12 would also leak ripple into the control loop. But when "+10V" rises to 8V, R2 R11 turn-on Q2 which clamps R5 and kills the buzz-leak. This 8V threshold is based on two temp-sensitive parameters of Q2, which may be a Clue.

> with a 47 uFarad electro between the emitter and collector

Huh. That's crude but not dumb. We can stand the DC leak, we just don't want the ripple getting into the toy-Moog's simple circuits. 12K+47uFd will muffle a lot of ripple. Then you don't need, and perhaps don't want, Q2. Try lifting one leg (any leg) off the board. If that seems happy, I could live with the hack.

For a serious diagnosis, lift a Q2 leg. Ignore the small ripple. Do you get -correct- voltage across CR3 (5.9V-6.5V) and can you trim "+10V" very close to 10V? If it has been mis-trimmed or sagged to ~~8V it's gonna be marginal. Considering tolerance and variation, it could get goofy at 9.5V.

"+Unreg" can be close to 10V, but must always be higher. If your power transformer, rectifier, or bulk cap are not up to spec, fix them. (My microwave oven lost a rectifier in a lightning strike, the display is dim and flickers due to half-wave operation. But it doesn't affect the result, like it could in a Moog.)

If we assume the "TIS-92" can really be many different parts of different area and doping, my "8V" estimate could be further off, even near 10V on a cool winter day. So change R11 to 1K or 1K2. That gets the starter kick-out down toward 6V nominal. I think we could allow kick-out to be as low as 3V. All we need to do is get Q9 into conduction.

> LM317

Why is "+10V" using a PNP device? In those days, Si PNP still cost more than Si NPN. The usual reason is that on a positive supply, a PNP will have sub-Volt dropout, an NPN will be dead to 0.7V and usually wants several volts leeway or several more transistors. I hate to think this fine machine is so marginal that a volt either way matters, but it is a MINI and excess headroom means bigger iron etc. The classic 3-pin regs may drop-out... I dunno but something to check before you dig-in.

Hang on. I can't face the future and I don't remember my past.

In an analog synth, everything drifts. This can't be cured, but is helped if the power supply is stable like that platinum bar in the cellar in Paris.

I was wondering why such an elaborate power supply.

If we leave R5 R12 in action, utility company voltage drift gets into the loop. The loop levers it down, but the ear's pitch sensitivity levers it up.

Unless your wall-outlet is VERY steady, you may "need" Q2 to cut-out the R5 R12 path. (Or add a start/run switch like a Hammond tonewheel organ.)

This also impacts the idea to use a commodity 3-pin reg. They are very good, they are not SOLID. They have every trick possible in a $0.39 pinhead die, but really solid regulation (line, load, temperature) is not sure to be good-enough for an analog music synth. At the least, you'd like the reference and power devices well apart, not molded in the same small bit of crystal.

BTW: I hadda giggle when I remembered what "CR" in CR3 means.

interesting circuit.

I don't see how Q2 could be causing a problem, unless R11 is open... and I don't understand why heating it up makes it better (heating should increase beta, if it's operating in a linear region, right??)

If you want to play with Q2 and caps, maybe try a cap across the base of Q2 to GND. that should slow it down, at least... and if not, you know the problem is not Q2.


I'm really interested to know exactly what the other half of the positive power supply is doing (the Q8,9, and friends) looks to be some kind of diff amp? comparing a voltage ref with a voltage divider, then adjusting current into Q5 which actually controls Q1?

Measurements are all done under load, right?

Negative rail all good??

PRR, you have a brain the size of the universe. I need to print out your posts and read them through a couple of times with a cup of tea to mine the gems, thank you so much for your efforts.

No-Fi, some good suggestions too, nice insights, I will now have some ammo to get to work with in the morning. Measurements are all under load, and the neg supply is not sensitive.

Q2 is massively heat sensitive, cool it down and the supply goes into a sawtooth frenzy, no such problem with Q5. I swapped them and the result is the same. This is a problem that other Mini's have had also.

The Moog sounds good when the room is hot, when it is cold the oscillators sound grindy and I can see the pitch wobbling on my CRO. (Which reminds me how dumb I was buying a digital storage CRO a few months back in a fit of pique that I missed a really good analog Tektronix on Ebay, digital scopes are terrible for audio analysis).

As for the 3 pin regulators, lots of synths use them, most ARP's, Prophet 600 and so on, I see them in synths all the time, but then ARP's VCO's were better than Moogs as far as stability anyway, so I will take another crack at fixing the Moog supply first.

