Forced Class A opamp

[bcarso]

[quote author="buttachunk"]The MC33078 output stage appears to be Class A, though swinging to the negative rail may still be beneficial;

www.onsemi.com/pub/Collateral/MC33078-D.PDF[/quote]

I didn't see it class A as such---it's actually pretty hard to judge just from a topology, without values and transistor areas, etc. It is at least class AB when the current out is sufficient, since the quiescent I is quite a bit less than the max output I.

PRR had some typically trenchant remarks about the various forms of class A etc. in some not-too-old thread, but I forget where.

 

[featherpillow]

PRR had some typically trenchant remarks about the various forms of class A

:grin:
Ha Ha! Good ol' PRR...we sure do appreciate him!

I was reading some of D. Self's work before my house burned down, and he doesn't seem terribly fond of Class A either. He says the distortions that come about as a result of the ouput pair shifting into AB when under load is quite jagged and ugly.

He seems to favor a well-compensated Class B output stage.

Here's what I haven't been able to figure out thus far: If a complementary pair biassed for Class A shifts in and out of AB under load, then won't a complementary pair biassed for AB shift in and out of Class B under load as well? And given that, will a Class B stage shift in and out of...class BC?

As I understand it, these types of loading distortions are pretty gross, albeit extremely small in amplitude. So, does it really even matter?

What gives?

 

[bcarso]

[quote author="featherpillow"]

PRR had some typically trenchant remarks about the various forms of class A

:grin:
Ha Ha! Good ol' PRR...we sure do appreciate him!

I was reading some of D. Self's work before my house burned down, and he doesn't seem terribly fond of Class A either. He says the distortions that come about as a result of the ouput pair shifting into AB when under load is quite jagged and ugly.

He seems to favor a well-compensated Class B output stage.

Here's what I haven't been able to figure out thus far: If a complementary pair biassed for Class A shifts in and out of AB under load, then won't a complementary pair biassed for AB shift in and out of Class B under load as well? And given that, will a Class B stage shift in and out of...class BC?

As I understand it, these types of loading distortions are pretty gross, albeit extremely small in amplitude. So, does it really even matter?

What gives?[/quote]

See also the Renardson site I posted a link for earlier in the thread for some views of the distortion residuals when passing from class A into B, and his interesting solution (warning: his site is part of some freebie service and there are obnoxious but not so far harmful distractions).

The crossover distortion thing is complex. The worst case clearly is when both devices are cut off and the amp has no gain at all in that region.

The next worst case is when the gain changes abruptly, when a device that has been off turns on and there is nothing else locally compensating for this (which is the usual case).

You avoid this with various forms of full brutal class A but of course the penalty is horrible efficiency.

Class B done just right, and where the upper and lower stages have matched gain, has no other region of switchover so seems better than a clumsy class AB. But with local correction maybe there are better ways still. Also a major consideration is the shift in parameters with local self-heating, which get significant at low audio frequencies and cannot be corrected by temp sensors on heatsinks or even glued to the device cases.

 

[cwatkins]

Last night I put in my second channel of my SSL, except for the noise being a problem (haven't found the thread about doing two in one chassis.)
I had been waiting to shoundcheck them, because I had a brown-dog with two AD797's class A's (I'll have to look at the resistor tonight) in the spot
where the tl072 goes.

been meaning to try this stuff out again as I did alot of the testing last year in relation to the headphone amps, which is what landed me here in the first place.

But I'm curious NYD, as far as looking for changes and regulating them back, like I'm about to test this with channels on my ramsa WRT820, it's been my plan for awhile, eq included and at 8 opamps a channel, I'm not going to use dual-brown-dogged ad797's class A's unless I can absolutely prove by ear and on paper that it's better, so, what do you look for,
total volume of the channel being different, distortion?
I mean I can hear a tone difference and match volumes and say, ok, it does or doesn't sound different, and ok, the eq is or isn't acting different, but I can't maybe say/prove/where why this is.

make sense?

 

[featherpillow]

See also the Renardson site I posted a link for earlier in the thread for some views of the distortion residuals when passing from class A into B, and his interesting solution

That's a great site! My curiousity is piqued by his alternate methods of testing for distortion, thanks for that recommendation.

