Melcor Deconstructed

[clintrubber]

I thought it was in Rafaels mail that there was no 100nF decoupling as drawn in the updated schematic, but decoupling in the form of those 10nF caps, but located to the power-transistor-collectors. Don't know if it will be critical, see Bcarso's info. The 100nF caps as drawn seem a good idea.

Heck, if everyone was already satisfied with that nasty 1-diode-D2 version, then probably the delicate influences of the exact decoupling schemaewill go largely unnoticed ! :wink: :grin:
Same goes for the left-plate-connection of C2 I guess, according to previous remarks (PRR suggested that its value or existence could very well be layout-dependent).

OK, let's talk further when we've soldered the corrections & have tested or better: heard the actual units.

Regards,


Peter

 

[rafafredd]

Well, I guess an important thing is about the notes PRR did reguarding the 18k resistor... Maybe I really Fucked up in there... I can´t certainly confirm I got this part perfec. Maybe PRR is righ about the 18k resistor. Someone should try it...

 

[clintrubber]

Even with completely different transistors the whole circuit simulated fine. But I can throw in the different 18k-situation, see what that gives. After having thought some more about it of course :thumb:

 

[bcarso]

The 18k (R9) provides a way for Q6 to turn off, and sets the current in Q4. The intention is probably to have the Q3 and Q4 collector currents equal. In that regard note that R8 and R9 constitue a current divider for the output of Q5. The designer may have been attempting to equalize the gain through the two different paths, that is, Q3-Q5 and Q4-Q6.

 

[clintrubber]

from Brad:

The 18k (R9) provides a way for Q6 to turn off, and sets the current in Q4. The intention is probably to have the Q3 and Q4 collector currents equal. In that regard note that R8 and R9 constitue a current divider for the output of Q5. The designer may have been attempting to equalize the gain through the two different paths, that is, Q3-Q5 and Q4-Q6.

Nice analysis, thanks.

and sets the current in Q4.

Sorry, I don't get this. Simply speaking & ignoring the fact that it's all in a feedback-loop, the diff-pair bases its current-division of the tail-current over on whats being fed to the base of Q3,Q4 and not on what it's seeing as its load, right ? But you'll have looked a step further here I suspect.


W.r.t. the connection of the lower-end of R9:

With the circuit as on the first page of this thread I did a simple comparison to find out more about the attachment of that 18k(R9) resistor. See PRRs & other comments on this; but I'm not sure whether PRR had doubts w.r.t. the 18k of the handdrawn-situation or the Flatpicker-drawn-circuit.

Anyway, connecting R9 to the emitter of Q5 doesn't seem to make sense without further actions. FWIW (=if accepted as evidence), sim says the seemingly good working situation turns into a bad one by doing that: output goes to -4.5 V and only the positive signal-peaks are put out.


As I see it there are now only two things left and both seem of lesser influence: the connection of the left-plate of C2 and eventual 10nF decoupling caps to the collectors of the power-devices.
That's it, right ?

O no, one more thing: DC-current consumption.
Do we get alike results ? I find 3mA in total (+/-15V), of which around 2mA in the power-devices (but as said my Q-models are different).
I'll comparte with the soldered 1731.


Have a good weekend,

Peter

 

[rafafredd]

Man, about the connection of C2, Yes, I can CERTAINLY CONFIRM that it was a feedback cap connected directly to the negative input.

Also, i CERTAINLY CONFIRM that the 10nF caps was decoupling the colectors of the output devices.

And it seems that the 18k is right in his place, in the end.

At least, that´s what was on the unit I´ve reversed. I hope it´s of any help.

 

[pmroz]

.
So next is some actual testing, right?
Dragged some of this over from other thread. Thought of a few revisions based on what some may have already built.
These suggestions were made before more-recent revelations...
My preference would be to leave the circuit as close to 'original' as possible for testing, now that we have a better idea what 'original' is.
Then apply changes and compare.

* internal cap C3 = 10pf (vs 680pf)
* supply = +/- 15V
* resistive load of = 20K
* feedback-cap = 200pf for the 1731.
Maybe we could use 220pf for 1731? Seems to be most common value currently in use, but i could be wrong.

* 22k & 22k gain-resistors (gain of 2)...
Is that R3 and R4? Originals were 39K2 as noted?
Fabio speced 20K for stepped gain. I'm dumb, but noticed on 2.5.2.0.s these Rs were also changed from 80k to 20k...

* raw opamp, so no TXs in front or after...
Could someone offer a good test-point for this on Fabio's pre?
Closest i can figure is pull input trafo secondary and RL (and RC, CC if used). Can i use test points somewhere without pulling these?

