Analysis of a certain class A driver

[clintrubber]

[quote author="PRR"]But with all the other suspected transcription flaws, and the fact that this company's products are more than the sum of the parts, so they hate to see schematics in public, I'd rather drop it. If you want to copy another Man's product, you need more than the schematic, you need the parts-selection, the build-quality, the critical inspection, and the karma (I believe there is good karma on that Man's assembly line).[/quote]
Agreed, looks like there are too many things not surely known for that circuit. Original idea was to have an active stage and since that one was designed for driving EQ-sections as well it seemed right for the task and a good kickstart. Cloning that to get a bit of M' aboard wasn't the intention and would have been naive. Having observed those several odd things though, the curiousity kicked in :wink:

> it'll just work, no mysteries.

Well, it will work, but there are issues.

At a glance: the 5532 is asked to drive R1||R2 through C2 C7 and the AC-grounded supply rails. 235Ω, woof. If the 5532 could drive that, two in parallel could do the 100Ω simpler than all that other stuff. (In fact it would take 6 to 10 sections of 5532 to drive 100Ω well.)

As drawn here it's unlike the ESP-design:
that one does have those caps indeed connected to the 5532
( http://sound.westhost.com/p24_fig1.gif ), but here they're connected to the final output.
A simulation says the load of the 5532 is over 2k5, but I guess that'll vary with circuit-params. It looks however well enough above 600 Ohms.

In fact it is not necessary to bootstrap to both supply rails. Let the opamp drive the top base, and use diodes and resistor/bootstrap/CCS the bottom base.

I started from Jakobs headphone-driver topology but I've been wondering indeed when one chooses to drive from the top and when entering the diode-stack midway.

Idle current of 40mA is not bad, but is neither Class A nor Class B for large swing into 100Ω. It is class A up to 8V peak.

You're right, biasing should increased for A (FWIW, but let's say we want that).

The input impedance of these one-opamp "balanced" inputs is never balanced, though it never seems to matter in studio work.

You're right, all those circuits out there using this aren't exactly correct. I've been thinking of doing it 'more correct' but IIRIC this way of doing it isn't that bad since source impedances will be low. Ad when using it as an insert I'll be running unbalanced anyway, so less circuitry would be better then.
I recall a short article about simply redimensioning the circuit to solve the incorrectness, but I see nobody doing it.

> VDC as presented to the EQ below say 2mV.

Find out how much DC your inductors can stand, and their DCR, and how much of the amp offset voltage reaches them.
...
But I'm not sure how to scale this to treble coils, you better check the actual parts.

Thanks, I'll try to get hold of specs for the coils (likely Toko's). That OP07C opamp in the DC-servo has an insane offset-spec, I want to use it just because I have a few available. So the output will be low in offset (will check/think whether the offset-spec of that opamp guarantees that the complete circuit will be low in offset, but I guess so: input-current is low as well so as long as R8 & R15 are well matched all should be fine.

And I could keep overdoing things by adding an offset-adjust to the OP07).... sigh, like you already mentioned just stacking those 5532 and a few resistors wasn't such a bad idea :wink:

Thanks,

Peter

 

[clintrubber]

[quote author="NewYorkDave"]I haven't studied the slew of messages/threads relating to this in detail. But am I correct in understanding that all this circuitry is just to drive a Pultec filter section? :shock: [/quote]
You're right, things are getting out of hand here - parts count is starting to
equal that of the actual Pultec-EQ :grin: :roll:

I know that hashing out these circuits is fun, and I'm not trying to piss on your parade, but...

That's indeed what's going on: for the fun of it let's overdo it and learn something along the way. While the initial idea was to simply re-use that M'-circuit.

Honestly, I don't think very low-Z, high-level drive is what you want to feed a Pultec, not if you want a "classic" sound. Consider that the sources fed into a Pultec in the old days would be of fairly high source impedance, 600 ohms, maybe as low as 100 Ohms. With a 600-ohm source, 6dB was being lost at the input under "best case" control settings. Drive that filter with a high-level amp having a very low source Z and your inductors will be saturating like mad.