> I don't understand why heating it up makes it better (heating should increase beta, if it's operating in a linear region, right??)

It's not real linear; and heat changes Vbe which is the trigger level.

Anyway it "can't" run linear. That gives a three inverter feedback loop. Sure, at DC it is negative feedback, but it is bound to find a 3-phase-shift zone, go positive, and oscillate. Instead it is supposed to transition abruptly to Q2 full-on, breaking that loop and muting the ripple path. Somehow it has forgotten to do that. (Since there is one non-stock cap found in it, I wonder what else may have changed since it left the factory.)

> exactly what the other half of the positive power supply is doing (the Q8,9, and friends) looks to be some kind of diff amp? comparing a voltage ref with a voltage divider, then adjusting current into Q5 which actually controls Q1?

Well, yeah. Not the simplest regulator ever drawn, but nothing odd.

> Negative rail all good??

To a first approximation: should not matter. Neg is slaved to Pos; Neg does not load Pos much. Of course with 20 year old stuff there could be major part-failures or bodged repairs; that's how Steve earns the big money.

I said that Q2 has an action threshold near 8V, temp-sensitive. Here is raw voltage ramped 0V-10V, temperature stepped 10C to 40C. Of course at low voltage, R5 R14 leak into Q5 to get things started. Above 8V, Q2 clamps the R5 R14 junction to kill the trash path. I don't really like these numbers, 2nA seems insignificant, and I'm too lazy to sim the whole shebang. But the range of action thresholds as temperature changes echos what you are saying about temp sensitivity.



I'm thinking that this is not one of Bob's best-thought designs. That it worked on his bench and generally in production but was really on the edge of not-working right.

Gut-thought: R11 up to 1K or 1K2. Sim says "no" but sims lie a lot.

Another point about 3-pin regs: note that the Neg is slaved to the Pos. This could be just to share the one high-quality Zener. However there may be some circuit in the synth which depends on the two supplies being precisely equal/opposite. Or at least both drifty in the same way. Independent regs won't give that, and it may degrade drift.

As I said, brain the size of the universe. Bumped Rll up to 1K this morning, it made a huge difference, so far it seems to be clean and stable, I have to freeze Q2 down well below any ambient temp that it will ever see to get it to start oscillating. I will calibrate the oscillators and test it for a day and see how it goes, but I think you cracked it PRR. Also, with that 1K resistor in place it is much easier to get the PSU supplies symmetrical.

Does changing this value up move Q2's response to a point where there is more hysteresis in the circuit and hence less oscillation?

BTW, if every educator could make their descriptions as clearly and sensibly as you there would be a lot of happy university students out there.

As for the big money part of synth repairs, well, it's a nice theory. With vintage gear more and more the job is a restoration more than a repair, especially with the big poly's. It is rare that a machine comes in with only one fault, but the customer only asks to have the biggest one, ie: the show stopper fixed. I repair the fault and then check everything and find several other things not working, so I go ahead and fix them anyway without charge as I can't stand the thought of a piece of gear that I have serviced going out with problems.

[quote author="PRR"]> It's the starter, isn't it? Actually the starter cut-out.
[/quote]

Yes clearly so---it's one of those bootstrapped reference topologies which is great if it starts---which it won't without a kick in the head. There are about 500k speaker systems out there that I designed, which have power supplies along those lines, that would keep me awake at night if I still worried about this stuff.

From a design standpoint, the best redesign strategy could be to make the threshold of cutout of the starter circuit less dependent on Vbe and beta.

> As for the big money part of synth repairs, well, it's a nice theory.

He, he. I know the reality.

me> I don't really like these numbers, 2nA seems insignificant

OK, I found my mistake, just over-simplification.

Corected, it says the unloaded raw supply voltage must rise above about 14V to bring Q5 on and get power to the regulation servo. We would expect over 14V for a nominal 10V out, but older gear in faraway lands, this has to be checked. (My Leader 'scope, as it ages, has trouble making enough raw DC to be stable.)

> make their descriptions as clearly

You would enjoy both books by T. K. Hemingway. His topics are not synths, but basic and related enough you will gain something. abe.com probaby has a few copies at prices a synth-fixer can swallow.

> great if it starts---which it won't

Hemingway has a few pages on fallacies including this one.

