crazy high rail voltage = better????

[Winston OBoogie]

integrity with such foolishness. How much time do you think he sets aside to waste?

None.  I've chatted with him several times.  There are better things he can be doing with time at his age.

I can say that he IS concerned with keeping upper (bad sounding) harmonic distortions down and isn't too concerned about small amounts of lower ones.
I suspect he's not in the zero feedback camp but, still, he knows a thing or two about what it entails to design a sonically pleasing circuit.  Yes?

If you know what you're doing, higher rails will push down whatever artifacts you have.  It isn't necessarily about having an extra 6dB of headroom in the usual sense.

Here's another approach, although a wasteful one:  Design for 6dB or more extra headroom and then put a 6dB attenuator on the output.
Noise and crap are lower by as much.  Haven't done this myself but it was done sometimes on older valve stuff.


Edit: let's define what is a "crazy high voltage".  Remember, twice the voltage is only 6dB more swing.  Older class 'A' Neve was +24V but the fact that the output was impedance coupled meant the output could swing more, closer in headroom to having a +48V rail.  If he's now using +90V, that's not a huge leap.

 

[JohnRoberts]

I get it that you like the idea of high voltage rails... I will throw you one bone in that regard. To the extent that crossover distortion is finite, scaling voltage swing up makes that proportionately less.  That said, I would just be inclined to work on getting the amp to hand over drive better at zero crossings.

I glanced at the the Cheever paper (thank you for posting it) and liked it enough to save it, presumably to read later.. That said now is probably when I should be reading all the stuff like this that I've put aside for decades (or throw it out).

The observation about human nonlinearity masking certain types of distortions sounds very likely. However this doesn't make me second guess modern (high negative feedback and moderate PS rail voltage) electronics that deliver all linearity so low it is impossible to measure without tricking the circuitry up to some high noise gain.

This does offer a plausible explanation for the "measures bad- sounds good" conundrum, but the different flavors of distortion have been generally understood for quite some time (low order-si, high order non), albeit not with such precision and certitude.

I don't find it important how a circuit is made linear as long as it truly is accurate (straight wire with gain) over the range of interest. I don't like any extra errors even if they are masked by our inherent distortion

JR

 

[Winston OBoogie]

"I get it that you like the idea of high voltage rails... I will throw you one bone in that regard. "

Thanks John.  And I get it that I am, in this thread, probably behaving like a 'Dog With A Bone'    Sorry :-[

"To the extent that crossover distortion is finite, scaling voltage swing up makes that proportionately less.  That said, I would just be inclined to work on getting the amp to hand over drive better at zero crossings."

Right.  However, I'd probably err on having zero zero crossings   ;D

"I glanced at the the Cheever paper (thank you for posting it) and liked it enough to save it, presumably to read later...'

I think it's a good one.  

"The observation about human nonlinearity masking certain types of distortions sounds very likely. '

As I mentioned in another foum (and was poo-poo'ed), to the ear's nonlinearity, we could add that of air:

http://hal.archives-ouvertes.fr/docs/00/21/95/66/PDF/ajp-jphyscol197940C860.pdf



"This [Cheever paper] does offer a plausible explanation for the "measures bad- sounds good" conundrum, but the different flavors of distortion have been generally understood for quite some time (low order-si, high order non), albeit not with such precision and certitude."


I think we have known the basics for, what, 100 years?

"I don't find it important how a circuit is made linear as long as it truly is accurate (straight wire with gain) over the range of interest. I don't like any extra errors even if they are masked by our inherent distortion"


I totally respect that approach.

Thanks John.

I shall now go back in my kennel and suck on some marrow...

John

 

[1954U1]

I've read the Cheever's doc,. interesting.. but there is something that puzzles me a bit..

Why else would state-of-the-art audio amplification designers move towards lower feedback, when advancements in output device bandwidth
allow the application of greater levels of feedback?"

This argument is always a bit risky.. we know that people do buy things like audiophile USB cables..

as soon as some feedback is applied a third harmonic appears, which is again fed back and creates sum products at f+3f,
or fourth harmonic, and 2f+3f, fifth harmonic.

I've not understood well this mechanism.
Cheever talks about the negative feedback loop as if it was a signal path going out&in out&in continuously [edit]with all the other parts of circuit being freezed[/edit]..
My electronics skills are far from being deep.. but I have some doubts that in a real circuit we can look at the feedbak loop, stop the time,
and think about what happens at the 1st tour, 2nd tour, 3rd tour.. hope that what I mean is clear..