I wasn't sure about this - I understand the 600 Ohms source was unavoidable then, but Jakobs recommendation about higher-source Z spoiling the max-peak value sounds reasonable as well.

As a shortcut to all this:
if a higher sourceZ (say 600 i.s.o. 10 ) just gives more total attenuation and no changes of EQ-curves then that 10 Ohm and all the effort & mA's it takes is indeed overdoing things. I should simulate the EQ-circuit.

If you're really opposed to the idea of a stepdown transformer, for whatever reason, and you want to isolate the signal source from the filter, how about a "taper pad?"

The stepdown-TX starts to sound more and more attractive :wink:
The additional required makeupgain will be no problem, it's not that much as for that 'other' EQ which seems to need some 50dB.
To be (too) safe though, to change that worst case 75 Ohms into something managable by the previous stage, turns-ratio may have to be higher than liked. That 75 Ohms won't happen for hi-energy bass-frequencies here though, so we might be chasing ghosts.

I did(/do) want to address the potential problem of a source that can't drive the Pultec-EQ-box not properly enough.

Or, a simple active solution: a 5532 or 5534 set for a gain of 6dB, with a build-out resistor of 590 ohms on the output.

I"m a "keep it simple" kinda guy, myself :wink:

That could be a nice approach. As said, as long as source-Z doesn't change EQ-curves (too much), this way could make the circuit a lot simpler.


Thanks,

Peter
(suppressing the urge to add yet more: another control to adjust source-impedance :wink: )

 

[clintrubber]

[quote author="Samuel Groner"]

But am I correct in understanding that all this circuitry is just to drive a Pultec filter section?

I don't think so--it's for driving a massive 150 ohm passive EQ.[/quote]
No no, this just only to drive the EQP-Pultec ! :grin: :twisted:

A few thoughts on the new revision:
* Add a 10 ohm in series with R3 and a 100 pF (or a 47 pF to correct for C1 :?: ) across R3 to improve CMRR.

Interesting, hadn't given this much thought so far. So in addition to the cap directly across the 5534 you want to add another HF-stabilizing cap.
Just curious, why the 10 Ohms 'outside' that added cap ?

* I would replace C8 with two 220 pF caps going to ground (actually chassis/pin 1)--most RFI is CM and as shown this filter is acting for differential signals only and at a pretty high frequency.

You're right, when wanting to do something here then better do it completely. Since the incoming line won't be long I wanted to skimp on the frequency... could probably leave it out altogether but let's make it more universal.

* I agree with PRR that the bootstrapping is probably overkill--for this application, two red LEDs with one 6k8 to the negative rail could work very well. The high forward voltage drop makes you use large emitter resistors (about as shown now) which is good for temperature stability (they could negatively affect class AB distortion so make sure you get enough current for class A). The temperature stability is improved by the matching tempco of the red LEDs.

You mean that one red LED better matches (temp-wise) the BJT & R-emitter-combination ? Better than the 2*1N4148's ? Interesting, wasn't aware of that.

This does remind me of the following: there's this ('71 HP-)story to 'put around 25mV across your emitter-resistors' to keep output-impedance around zero drive low. Since it doesn't apply to class A stages the voltage drop across the emitter resistors has been chosen differently (way higher).
Hmm, since at the present dimensioning it's not class A all the way there will probably be a transition, as in the output-impedance changing. I figure this to be well away from the midpoint so will do less harm as is the case for the usual crossover-dist., but should give this one some more thinking.

* Personally I prefer servo opamps with input bias current compensation (I use OPA177); that saves me R15/C9. But nothing wrong as shown.

The intended OP07C has input bias current compenation as well, so they could be skipped here.


Thanks,

Peter

 

[clintrubber]

[quote author="Viitalahde"][quote author="NewYorkDave"]I don't think very low-Z, high-level drive is what you want to feed a Pultec, not if you want a "classic" sound[/quote]

The manual of this thing (Ma$$ive Pa$$ive) says that the "trick is to make the coils saturate just the right way" .. this is probably what they're after.