 

[eikoow]

[quote author="NewYorkDave"]So, while everyone's oooh-ing and aaaahh-ing over the newfound soundstage, did anybody remember to take some actual before and after measurements? It would be interesting to compare the amount and spectra of the distortion products before and after the mods.[/quote]

We did do broadband THD+N measurements to make sure we hadn't screwed anything up, but those numbers didn't change much as a result of the class-A mod. The broadband measurement is mostly measuring noise, anyway -- except for the channels which had a bad capacitor, or where we had soldered the resistor in the wrong place, etc. That's all I was looking for (and we did find a few). We didn't have time to do spectrum analysis of the distortion products.

I think that harmonic distortion measurements are not very useful for evaluating this modification. (I think the same thing about two-tone IM measurements, but I haven't had time to document that yet.) Some laboratory harmonic distiortion measurements of class-A conversion of an AD797 can be found at http://stuff.jkc-lab.com/audio/evm/hd-fca-050822.html, These do include spectrum analysis of the distortion components.

Perhaps spectrum analysis of a more complex test signal will provide useful results. However, I don't yet know what that signal might be. I'll continue experimenting......

 

[bcarso]

In the long run you might want to look at the papers from a ways back by Theile, "Measurement of Nonlinear Distortion in a Bandlimited System, " J. Audio Eng. Soc., AES, NY, NY, vol.31, pp. 443-445, June 1983, and Small, "Total Difference Frequency Distortion: Practical Measurements, smae as above vol. 34, no. 6, pg.427, June 1986.

 

[cuelist]

:? Surely the whole point is to reduce or eliminate the audiable effects of crossover distorsion?

I am surprised that no one has yet mentioned the technique of slightly shifting the quiescent point of the opamp up or down. I believe this has been done in some of Mr Neve's designs using opamps.

The idea being that crossover distorsion is mostly audiable at low levels and by shifting the quiescent point, the low level signals are always handled by one of the halves, effectivly this is class A.

It's been a while since I looked at this but I believe it was done by a slight offset of the input(s) and nothing to the output, thus no excessive power draw like that associated with brute force resistor loading as previously discussed in this thread.

I guess I have to go and dig up my 9098 schematics and have look...

 

[Learner]

[quote author="eikoow"]
I think that harmonic distortion measurements are not very useful for evaluating this modification. (I think the same thing about two-tone IM measurements, [/quote]

May be you could try inputing a swept signal of 1hz~20Khz to measure the THD of your sweeping oscillator, then run the sweeping signal from the oscillator into the op amp without the mod.

By subtraction the THD figure of your sweeping oscillator from the signal coming out of the op amp with no mod, it should be the THD figure of the op amp.

By comparing the difference between the THD figure of the op amp done with mod and no mod this should determine if there are any artifects produced by the mod done to the op amp.

It would help to determine the effect of the mod in the frequency domain, if you compare the sweeping spectrum of signal coming out of the op amp with mod and no mod.

 

[TedF]

When this thread first started, I thought it looked interesting, so I set up some tests on the bench, operating various op-amps at about 15dB gain and pulling the outputs about. I looked at sine and square waves on scope and spectrum analyser, and measured distortion.

The op-amps tested were: TL072, TL062, 5532, and MC33078, and I had a quick look at a few others in the junk box.
The results were distinctly non startling.... Dragging current made virtually no difference to any of the chips except for making slight differences to the clip points. As one would expect, the TL072 was not very happy and showed 3rd harmonic products at low level as soon as it looked at anything less than 10Kohms. The 62 is happier. The 5532 actually showed signs of 2nd harmonic distortion at low level, but well down, and unsurprisingly the MC33078 was the best behaved; I believe that it has a predominantly class A output stage anyway.

I suspected before I started that this was a bit of a wild goose chase; after all the only thing the resistor does is to alter the conditions of the output devices, it does nothing for the internal architecture.
By altering the DC balance position of the amplifier it's entirely possible to eliminate the 3rd order products caused by instability at crossover; but it get's more complicated when there are gain controls involved! Even so, I think it's a more productive area for study; and as someone already pointed out, Rupert has done it already.

 

[bcarso]

[quote author="cuelist"]:? Surely the whole point is to reduce or eliminate the audiable effects of crossover distorsion?

I am surprised that no one has yet mentioned the technique of slightly shifting the quiescent point of the opamp up or down. I believe this has been done in some of Mr Neve's designs using opamps.

The idea being that crossover distorsion is mostly audiable at low levels and by shifting the quiescent point, the low level signals are always handled by one of the halves, effectivly this is class A.