What is more subtly hidden is sharp increase of higher harmonics as even moderate feedback is applied..
In Chapter 3 actual device measurements results of F.E.T's and B.J.T's are included and show good agreement with the calculations
in this section although slight de-generation of the FET gate drive results in some H3 and higher distortion with no feedback in place[italic is mine]

Then where are the actual measurements??

inability of even the finest instrumentation being able to quantify the difference between a Stradivarius violin and a more modest instrument

If thats true.. then no furher discussion is meaningful, because we cant prove anything..

The desing will contain harmonics at much higher levels than the same device used in a feedback amplifier, but, the harmonics
may match the ears self-generated envelope of harmonics, and result in a better TAD figure.

TAD = Total Aural Disconsonance


So, the true point of Cheever is that its better to have more distortion, if the harmonics distortions are distributed on a supposed human, fixed, TAD curve,
instead of having, with negative feedback loop, a better total harmonic distortion figure, but with the high harmonics distortion reduced less than the low harmonics distortion,
if I've understood well.
It shouldnt be so difficult to make some blind AB tests for that..

 

[JohnRoberts]

The GE (Golden Ear) vs, MR (meter reader) debate has been going on as long as I can remember and will surely go on especially when it supports merchandising silly wire and such.

I am a long time advocate of NF (negative feedback) but in nature there is no free lunch so it is useful to understand the cost/benefit.

The basic transaction involved when using NF is to trade open loop gain for improved accuracy, the cost for this improvement is the injection of an error voltage effectively in series with the amplifier + input.  This error voltage is the reciprocal of the open loop transfer function and gets multiplied by the closed loop gain.

In principle a perfectly linear open loop transfer function, would inject a linear error voltage causing a simple gain error. In practice even the best open loop transfer function requires compensation for stability to prevent oscillation. Since this compensation generally takes the form of a one pole integrator, the error voltage will be the reciprocal or the derivative of the (output) signal, above this compensation pole frequency. This is generally ignored in sine wave bench testing since the integral/derivative of sine waves is a phase shifted sine wave, so the net impact is just some modest phase shift, automatically ignored in THD analyzers. OTOH looking at complex non-sinusoidal signals results in a complex and harder to ignore error voltage. For an extreme example the error voltage generated by a triangle wave is a square wave. There is no way to add a small square wave to a triangle wave and get a pure triangle wave.

This doesn't mean that NF is bad, only that the open loop transfer function is important, loop gain margin is important, compensation is important, etc. Using an amplifier with lots of loop gain margin means these errors will be arbitrarily small. Starting with a clean transfer function means the this error voltage will be less nasty, before it is made smaller.

A final consideration regarding slew rates and gain bandwidth product, is that we don't need the speed or bandwidth for the audio passband (in fact we want to limit that), but by having a very fast circuit the stability compensation pole can be pushed out high enough to be less intrusive at audio frequencies. Note: it matters how the amplifier is made fast. Some fast S/H amps that are made fast by saturating the input stage will not be great audio performers.

So there is no free lunch but NF properly used can give us a very clean result.

JR



 

 

[abbey road d enfer]

I have some doubts that in a real circuit we can look at the feedbak loop, stop the time, and think about what happens at the 1st tour, 2nd tour, 3rd tour.. hope that what I mean is clear..

This is very clear. Unfortunately, that's the way Matti Otala's work has been presented to the believers, similar to Zenon's paradox (which demonstrates that Achilles could never outrun the turtle). The input signal hits the input stage; the feedback signal is delayed (for whatever reason, including the amplifier's delay), so the input stage is overloaded, the amp itself is overloaded, thus it cannot produce a proper feedback signal; conclusion: an amp relying on feedback to operate cannot work properly.
My dear friend, we've been living with that since the 1960's, and I reckon it's gonna last.

 

[JohnRoberts]

I have some doubts that in a real circuit we can look at the feedbak loop, stop the time, and think about what happens at the 1st tour, 2nd tour, 3rd tour.. hope that what I mean is clear..

This is very clear. Unfortunately, that's the way Matti Otala's work has been presented to the believers, similar to Zenon's paradox (which demonstrates that Achilles could never outrun the turtle). The input signal hits the input stage; the feedback signal is delayed (for whatever reason, including the amplifier's delay), so the input stage is overloaded, the amp itself is overloaded, thus it cannot produce a proper feedback signal; conclusion: an amp relying on feedback to operate cannot work properly.
My dear friend, we've been living with that since the 1960's, and I reckon it's gonna last.