I don't like coil saturation much myself.. but maybe it's a matter of having the correct type of coil with just enough drive. A Toko inductor is probably not the right type of a coil. :razz: :razz: :razz:[/quote]
Wasn't it said that that box used Toko's as well ?
(I couldn't find it back here though)

However many components I'll be spending on just that input-driver, I won't be giving it anything else than unity-gain :thumb:

Bye,

Peter

 

[Samuel Groner]

Just curious, why the 10 Ohms 'outside' that added cap?

For best CMRR the two networks for the inverting and noninverting input should have equal common mode impedances. One is formed by:
* R19-R13-R6-C6-R5-R17-C4 (let's neglect R16 as it is heavily shunt by R17)
The other by:
* R18-R14-R9-C5-R3
As you easely can see, these networks are not well balanced--adding the suggested resistor and cap solves this. The additional cap goes across R10 only because C4 is across R5 only. If C4 were grounded, the additional cap would go across the 10k+10 combo.

C8 does not show up because it acts for differential signals only--two grounded caps would.

Note that the additional 10 ohm only matters if you match the larger resistors to 0.1% or better. In practice, it might be better to add a 100 ohm trim instead. Or to just forget it if you are happy with 40 dB CMRR...

You mean that one red LED better matches (temp-wise) the BJT & R-emitter-combination? Better than the 2*1N4148's?

It looks to me as if the bias network (i.e. the diodes) should have similar tempco as the Vbe of the transisor--two diodes have about twice the tempco.

Samuel

 

[Viitalahde]

[quote author="clintrubber"][quote author="Viitalahde"]A Toko inductor is probably not the right type of a coil.[/quote]Wasn't it said that that box used Toko's as well ?[/quote]

Yep, that's exactly what I ment. :wink:

 

[clintrubber]

[quote author="Samuel Groner"]

Just curious, why the 10 Ohms 'outside' that added cap?

For best CMRR the two networks for the inverting and noninverting input should have equal common mode impedances. One is formed by:
* R19-R13-R6-C6-R5-R17-C4 (let's neglect R16 as it is heavily shunt by R17)
The other by:
* R18-R14-R9-C5-R3
As you easely can see, these networks are not well balanced--adding the suggested resistor and cap solves this. The additional cap goes across R10 only because C4 is across R5 only. If C4 were grounded, the additional cap would go across the 10k+10 combo.[/quote]
I see, you're better matching both strings. Hadn't thought of it that way, thanks for the suggestion.
Final circuit for the task at hand should be way simpler (like various people already suggested) or when going for the complicated version (just for the fun of it), I should then add simply everything that makes sense, like those 10 Ohm & 100pF. :wink:

Note that the additional 10 ohm only matters if you match the larger resistors to 0.1% or better. In practice, it might be better to add a 100 ohm trim instead. Or to just forget it if you are happy with 40 dB CMRR...

40dB would be OK since the lines will be short. Most likely I'll even be using it unbalanced most of the time (channel-insert) so since the neg-input will be grounded then all additional efforts are kind of waisted (a dummy source-impedance will most probably not be present).

You mean that one red LED better matches (temp-wise) the BJT & R-emitter-combination? Better than the 2*1N4148's?

It looks to me as if the bias network (i.e. the diodes) should have similar tempco as the Vbe of the transisor--two diodes have about twice the tempco.

Sorry, I should have put it otherwise: you mean that the LED (GaAs, right ?) better tempco-matches the BJT+resistor ? I'll check for a GaAs-tempco.
Hmm, red LED, isn't that some 1.7V..2.1V ? More BJT-biasing already, yummy :wink:
I don't recall having seen LEDs used in the actual signal path (apart from a Rat-distortion pedal) but it'll be fine I guess.

BTW, when using the Si-diodes I realize it'd be better to mount only two of them on the heatsink (D1, D3).

Thanks,

Peter

 

[clintrubber]

[quote author="Viitalahde"][quote author="clintrubber"][quote author="Viitalahde"]A Toko inductor is probably not the right type of a coil.[/quote]Wasn't it said that that box used Toko's as well ?[/quote]

Yep, that's exactly what I ment. :wink:[/quote]
D :twisted: mn, don't spoil the fun now ! :wink:
But hey, if they're fine enough for that one...

