Global Feedback - A huge benefit for audio

[bcarso]

sinusoids

Agree that sinusoidal distortion measurements are a good place to start.

When the RF guy in EW (UK) lectured the unwashed audio masses about the supposed stupidity of HD measurements, and somehow missed noticing that IM measurements had become quite common, I pointed out in my caustic LTE that sine wave distortion did give information on IM distortion too.

If the device under test is exhibiting time-invariant nonlinearities things are relatively easy to interpret. If however there are a variety of other effects and they are of sufficient magnitude, like signal-induced self-heating effects, then the symmetry of sine waves may conceal some effects. At one point as Audio Precision migrated to a lot more surface-mount components, they started to deal with issues where the resistors became significant distortion sources. Most of these were attributed to self-heating effects, and had to be accounted for.

It's interesting that little has been mentioned about resistors in this thread, but the usual assumption about GNFB is, at least to begin with, that they are perfect. But although there are differences among types, most show not only temperature effects but also voltage-coefficient effects and excess noise generation from current flow. The latter leads to the attempt to reduce any such d.c. at zero signal, but then we have actual signals modulating the noise.

Usually these are small effects with modest disadvantages compared to the benefits of reduction of undesirable active device behavior.

 

[atmasphere]

But they're not working at constant power by any means.

Within 1db its 'good enough for government work' 😀

Whoa. Stop right there. Audible distortion is unavoidable? Data to support that remarkable claim?

Norman Crowhurst?

FWIW often .001% THD is audible... well, something is causing that brightness, and its not an FR error. Since the ear converts a number of forms of distortion into tonality and uses the higher orders to figure out how loud a sound is, it turns out that we can here trace amounts of higher orders, which is what you are likely dealing with when the distortion figure is that low. Its important to understand that we don't hear the distortion as such, we hear it as a brightness.

Here I don't agree with you Jan, as there are a lot of good recorded music, classical or jazz with good recording mike setups, quite close to live play. Not, of course, Rock concerts with powerful PA systems, and for years I don't go on that kind.

One of the best Pink Floyd LPs I've heard is a bootleg of a BBC performance- recorded in front of a Decca stereo tree live with no spotlights or overdubs. The LP is called 'Rhapsody in Pink'

I think that I understand where Atmasphere is coming from: In the early days, before we had lots of loop feedback in amplifiers, designers had to make do with what they had available, and I recall learning in electrical engineering about maximum transfer of power with matching. I can imagine that a 800 ohm speaker would have to work that way. OR, what was the damping factor of the hi Z speaker directly driven by vacuum tubes?
It is known that horns are much more forgiving of damping factor, because they have a separate damping mechanism, and virtually all serious early designs were horn loaded.
When the direct radiator speaker was designed by Rice and Kellog (sp) in about 1925, a LOW drive impedance became necessary to meet the conditions for a flat response, even with a widely varying load impedance. This is not an EFFICIENT power transfer, just a practical way to make a loudspeaker in a cabinet work reasonably well. We have forgotten its origin and nature over the decades.

Thanks for that.

What people seem to forget is that >10% THD will always sound awful and <0.001% THD will always be inaudible regardless of the distribution of the spectrum non-linear distortion products. 0.001% is -100dB so if you are listening at anything below 100dB SPL all the distortion products are below the threshold of audibility. It's not hard to design an amplifier which has inaudible distortion for a practical power output range.

Where more information about the spectrum is required is in between those ranges where distortion is audible but not strongly unpleasant - i.e. ~0.1%-1% THD.

The other problem is when the THD is only quoted at one power output level - that tells us very little about the level of distortion for dynamic signals. It can be deceiving to quote 0.01% THD at 100Watt if it's 1% at 1Watt. Likewise an amplifier 10% THD at 100W could be 0.001% at 1W and 10W while another amp quoted as 0.3% THD at 100W could be 0.3% at 1W and 10W. The one which looks worse on paper (10% @ 100W) is probably a better amplifier for practical home HiFi use since it is cleaner at low power.



Such subjective observations such as 'image' and 'depth' are merely a grouping of the various objective qualities of a system.

e.g.
'This tweeter sounds harsh' could mean 'This tweeter has an awful peaking frequency response' or 'this tweeter has horrible levels of 3rd and 5th order distortion' or 'i'm used to a tweeter which rolls off hard at 10kHz so any tweeter which is flat to 20kHz sounds harsh to me'.