The feedback path is delayed a little bit by propagation delay (finite speed of electricity and device turn on effects). The concept of dominant pole compensation is to swamp out these small delays with a well defined capacitor integrator.  While I don't accept step functions as valid naturally occurring input waveforms, real signals can be faster than poorly designed real circuits. At issue is the the input stage's instantaneous Vp-p range (Vth?). This can be a few volts in FET input amps and tens of mV in simple bipolar LTP. The addition of resistor (or inductor?) emitter degeneration buys volts of input peak handling before saturation. Again nothing new or magical. The old LM308 opamp had 2k resistors in the LTP emitters.

I recall a snarky comment from one of the old school designers when this slew induced nonsense hit the hobby press... Something to the effect that this was common knowledge to all the designers working on WWII radar circuits.  So known about decades before the '60s.

JR

 

[Winston OBoogie]

One reply to one question:

Then where are the actual measurements??

To measure this effect is little more than child's play. The measurements have been presented to death many times and by many people.  It is a real effect.  However, I think you should try it yourself as proof...or not.


Take the most linear device you can lay your hands on and find the operating point that produces the most benign transfer curve.  Ideally, your first harmonic (2nd) will be, say, 3 decades down.  The third sitting beneath that and 4th being lower still.  And so on. Higher orders will no doubt be absent unless you have sufficiently sensitive measurement capability.

Apply feedback around the device.  Look at the changes.  Some nominal amount of feedback (a few dB) will still leave you with a fairly benign and monotonic transfer.  
Beyond a point however (12dB?), things will drastically change.  Some higher harmonics will increase so much that it will require 40 - 50dB of feedback to decrease them to the level they were before feedback was applied.  You will then need more singularities (stages of gain) to realize your required amount of gain.  This will, of course, no doubt require compensation techniques and everything that goes with the territory of such a topology.  

If you have the means, build two circuits that employ these opposing methods and measure and listen to them.
Hopefully you're listening on a half decent system/speakers.
If you're on an iPod or a Mackie, all bets are off.

Anyway, pick which one you like best and use that approach in stuff you build or design.  Simple really.

Besides Matti Otala and Cheever, we should also not forget that work was being done on this in the 1950's by Del
Shorter (BBC) and Norman Crowhurst.
Peter Baxandall also observed the phenomenon.

Edit for spelling.

 

[Winston OBoogie]

Another reply to another point:

So, the true point of Cheever is that its better to have more distortion, if the harmonics distortions are distributed on a supposed human, fixed, TAD curve,

I actually feel that the point of the paper is that the usual 'lumped' T.H.D. measurement in invalid as regards meaning and correlation to the D.U.T. being a good one. 

Also, this distortion spread we're talking about is not just a 'human' condition.  It is, however, a natural one.  It occurs in air and inherently in the active device before large amounts of feedback are applied.

Another part of the problem I have with amplifiers using large amounts of n.f.b., or a non "A" class topology, is that the distortions are non monotonic.  They do not rise gently with amplitude.  Monotonicity is also is a natural phenomenon. 

 

[1954U1]

One reply to one question:

Then where are the actual measurements??

To measure this effect is little more than child's play. The measurements have been presented to death many times and by many people.  It is a real effect.  However, I think you should try it yourself as proof...or not.

I was only arguing about how the phenomenon is presented in the Cheever's paper..
I've seen the fig 3-2 on page 66[pdf's 73] of the doc.
I've only thought it was a bit unfair and un-realistic to make a test with only a single-ended circuit with an external current source as the one pictured in fig 3-1..
Excuse me for that, I know I'm a nothing in electronics experience, I try to express things that seems obvious even to me..
Anyway, yes and thanks, you're completely right.
I have [now] enough measurement instruments, listening system and raw skills to make the tests by myself, using real circuits.

The thing, in the Cheever's paper, that interested me at most is the one preparing the conclusion:

In review, global or single stage negative feedback is not required for audio
amplification if a single-ended design is chosen. The design will contain harmonics at
much higher levels than the same device used in a feedback amplifier, but, the harmonics
may match the ears self-generated envelope of harmonics, and result in a better TAD
figure. In general only two gain stages will be required, in contrast to at least five with
modest feedback level amplifiers.

Also, this distortion spread we're talking about is not just a 'human' condition.  It is, however, a natural one.  It occurs in air and inherently in the active device before large amounts of feedback are applied.