* some Hi-Z biasing-resistor to gnd (say 1MOhm) = If you're talking about the resistor from the non-inverting input to
ground I'd recommend using a 20k.
Not sure about this... you don't mean R1? Or separate 1Mohm connected for testing.

* capacitive load =nF ?????

Just a few more test conditions to confirm, and I could try applying the great sim work you all are doing to actual units.

Also, if anyone was willing to treat me like an idiot and give exact test/connection points, (esp 1Mohm and cap load) would appreciate.

 

[bcarso]

Peter said: "Sorry, I don't get this. Simply speaking & ignoring the fact that it's all in a feedback-loop, the diff-pair bases its current-division of the tail-current over on whats being fed to the base of Q3,Q4 and not on what it's seeing as its load, right ? But you'll have looked a step further here I suspect."

Open-loop, you are correct. What the collector currents do won't have much effect unless one or the other device saturates.

Within the overall global closed feedback loop, with that loop happy and stable, what the effects of Q3-Q4 collector currents are will determine about how the emitter current from R5 is divided up. And the differential input voltage at the bases of Q1-Q2 will change slightly to make what's required at the Q3-Q4 bases. Since the base current of Q6 is usually small, the 18k, R9, is the major determinant of what Q4's Ic has to be.

When I looked at the static values of Q3-4 collector current they weren't exactly equal, but I need to go back and set the rails to +/-15, now that we have settled on that. Of course device betas etc. will have some effect, but not much if it is a good design.

Brad

 

[bcarso]

Further pondering and sim shows that the effect of R8 on actual open loop behavior is small. It probably has some other role such as behavior in overload with further reduction in the available base current for Q8.

As far as its small effect on open loop behavior, it is probably best analyzed as adding a bit of extra bootstrapping gain to the 18k R9. The Q5-Q6 stage could be thought of as a type of totem-pole one, with the impedance seen at Q6's base much higher than if the equivalent of the 18k (with an appropriately larger value to set Q4 collector current) were tied to the negative rail. Thus the low-frequency open-loop gain is much larger.

If on the other hand R8 is reduced and R9 taken to zero, the difference is small in l.f. open loop gain. With R9 taken to 10.5k, R8 0, the gain is down about a dB (from 103 to 102). Clearly not a good reason to incorporate R8.

BTW these open loop sims include a stiffish (2k) load for the gain-setting divider and a 600 ohm output load as well. Gain will be higher with higher loading Z's.

The phase behavior is not ideally 6dB/oct. but at least indicates stability at unity gain: 0 dB open loop occurs at about 4.3MHz, where the phase is -164 degrees. This does argue for watching any loading and feedback divider components at low closed-loop gains as it clearly won't tolerate much phase shift at high loop gains.

 

[clintrubber]

from Rafael:

Man, about the connection of C2, Yes, I can CERTAINLY CONFIRM that it was a feedback cap connected directly to the negative input.

Also, i CERTAINLY CONFIRM that the 10nF caps was decoupling the colectors of the output devices.

Thanks, both cool. We may move those now (in the clone) present 100nF-caps to the other side of the 10 Ohm resistors and perhaps keep their value.

And it seems that the 18k is right in his place, in the end.

Looks like it, nice - again, my sim-Q's are not accurate, but the alternative connection of the 18k gave a faulty circuit operation.

At least, that´s what was on the unit I´ve reversed. I hope it´s of any help.

Sure ! You were there, and we tried to find out and now we know about everything again. Thanks !

 

[clintrubber]

from pmroz:

* 22k & 22k gain-resistors (gain of 2)...
Is that R3 and R4? Originals were 39K2 as noted?
Fabio speced 20K for stepped gain. I'm dumb, but noticed on 2.5.2.0.s these Rs were also changed from 80k to 20k...

No, those are the 'external' resistors, around the opamp so not on the 1731-schematic.

* raw opamp, so no TXs in front or after...
Could someone offer a good test-point for this on Fabio's pre?
Closest i can figure is pull input trafo secondary and RL (and RC, CC if used). Can i use test points somewhere without pulling these?

* some Hi-Z biasing-resistor to gnd (say 1MOhm) = If you're talking about the resistor from the non-inverting input to
ground I'd recommend using a 20k.
Not sure about this... you don't mean R1? Or separate 1Mohm connected for testing.

No, also outside the 1731 itself. Fabio's R.3.1.20-schematic and the Bo Hansen-info show the idea.