'This system has great imaging' could simply mean 'this system has flat frequency response, low non-linear distortion and the levels of the left and right speakers match well'.

'This system is lively and engaging' could mean 'this system has a rising response which I prefer because I have hearing loss above 5KHz' or 'this system has high non-linear distortion which I am currently enjoying as false detail but in 2 hours I will perceive as distortion and listening fatigue'.

The problem is that it is very difficult to demonstrate such subjective descriptions of since they are broad and often are used to describe multiple unrelated technical deficiencies of a system. The ear is easily fooled and multiple unrelated sources of distortions can often manifest themselves as the same type of subjective sound even though they measure very differently. The best advice I can give is to invest in a measurement setup and try to correlate the measurements to what you hear. Then you will hear what linear and non-linear distortion issues sound like and be able to describe issues without resorting to using vague non-technical jargon like 'harsh' or 'depth'.

Measurements are also important to prevent expectation bias when making either scientific or unscientific changes to a system.

Nice post! One of our customers uses a microphone in the room to measure distortion. Then he makes changes, quantifies the result and compares it to what he hears. There are a number of inherent bugs with that approach, but he does get results.

 

[bwaslo]

The idea that you'll be hearing just what the recording or mastering engineer heard, once you have your amp distortion below audibility, seems to miss the facts that you don't have his room or (probably) his speakers, nor are you likely to listen at the same volume. Nor his ears, for that matter. So really, no effing chance.

Should you get closest you can? Not a bad plan, but there's no reason to expect that all improvements in technical accuracy will get a result that improves (rather than degrades) the perceived sound nor even gets it closer to what the recording guy went for. Amp distortion is probably never going to be anywhere near the majority of the problem there.

I think you could use Diffmaker (if you actually read its directions, BTW!) to find amps that don't contribute audibly other than maybe frequency response.. I quickly found a few that nulled inaudibly (an old, modified Hafler, a UCD amp, and an ICEpower module) and a few that couldn't (an AudioSource mono and an old, possibly needing service, Kenwood integrated).

BTW, I could very easily, by ear only, tell the Hafler sound from the ICEpower - the Hafler's transformer physically hummed and the ICE uses a SMPS! We don't hear only with our ears, nor hear by ear only what goes through the speaker wires!

 

[fireworks]

Lots of input from people who view 'GNFB' as yet another black boax as if the effects had nothing to do with the topology (and hence, in some sense technology, as it's not equally easy and indeed sometimes not even possible to make completely equivalent circuits using semiconductors and tubes, also not even specific types of semis and tubes in some cases). Lots of oversimplifications lead to incorrect modeling and results.
An example - amp dominated by input stage gm. Really? As if there is only one possible type of output stage, capable of only one interdependence with the actual load. Not to mention mixing up things like single ended and class A, follower type and gain type output stages (or for that matter transformer coupled output stages). How GNFB will behave and what it will do varies GREATLY depending on where and how it's applied (with caveats that there are quite a lot of cases in which certain combinations of where and how can be shown to give bad results just by using common sense).

GNFB - the usual basic stuff it does always applies - lowers distortion and improves linearity and sensitivity to parts tolerances, at the expense of gain. BUT that is not the only expense. In particular, our ear-brain combo does not perform linearity analysis according to classical math, in stead it has evolved to tolerate some non-linearities more than others, also be fairly insensitive to some and very sensitive to others. A mathematically 'more linear' transfer function does not by any stretch necessarily translate to being perceived as more 'natural sounding.

Here are some 'rules of thumb' that I have collected during the years from experience of designing and restoring amplifiers:

1) Generally, you want your amplifier to be as linear as possible before feedback is applied.
2) The less linear the amp is without feedback, the more excess gain and bandwidth is required for it to sound good once GNFB is applied. Note I said gain and bandwidth, not gain-bandwidth product.
3) The more stages the GNFB loop contains, the more problems you can expect.
4) Do not even try to fix time delay/hysteresis problems with GNFB, in other words, avoid putting gnfb loops around, for instance, transformers or any circuit that has non-amplification related time constants, such as blocking distortion, sticky clipping and other dead zones in the transfer function like excessive crossover distortion, ESPECIALLY if the actual manifestation of dead zones depends heavily on the loading of the stage (mot notably output stages suffer from this). In some cases like output protection schemes you can't avoid this but then have to thoroughly test for stability when these cases happen, the good thing being one does not expect the amp to continue providing signal fidelity, just be able to get back to it when the error condition is removed.
5) There can be a definite NEGATIVE sum effect when applying small amounts of feedback to fairly nonlinear circuits. For instance, if a stage produces fairly large but nicely distributed levels of low order harmonics, and a few of them, the ear will tolerate or even suppress them, whereas applying GNFB in small amounts around such a stage will produce harmonics of harmonics, which can indeed sum to a higher distortion leve even though the original harmonics can be suppressed, and even worse, upset their distribution in such a way that the ear will not be able to mask them. Consider a stage that makes only 2nd and 3rd harmonics. Small levels of GNFB will produce 2nd, 3rd, 4th, 6th, 9th etc, where the 2nd and 3rd will be suppressed by GNFB in some amount, not at alll necesairly both by the same amount, which can sometimes result in changing which of them is dominant. Conspicuously, some harmonics will be missing almost entirely, where the ear expects a falling series of harmonics, either all of them or even and odd falling separately but by a similar basic monotonous rule. This sort of situation is not what you find in nature and hence not masked. This is a very simplistic example, much worse things happen with intermodulation, which is FAR more significant where perception of 'naturally sounding' is concerned.
6) Be very careful incorporating 'filtering elements' - high or low pass - in feedback loops especially if the fiter corner frequency depends on input or output impedances connected to the amp that cannot be controlled for by the designer (this problem most often happens in tube amps). There may be unwanted behavior inside the loop that you will not see after an output filter (i.e. coupling cap). This can seriously worsen cases where the action of the 'filter' is still felt at the edges of the audio band - especially on the high side (but also can lead to some strange effect in bass reproduction on the low side). While the ear's sensitivity to sound falls drastically in the top octave of the audio band, experiments with location and reverb show that our ear]b]S[/b]-brain system (not ear as in single!) can discriminate delays between the left and right ear with a precision greater than the reciprocal of the higher frequency we can hear suggests, so this is something that should be carefully taken care of when you need two identical amps such as for stereo reproduction.
7) If you can avoid giving NFB extra correction jobs to do, avoid it - for instance, rather than rely on GNFB to sort out your PSRR problems, employ separate measures for this in the circuit topology or by adding filtered or stabilized power rails where appropriate.

That sound like good advice. Did you use it in amplifiers of your own design ?

 

[SY]

Within 1db its 'good enough for government work' 😀

Looking at a typical span of impedances for a speaker, that number makes no sense to me. Can you tell me how you got it?

FWIW often .001% THD is audible...

Cite?

 

[SY]

The idea that you'll be hearing just what the recording or mastering engineer heard, once you have your amp distortion below audibility, seems to miss the facts that you don't have his room or (probably) his speakers, nor are you likely to listen at the same volume. Nor his ears, for that matter. So really, no effing chance.

I don't think anyone claimed that this was sufficient. But it takes the electronics out of the equation, allowing you to put efforts into the areas that really do matter, like source material, formats, rooms, and speakers. Using an amp as a poorly-controlled and expensive signal processor doesn't seem to me to be optimally effective.

 

[Michael Rothacher]

Since we've come back to it. Could someone point me to a good source which covers the threshold of audibility for harmonic distortion? Preferably, a source which also describes the methods employed to test the assertions. I'd like to read up on it.

Best,

Mike

P.S. I promise I Googled it first, but I'm interested in the articles that have been vetted by the experts here.

 

[SY]

I'd start with the Geddes/Lee papers.

 

[atmasphere]

Looking at a typical span of impedances for a speaker, that number makes no sense to me. Can you tell me how you got it?

By measuring the output power of one of our amps.

Cite?

I'm sure there must be a link out there somewhere but of the millions you get when wading through the hits; well you know what I mean.

More to the point, we all know the single most-often heard complaint laid that the door of solid state electronics with distortion figures like I mentioned: brightness (this is, after all, why the vacuum tubes are still in business). After all these pages we know that this is not due to a frequency response error! Its distortion, and obviously not very much of it.