Yes, distortion spread is in nature too, not only in human ears..
Again from Cheever's doc, pag 66 [pdf's 73], he states:

Why plot 98dB? Because it is clear from the aural
harmonic envelope that the fall off matches most closely the zero-feedback data!

[the whole 98dB SPL  / 2 Watt / 2 speakers / 1 mt seems really, to me, a bit a trick to fit the data..]
Moreover:
why should we try to give the speakers a sound with a distortions distribution ultra-precisely matching that of the human ears?
There is some universal law behind our "aural harmonic envelope", so the musical instruments, the air, the speakers materials,
the without-negative-feedback electrons, all have exactly the same distortions distribution, and so its "right" to join that law?
And [I'm always referring to the paper], one thing is to state the obvious, i.e. that the 3rd and higher harmonics are unpleasant and to avoid..
another thing is to say that an amplifier should super-exactly mimic the human ear behavior in order to sound good.
[but I have to admit.. this last sentence sounds good indeed  ;D]

Another part of the problem I have with amplifiers using large amounts of n.f.b., or a non "A" class topology, is that the distortions are non monotonic.  They do not rise gently with amplitude.  Monotonicity is also is a natural phenomenon. 

I was only thinking that all these class A currents are a bit of waste, if we can do nearly the same with well designed circuits using NF..
but maybe I see your point.
How much "nearly" we can go, and how difficult it is?
Should not instead we use high voltages and only class A topologies, and stop worrying about "discrepancies" in distortions distribution?

Anyway, many thanks for let me better understand this issue.. really.

 

[Winston OBoogie]

Excuse me for that, I know I'm a nothing in electronics experience, I try to express things that seems obvious even to me..
Anyway, yes and thanks, you're completely right.

Actually, I think that most on here would not think me right    I am only expressing what I found to work for me.
I think it was in this thread that I said I was probably "pissing into the wind". 

Let me say this, I'm not an expert either.  But I try to learn from each thing I do.  Hopefully without sounding like I'm tooting my own horn, there are tens of thousands of pre amps and such that are on the market that I had a hand in.  Unfortunately, the ratio of what I consider good and what I consider so-so or 'tone boxes' errs, by a large margin, on the so-so side.
I tend to do the sort of design stuff that I'm talking about here more for fun.  However, very occasionally, when budget is not an issue, I get to do what I want. 

I have [now] enough measurement instruments, listening system and raw skills to make the tests by myself, using real circuits.

I think that's great!
The most linear stage will be a valve, and an ancient one at that.  No 12AX7's please 

I have yet to achieve good results using BJT's by themselves, although there are others that have done great work in that regard. 
I have gotten what I consider fine results with J-Fets.  I am not necessarily an original thinker but I do read and sift through the online posts and articles by people such as John Curl and then apply what I understand is being said.
 
Several years ago, I built a complimentary folded cascode circuit with zero feedback that was probably 70% Curl's design.  Although I have not seen his schematic.  That circuit is currently the only amplifier in a top mastering facility's transfer desk and was chosen over several contenders.
Another example: I have a valve mic amp that is used in, possibly, the most famous studio in the world and, besides being used for Rock/Pop, is frequently used as the main vocal amp for opera recordings.  It has switchable feedback but I hear it is used mostly in the 1dB or 7dB of feedback mode.

Moreover:
why should we try to give the speakers a sound with a distortions distribution ultra-precisely matching that of the human ears?
There is some universal law behind our "aural harmonic envelope", so the musical instruments, the air, the speakers materials,
the without-negative-feedback electrons, all have exactly the same distortions distribution, and so its "right" to join that law?
And [I'm always referring to the paper], one thing is to state the obvious, i.e. that the 3rd and higher harmonics are unpleasant and to avoid..
another thing is to say that an amplifier should super-exactly mimic the human ear behavior in order to sound good.
[but I have to admit.. this last sentence sounds good indeed  ;D]

I actually don't know that I prefer things to strictly match the human ear in order.  I also think the preference for higher 2nd or 3rd might be split fairly equally amongst the population that give a damn about such things.  I can only say that I prefer - record stuck in groove here! - low to zero levels of F.B. and less singularities that are linear and have wide bandwidth.  And, obviously, no upper orders.
I have been just as happy with complimentary or push pull circuits that canceled 2nd as I have with a topology that didn't. 


Yes, class "A" is very inefficient.  So is spending money for diminishing returns on the improvements you gain.  Some people are happy to do it though...

Best,

John

 

[Winston OBoogie]

Some interesting results of distortion tests on J-Fet stages:

http://web.telecom.cz/macura/diyaudio/jfetdist.htm