It's been a while since I looked at this but I believe it was done by a slight offset of the input(s) and nothing to the output, thus no excessive power draw like that associated with brute force resistor loading as previously discussed in this thread.

I guess I have to go and dig up my 9098 schematics and have look...[/quote]

I worked on a woofer amp design where I assumed that the only important audible region of crossover distortion would occur at low levels, and as you suggest be appropriately masked if it occurred shifted into a higher level region. The audible results proved otherwise, probably because of the significantly higher frequency of the distortion components. The results were good enough for a fairly uncritical client and eliminated concerns about bias adjustment and temperature compensation, but wouldn't pass muster for critical listening.

I can't see where shifting the input voltage, differentially, would do anything to the output bias to first order unlesss the output was d.c. loaded to ground, in which case it would accomplish the slight shift in the crossover distortion notch location with output swing. As I believe PRR pointed out the typical crossover notching is a function of output current, not voltage per se.

Actually I think the more bothersome audible artifacts from class AB (as opposed to blatantlty sub-class B) crossover distortion are more dynamic, that is, related to the finite speed of the output devices, which get quite slow when at very low currents or turned off. Stage designs that avoid either device turning off completely regardless of output current probably do about as well as pure class A.

 

[eikoow]

[quote author="cuelist"]:I am surprised that no one has yet mentioned the technique of slightly shifting the quiescent point of the opamp up or down. .....The idea being that crossover distorsion is mostly audiable at low levels and by shifting the quiescent point, the low level signals are always handled by one of the halves... [/quote]

If you want to put an IC opamp into that condition, you've got to draw a DC current from the output at least equal to the output stage bias current. I see no other way to force all the signal current to flow through one transistor and none through the other. That DC current can flow through a resistor connected to some non-zero voltage, such as the power rail, or it can flow through the load (if the load will tolerate it). If the DC current flows throught a load connected to ground, then the DC output voltage will not be zero.

If the feedback resistor is a reasonably low value *and* there is a DC path to ground through the feedback network, then deliberately offsetting the output voltage from zero could cause enough current to flow through the feedback resistor to force the output into class A. Presumably the audio output from this stage would then be capacitor coupled to eliminate the DC offset.

It's an interesting question whether the output *always* needs to be class A, or only needs to be class A at low levels. For our modification, I tested resistor values of 15k, 6k, 3k and 2k. 15k sounded better than no resistor. Going to 6k produced a smaller but still quite noticeable improvement. Changing to 3k produced a much smaller improvement. 2k produced very little further improvement -- in fact, I wasn't sure I could hear the difference. We could tolerate the power supply draw and heat caused by using 3k resistors. With 2k resistors, both of those were marginal. So, we used 3k resistors.

I did *not* analyze every amp in the console to see what load it was driving, and what resistor value would be needed to always keep it in class A. In general, the load on most of the opamps is 5k or higher, except for the output amps. Only a few pieces of outboard gear have 600 Ohm inputs, so the load on most of the output amps should usually be 5k or higher. Thus, with 3k resistors, only a few of the amps might change into class AB at very high signal levels.

I wanted to measure whether the whole console would stay in class A with normal signal levels, or whether some amps were operating in class AB with high signal level. Joshua brought up a mix in Pro Tools that put audio on every channel, and I monitored the power supply current with a Tektronix A6302 current probe (DC to 5 MHz). It was quite stable. Prior to the class A mod, power supply current varied quite a lot with signal level, as one might expect from a console full of class AB amps.

I think that some of the improvement in the sound of the console may be due to the huge reduction in non-linear garbage on the power supply rails. Some of that garbage will be transferred into the ground system, and thus back into the audio. The usual way of making distortion measurements on a console is to drive one channel at a time, with no signal on all the other channels. A realistic measurement would require driving all (or almost all) of the channels at the same time, and we didn't do that measurement before we did the class A mod.

 

[bcarso]

[quote author="eikoow"]

I wanted to measure whether the whole console would stay in class A with normal signal levels, or whether some amps were operating in class AB with high signal level. Joshua brought up a mix in Pro Tools that put audio on every channel, and I monitored the power supply current with a Tektronix A6302 current probe (DC to 5 MHz). It was quite stable. Prior to the class A mod, power supply current varied quite a lot with signal level, as one might expect from a console full of class AB amps.