* capacitive load =nF ?????

Just a few more test conditions to confirm, and I could try applying the great sim work you all are doing to actual units.

Also, if anyone was willing to treat me like an idiot and give exact test/connection points, (esp 1Mohm and cap load) would appreciate.

Let's skip the cap load for now. It'd be to mimic the external cap-loading, which is important when talking about opamp-stability, but at the time of writing there seemed to be more severe things wrong, so it was put aside then. maybe it's now time to revisit it, but why then not right away 'for real' (PCB, scope, ...)

Bye,

Peter

 

[PRR]

> indicates stability at unity gain: 0 dB open loop occurs at about 4.3MHz, where the phase is -164 degrees.

I suspect it was NOT intended to run at unity-gain, or they would not have used the simple resistor long-tail under the input emitters. Input balance goes way off with large common-mode input.

Whoops: they could legitimately have used it as an inverter, specifically a mix-sum stage, where it would run at noise gain around 2 and zero common-mode voltage.

But usually, amps like this were used to amplify. Too expensive to toss them around as unity gain buffers, the way we do chips.

The connection of R18 to a lower emitter popped-out of one early sketch (when we still didn't know transistor pin-out) and was probably wrong all along.

 

[clintrubber]

from Brad:

When I looked at the static values of Q3-4 collector current they weren't exactly equal, but I need to go back and set the rails to +/-15, now that we have settled on that.

You saw then being different - at which supply ?

If on the other hand R8 is reduced and R9 taken to zero, the difference is small in l.f. open loop gain. With R9 taken to 10.5k, R8 0, the gain is down about a dB (from 103 to 102). Clearly not a good reason to incorporate R8.

Interesting (re-)reading this all. Sure, one can't get away with assuming open-loop were it isn't, the closing should follow and its consequences analyzed. :wink:
Reminds me of my old boss - familiar with BJTs and unfamiliar with FETs. So at analyzing a circuit he went: OK, it's a FET, but let's pretend it's a BJT :wink:

The 103/102 [dB] is the simulated openloop-gain @ 0Hz, correct ? FWINW, I found ~85 dB but could only use low-beta Q's.

Clearly not a good reason to incorporate R8.

Makes one want to ask the designer - but who has actually designed this circuit ? Perhaps it's not even one person and/or there's no sound idea behind it all. But let's not hope so.


from PRR:

I suspect it was NOT intended to run at unity-gain, or they would not have used the simple resistor long-tail under the input emitters. Input balance goes way off with large common-mode input.

It's tempting to add a current source, but then we're leaving our present path to the 'original 1731'.

The connection of R18 to a lower emitter popped-out of one early sketch (when we still didn't know transistor pin-out) and was probably wrong all along.

Good to hear this as being not very firm - guess that solves the 18k-issue.

Regards,

Peter

 

[bcarso]

"The 103/102 [dB] is the simulated openloop-gain @ 0Hz, correct ?"

Yes. It's -3dB at about 1kHz.

Brad

 

[pmroz]

.

Makes one want to ask the designer - but who has actually designed this circuit ? Perhaps it's not even one person and/or there's no sound idea behind it all. But let's not hope so.

There's this:
http://www.thelodgestudios.com/HistoryOfAPI.html
also Paul Wolff, formerly of API, posted a quick history at PSW, but 20 search minutes later i can't find it.
He said some people left Melcor to start API, including the few people who had designed the 1731. Same few set out designing a new opamp.
IIRC he did not include himself in the 'opamp designer' group.

 

[gar381]

Hi All

Has anyone had a chance to listen to a 1731 after these changes yet?

GARY

 

[fum]

I've tested mine with the double diode, but I still have to go back and update C1 to 680 pf.

I've got some built that are still original schemo (single diode, C1 1000pf), that I might drop in to the circuit, and may even possibly post a soundclip comparision, if I'm feeling really saucy.

Regards

ju

 

[bcarso]

from fum: "if I'm feeling really saucy. "

Better you than me. I'm feeling decidely non-saucy today (perhaps from too much of the sauce yesterday at a wedding reception...)

 

[clintrubber]

Has anyone had a chance to listen to a 1731 after these changes yet?

Not yet, but soon will. I deliberately went for stuffing a second PCB with the 'updated insights' to be able to compare it to the non-corrected version. That second PCB hasn't got all parts yet now.

 

[gar381]

Hi Ju

Keep us informed. I'm interested in knowing your thoughts
on the sound comparision befor I order caps and mod mine.
Sound clips would be wonderful as well if possible. :grin:

Thanks

GARY