Hemingway on Non-Starters, 1.5MB PDF file

> the best redesign strategy could be to make the threshold of cutout of the starter circuit less dependent on Vbe and beta.

Yes, but how to do that without adding more parts and new bugs?

I eventually realized that we can't leave the R5 R14 path because it leaks turn-on current to Q5 base which R27 might not be able to divert even if Q9 is full-off. Then light-loaded output soars. True, it has substantial bleeder, but still I mistrust a reg that floats high on light load.

The R5 R14 junction must be pulled down fairly far. This conflicts the idea of standing Q2 up on a stable voltage (Zener) for stable action-point.

We can add another and another transistor to get the sensing, reference, and clamping all on different parts. Aside from practical cramming difficulty, many transisors together hanging on a slow analog loop wants to find some quasi-stable point.

There may be a modern Low Battery IC which could be twisted to this function, but that could be hours skimming datasheets and working out the trick, only to find that the part is not really available.

And all on the not-big money Steve can collect from his customers.

While I can't make my cocktail-napkin or my simulator show trouble near nominal 10V, I was a bit surprised how hard it was to to get a soft or hard answer worth trusting. And working out the several temperature effects, and the device variation effects, requires an impractical number of cocktail napkins.

You are right that the downfall of these starting circuits is that they will find an unwanted stable state between fully on or off, if you give them the chance. In this case the unwanted state also oscillates around that region I guess.

One could enhance the threshold by putting a zener in series with a smaller-value R2---in fact it's tempting to try to run the top of a smaller value R2 from the anode of the voltage ref diode CR3, although that part wants a fairly precise current to give the specified 6.2V reference V.

But you really want a nice big hysteretic snap in that starter. There's probably a way without adding parts, but I wouldn't want to be flying a plane that depended on the circuit (or a hastily devised variant) anytime soon.

One approach that might work---make the Q2 turnoff path have net positive feedback by getting its base current from the collector of Q8. I don't have time to simulate that at the moment but it looks interesting. The tendency to find an intermediate state would be thwarted by the net four inversions.

The more I look at this whole regulator the more worried I get though---you have four capacitors involved in determining stability margin!

But then a number of the low-dropout IC regulators require a very specific ESR output filter cap, or they will oscillate too. And this synth was made a long time ago.

OK. It does appear to work well with feeding the base of Q2 from the collector of Q8. And you can eliminate R2.

I tried to trick it into oscillation with a slow input ramp around the threshold, so far without success. BTW if someone tries this sim, beware that many zener models in Spice are bogus---they actually behave as if they have little batteries in them!

One of the unseen parameters with this circuit that can foul up the analysis is the unspecified unregulated supply---we would really need to know its impedance, typical rise time when mains power is applied, etc.

Also, I don't know what the typical load impedance looks like.

It is also troubling that the threshold for turnon of Q5 and hence the whole shebang is itself also pretty temp sensitive. Bob may have been walking a little close to the edge of the parapet on that one too.

> I don't know what the typical load impedance looks like.

I wanted the total current demand.

The MPS U55 is one of those small heat-tab things, 1W free air 10W on infinite sink.

It says 2 Amps but is going downhill by 400mA.

If we assume 10V waste and 2W practical heat, or 20V waste and 4W heat, that's 200mA, which makes sense to me.

If most of that is resisty, 50 ohm load. The dynamic resistance may be higher due to all the CCSes, though there may be some heavy current-dumps to damp.

Working around the three gain stages, he got DC gain of 5,000-10,000. Output impedance could be a milliOhm. At DC. GBW seems quite high, though not designable on a napkin. He may have gain of 1K at 10KHz, Zout 0.1 ohms, which ain't bad.

> the threshold for turnon of Q5 and hence the whole shebang is itself also pretty temp sensitive.

Near 2mV/DegC divided by Q8Q9 gain, or near 0.1 mV/DegC, drift due to Q5? That's why the cheap 1N821 instead of the costly 1N829.... Q5 drift swamps CR3 drift. It's on the edge of what would be needed to make an analog synth's drift a non-issue.

He's certainly got all the major errors covered and it comes down to the "negligible" things. How much Early Effect in Q5 and Q1? How low can Q2 go in reducing input trash? Is the GBW rolloff smooth and stable?