I think that some of the improvement in the sound of the console may be due to the huge reduction in non-linear garbage on the power supply rails. Some of that garbage will be transferred into the ground system, and thus back into the audio. The usual way of making distortion measurements on a console is to drive one channel at a time, with no signal on all the other channels. A realistic measurement would require driving all (or almost all) of the channels at the same time, and we didn't do that measurement before we did the class A mod.[/quote]

This is very astute IMO. The power supply comtamination is often addressed by more and more bypassing, local and separate regulation, etc. etc., but there is nothing like minimizing the disturbances to begin with!

A complex optoelectronic instrument I did years ago had a lot of high speed switching alongside very critical charge-sensitive preamps etc. I finally arrived at a design wherein virtually all of the electronics drew constant current from the power supplies. Locally to some modules, simple R-C decoupling networks took care of the residual fluctuations. The shifts and droops that would have been unacceptable downstream of those networks were alleviated by the basically constant local current drain regardless of state.

Ironically about the only remaining fluctuations were due to the few purchased modules in the system, the track-hold and the A/D converter. Had I rolled my own versions of those I could have eliminated that contribution as well, but the hellhounds were at my throat to finish the project.

 

[TedF]

I agree!
So often apparent distortion artifacts are actually poor groundpath routing.... but that's getting off the topic.

 

[eikoow]

[quote author="bcarso"]Had I rolled my own versions of those I could have eliminated that contribution as well, but the hellhounds were at my throat to finish the project.[/quote]
Ah yes..... There's so rarely enough time.

My wife has been bugging me for years now to invent the time warp machine, so she can get *her* projects done on time, but somehow I just haven't found the time to do it. :grin:

 

[Svart]

hmm, there is no doubt of your measurements but I do HEAR a marked difference in the signal.. maybe the sine/square waves don't quite tell the whole story here. I have NON-scientifically tested by having an assistant swap channelstrips out without me watching and I've been able to pick out the channels by ear 100% of the time. I have also had others pick them out as well..



:shock:

If i were to do this mod on SSL9k preamps, would I also want to mod the tl052 opamps as well? I'm thinking yes, but i've been so stressed lately I'd rather not think right now..

 

[Flatpicker]

[quote author="Svart"]...NON-scientifically tested by having an assistant swap channelstrips out without me watching and I've been able to pick out the channels by ear 100% of the time...[/quote]Well, I wouldn?t call a measuring device as sensitive as the ear connected to something as complex as the brain unscientific, but judging from yours and others aural experiences, I take it that there is a difference that has yet to be measured on something man-made. :wink:

 

[TedF]

Pls forgive my scepticism, but being able to tell the difference is not necessarily showing improvement! I can hear the difference between a transformer coupled mic amp, and a conventional '2017' type, but it's not always better.
In my design work I'm tending to go for simple descrete class A amplifiers in preference to any form of op-amp, wherever there is gain involved (I just use op-amps as impedance converters). This arrangement certainly sounds 'better' even though the analyser tells me different! :grin:

 

[Svart]

Well i mean non-scienific as in there are no tangible measurements, but only educated opinions.

and there is a difference that I do find pleasing, if it's not what we/I think it is, it is none-the-less useful and I would still like to get to the bottom of what is truly happening. I don't have the measurement devices i need to do testing on my own or else i'd be on this.

TedF, scepticism is understood as I am still skeptical myself. However there is a difference between the audio from a "forced ClassA" channel strip and a non modded strip in my setup. nothing else is different besides the mods.

I still have a sneaking suspicion that square waves and sine waves are too "simple" to actually show what is going on here and that further testing needs to be done.

:guinness:

 

[TedF]

Hi Svart, you are straying into my favourite area! I believe that sine/square testing is totally inadequate; it's useful only to compare systems in a gross way.
Operational amplifiers have an unfortunate and disturbing attribute; all the time that transient signals remain within the constraints of the gain/bandwidth, everything is fine, but the tiniest transgression out of that domain, and they can go unstable for a short moment, and introduce horrors into the signal path.... and I am certain that this is the area where most dis-satisfaction with quality originates.
With descrete circuits, a momentary overload causes a bit of a jolt and a squaring of a signal, but it is instantly corrected and does not significantly affect the feedback parameters. In an op-amp, it's very different; an overload resulting in loss of feedback can 'domino' through the architecture, rippling unwanted products for many milliseconds... and sounding 'thin', 'muddy', 'harsh',... recognise it??
It's not unreasonable to suppose that a bias on the output devices could shorten these bad effects, making the device sound better.... Yes?