This of course is at the very root of why amps that measure poorly (IOW: high THD, but mostly 2nd harmonic) can sound fine. General Electric did a study during the 1960s which showed that humans would tolerate a lot of even ordered harmonic distortion (up to 30% without complaint) but complained quite quickly when odd orders were introduced at much lower levels (less than 0.5%). I don't know though whether that study has ever been placed online. However its easy enough to prove with some basic test equipment!

I've been harping for years that we really won't make progress (or dispense with threads like this one) until the industry places the human hearing/perceptual rules at the foundation of its goals. Right now they often get ignored- and its been that way for decades.

 

[bwaslo]

I don't think anyone claimed that this was sufficient. But it takes the electronics out of the equation, allowing you to put efforts into the areas that really do matter, like source material, formats, rooms, and speakers. Using an amp as a poorly-controlled and expensive signal processor doesn't seem to me to be optimally effective.

But there is the data point that for some, SETs and their distortion can make a noticeable improvement in the stereo illusion. Yeah, it would make sense to put that elsewhere than the amp probably, but only after you've figured out what those amps are doing so you can try to duplicate (and preferably, improve apon).

In case anyone asks, I'm all in favor of tone controls and eq (by listeners, not just recording staff), too. I think the overzealous focus on trying to be pristine is what caused these to disappear from audio gear.

 

[SY]

In case anyone asks, I'm all in favor of tone controls and eq (by listeners, not just recording staff), too. I think the overzealous focus on trying to be pristine is what caused these to disappear from audio gear.

+1000

 

[Michael Rothacher]

It could be said that selecting any audio component for it's merit as a tone-control is misguided I suppose. After all, there are alternatives like actual tone controls, but everyone knows those can't be trusted in the hands of amateurs. Perhaps there are worse things. Heaven knows there are some really bad expensive loudspeakers out there, and those are probably selected to tame the warts of a really crappy room anyway.

My mom got me interested in hi-fi when I was a kid. "I Robot" was one of the first records I listened to over and over. We had one speaker behind a couch and the other probably under a side table or something like that. As I recall, the tone controls were always set for a little treble lift and a little bass cut. People like what they like. So it goes.

But I digress. I think we were saying something about feedback being good because it reduces distortion, but then we said distortion isn't really a good measure of audio quality anyway, and you can't really hear it, so...

I'm not sure I actually get the rules of our little game anyway, so I'm going to go read some stuff and build some amplifiers while we work it out 🙂

 

[jcx]

audibility is complex

a big recent contribution is the development of lossy perceptual compression

presumably studying the algorithms, motivating theory and the testing, tuning evolution would get to a level of useful knowledge re audibility

 

[Nelson Pass]

I'm not sure I actually get the rules of our little game anyway, so I'm going to go read some stuff and build some amplifiers while we work it out 🙂

It's entertainment, not dialysis. Whatever entertains, succeeds.

😎

 

[ilimzn]

1) Generally, you want your amplifier to be as linear as possible before feedback is applied.
2) The less linear the amp is without feedback, the more excess gain and bandwidth is required for it to sound good once GNFB is applied. Note I said gain and bandwidth, not gain-bandwidth product.
3) The more stages the GNFB loop contains, the more problems you can expect.
4) Do not even try to fix time delay/hysteresis problems with GNFB, in other words, avoid putting gnfb loops around, for instance, transformers or any circuit that has non-amplification related time constants, such as blocking distortion, sticky clipping and other dead zones in the transfer function like excessive crossover distortion, ESPECIALLY if the actual manifestation of dead zones depends heavily on the loading of the stage (mot notably output stages suffer from this). In some cases like output protection schemes you can't avoid this but then have to thoroughly test for stability when these cases happen, the good thing being one does not expect the amp to continue providing signal fidelity, just be able to get back to it when the error condition is removed.
5) There can be a definite NEGATIVE sum effect when applying small amounts of feedback to fairly nonlinear circuits... applying GNFB in small amounts around such a stage will produce harmonics of harmonics, which can indeed sum to a higher distortion level even though the original harmonics can be suppressed, and even worse, upset their distribution in such a way that the ear will not be able to mask them...
6) Be very careful incorporating 'filtering elements' - high or low pass - inside feedback loops especially if the fiter corner frequency depends on input or output impedances connected to the amp that cannot be controlled for by the designer (this problem most often happens in tube amps). There may be unwanted behavior inside the loop that you will not see after an output filter (i.e. coupling cap). This can seriously worsen cases where the action of the 'filter' is still felt at the edges of the audio band - especially on the high side (but also can lead to some strange effect in bass reproduction on the low side)... this is something that should be carefully taken care of when you need two identical amps such as for stereo reproduction.
7) If you can avoid giving NFB extra correction jobs to do, avoid it - for instance, rather than rely on GNFB to sort out your PSRR problems, employ separate measures for this in the circuit topology or by adding filtered or stabilized power rails where appropriate.