Cheers

 

[bcarso]

[quote author="clintrubber"]I recall a short article about simply redimensioning the circuit to solve the incorrectness, but I see nobody doing it.

[/quote]

It's true the four-R single op-amp-based diff amp is not perfect, but neither is virtually any other such, including two-op-amp designs, until things get pretty hairy. But most of the time it, and they, are adequate.

The type of circuit in question has a different normal-mode Z than common-mode Z. For a ground-reffed signal one input R is twice the other, for example. But it is the common-mode Z and common-mode gain that counts when rejecting common-mode signals, and at low-mid frequencies (sufficient op amp open-loop gain) these are both equal. I've seen articles (or for that matter short notes*, a la "Ideas for Design"), that purported to discover and then correct deficiencies in the classic circuit, and they have been anywhere from misleading to flat wrong.

Whitlock's material, while inescapably a bit promotional, is pretty sound stuff.


Maybe it's time for a thread on diff amps, or a meta referencing discussions in other threads. A subtopic could be the general issue of bal-unbal conversion.



*Those little notes are frustrating, because most of the time there is no Letters section and no way to point out errors. Submit another note that says "Old circuit still works, as usual" ?---not likely to see the light of day.

 

[Samuel Groner]

I recall a short article about simply redimensioning the circuit to solve the incorrectness, but I see nobody doing it.

I often hear that the "four resistors and one opamp" circuit does have "impedance imbalances"--what people usually don't see is that this applies for differential mode signals only. The impedances for common mode signals are well balanced and that's what counts for good CMRR.

Having balanced impedances for audio signals wouldn't hurt, but with a good driver I wouldn't expect any misbehaviour.

The "trick" to balance the differential mode impedances seriously degrades CMRR though--once in a while something the audio industry seems to recognise (consciously or otherwise).

[Edit:]Sorry for repeating what Brad allready said--the server refused to accept my immediate edits...

Samuel

 

[clintrubber]

[quote author="bcarso"]It's true the four-R single op-amp-based diff amp is not perfect, but neither is virtually any other such, including two-op-amp designs, until things get pretty hairy. But most of the time it, and they, are adequate.[/quote]
Agreed, the schematic above kind of neglects the issue, but since it's been used like that in various designs and was doiung fine (the G-SSL and for that matter, in the 'M'-circuit as well) it was OK enough for me. And as said, cables will be short & most likely unbalanced.

The type of circuit in question has a different normal-mode Z than common-mode Z. For a ground-reffed signal one input R is twice the other, for example. But it is the common-mode Z and common-mode gain that counts when rejecting common-mode signals, and at low-mid frequencies (sufficient op amp open-loop gain) these are both equal. I've seen articles (or for that matter short notes*, a la "Ideas for Design"), that purported to discover and then correct deficiencies in the classic circuit, and they have been anywhere from misleading to flat wrong.

Sounds like you've been seeing the same EDN-idea ! :wink:

Those little notes are frustrating, because most of the time there is no Letters section and no way to point out errors. Submit another note that says "Old circuit still works, as usual" ?---not likely to see the light of day.

I see your problem. In general, not being able to respond like we do here is often missed when using 'other kind of media'.

Bye,

Peter

 

[clintrubber]

[quote author="Samuel Groner"]

I recall a short article about simply redimensioning the circuit to solve the incorrectness, but I see nobody doing it.

I often hear that the "four resistors and one opamp" circuit does have "impedance imbalances"--what people usually don't see is that this applies for differential mode signals only. The impedances for common mode signals are well balanced and that's what counts for good CMRR.

Samuel[/quote]

Oops, right, I guess you're very right !
Thanks for reminding, it's easily forgotten.

And cheers :thumb: to the Swiss BTW, just returned from a very nice gig, The Young Gods playing Paradiso in Amsterdam. No idea if you know them or how well known they are in your country.

Bye,

Peter