We could assemble a "better" circuit with parts he didn't have in 1969(!). I dunno what you charge, but I would not bid it for a cut of what Steve gets from one, or even many, MiniMoog jobs. Aside from making a plan, it would have to be tested with the funky MiniMoog; hours of various patches and drift-checking. There were 12,000 MiniMoogs made 1970-1981, and if most have a problem and many owners could be parted from some cash, there could be a Market for an upgrade power supply. But the karma of Moog is more enthusiasm than money.....

Hmmm. Vintage MiniMoogs seem to ask high eBay prices. 2.5X original list??? Who knew there would ever be Big Money in toy synths? But the loot is in owning one, not repairing......

It is winter here, and the workshop is a 9 degrees this morning, I calibrated the supplies yesterday when the temp was much higher to exactly +/- 10.000 Volts, this morning I measured it 1 minute after turn on and got +/- 9.999 on each rail, so the old dog can still shoot pretty straight. No sign of oscillation either with that extra 180 ohms.

I wish the VCO's were as stable as the power supply.

I sold my "last-rev-with-the-super-stable-oscillator-board" Minimoog about 6 years ago for $3500.00 and bought a computer with that money, the computer is now worth about $40.00.

[quote author="PRR"]>> the threshold for turnon of Q5 and hence the whole shebang is itself also pretty temp sensitive.

Near 2mV/DegC divided by Q8Q9 gain, or near 0.1 mV/DegC, drift due to Q5? That's why the cheap 1N821 instead of the costly 1N829.... Q5 drift swamps CR3 drift. It's on the edge of what would be needed to make an analog synth's drift a non-issue. ......[/quote]

I meant the threshold for the unreg input voltage to fire the thing up, with the -2mV/degree K being multiplied by the resistor divider effects.

But we are helped by the loose regulation of the unloaded raw d.c., so even with low line and cold parts it's probably not an issue.

The stability once turned on looks quite good.

[quote author="Steve Jones"]...
I wish the VCO's were as stable as the power supply. ...
[/quote]

When I was the engineer for the UCLA studio, circa 1968, the room temperature fluctuations were terrible. The complaints about Moog oscillator drifts were unending.

When Jim Cooper took over, after I fled to full time with Astronomy and another set of slightly-less-irritating prima donnas, he wisely made friends with Facilities and determined what number to call and switch to have thrown to at least get the A/C to come on, so it was comfortable.

[quote author="bcarso"][quote author="Steve Jones"]...
I wish the VCO's were as stable as the power supply. ...
[/quote]

When I was the engineer for the UCLA studio, circa 1968, the room temperature fluctuations were terrible. The complaints about Moog oscillator drifts were unending.

When Jim Cooper took over, after I fled to full time with Astronomy and another set of slightly-less-irritating prima donnas, he wisely made friends with Facilities and determined what number to call and switch to have thrown to at least get the A/C to come on, so it was comfortable.[/quote]

I have a Yamaha CS-80, even the Minimoog seems stable beside that...

[quote author="PRR"]> the best redesign strategy could be to make the threshold of cutout of the starter circuit less dependent on Vbe and beta.
Yes, but how to do that without adding more parts and new bugs?

[...]

There may be a modern Low Battery IC which could be twisted to this function, but that could be hours skimming datasheets and working out the trick, only to find that the part is not really available.[/quote]

A possible contender is the Microchip MCP100/101/MCP120. Available in TO-92 in a range of trip voltages. Stocked by Digi-Key, Mouser, Jameco, Allied, Newark, RS, Farnell and many others. Has both active-high, active-low and open drain versions. Costs $0.30 in qty 1.

Drawback: really doesn't like to see more than 7V on its input or output. Output should be <= the input.

Suggested fix: Remove R2 and R11. Get a MCP100-450DI/TO (active low reset, 4V5 trip voltage). Connect a 2k2 resistor from +10V to the VDD of the MCP100, add a 5V6 zener in parallel with a 100n capacitor between VDD and VSS, connect VSS to circuit ground and hang a 4k7 resistor between the /RST output of the MCP100 and the base of Q2.

JDB.
(the zener isn't strictly necessary, and ~1mA base current for Q2 may be a bit much, but for QTY1 any time spent getting this <$1 fix even cheaper would be a waste of (non-)billable hours. Main issue with the design as presented is that the turn-on of that 1mA base current could cause a glitch at the input of the MCP, but with the given values that shouldn't lead to oscillations within the worst-case envelope of both a typical zener and the MCP. Replacing Q2 with a 2N7000 or other NMOSFET would fix that, but that's IMO overkill. Plus then you get to worry about the effects of hi-temp FET leakage...)