That sound like good advice. Did you use it in amplifiers of your own design ?

Yes, of course.
In fact, there are many cases where designs are published and you can see that some typical cases of running the amp to some extreme have not been dealt with, neither simulated or tested.
JCX I think mentioned that clipping can be handled well before a signal enters the power amp (also known as pre-clipping) and I agree, in fact it is (regrettably, only sometimes) used in pro audio, but I challenge you find me a single example of this amongst the hunderds of amps published on this site.
In fact, many designs here start off with the 'Blameless' D. Self amp, and this one is far less than blameless when it comes to clipping behavior - there is nothing to prevent the VAS transistor from going into maximum possible saturation on clipping. In fact, many amps use the same basic layout and add an output protection scheme, which shunts VAS current to the output on overload, so now the saturated VAS transistor ends up driving the load, with predictable effects on it's integrity if the condition is maintained long enough. And it only costs two diodes to implement a baker clamp...

 

[banat]

But there is the data point that for some, SETs and their distortion can make a noticeable improvement in the stereo illusion. Yeah, it would make sense to put that elsewhere than the amp probably, but only after you've figured out what those amps are doing so you can try to duplicate (and preferably, improve apon).

In case anyone asks, I'm all in favor of tone controls and eq (by listeners, not just recording staff), too. I think the overzealous focus on trying to be pristine is what caused these to disappear from audio gear.

SET distortion profile is very simple , playing at low output power level consist only from second harmonic ,higher ordered harmonics almost not exist , BTW ...
if SS industry production long time ago was not stopped development of those SIT devices today standard available SS amps sound will be way better I`m sure .

 

[chrisb]

It's entertainment, not dialysis. Whatever entertains, succeeds.

😎

how Zen of you 😉

 

[ilimzn]

But I digress. I think we were saying something about feedback being good because it reduces distortion, but then we said distortion isn't really a good measure of audio quality anyway, and you can't really hear it, so...
I'm not sure I actually get the rules of our little game anyway, so I'm going to go read some stuff and build some amplifiers while we work it out 🙂

Bwaslo I think mentioned that there was merit in single sine measurements and of course I agree with him, this would certainly be the 'first order approximation' of a complete battery of measurements. If things are wrong here, your other more sophisticated measurements certainly are not going to show better results 😛
That being said, when we say distortion, these days we usually do go at least one step further than 'the old times' when we'd think of it as a single percentage. FFT setups are possible almost as a rule if you are able to even visit this site on a computer. So, I would say we also imply some reference to a number of harmonics and their amplitudes. And though these measurements certainly imply a lot about IMD behaviour, we can also measure that much easier than it used to be. The funny thing is that even sophisticated measuring rigs tend to implement 'the traditional measurements' when it would be quite possible to do some 'non-traditional' ones using the same hardware. Here's an example - do an amplitude sweep IMD run and plot it as FFT on a spectrogram scale - the WHOLE sweep. It is VERY revealing. Similar thing with a single frequency sweep. Of course, there is the matter of interpretation, but to someone with even modest experience, some things will be obvious.

For one thing, someone in another thread recently mentioned how crossover distortion was just another form of harmonic distortion. If I was in a cartoon, I'd have little question marks popping over my head at that point. For HD, if yoi nput teo sine signals, you get harmonics of both and the IM products in all combinations. Try this with say 200Hz and 2kHz at 1:5 ratio at a level safely below clipping of the amp and look at the spectrogram (a proper number of FFT points has to be chosen to get enough time resolution in the spectrogram) and you can clearly see how IMD varies drastically depending on the 2k sine part crossing zero or not as it rides on the much larger amplitude, generating THD and IMD chirps. A regular static THD and IMD analysis will completely miss that - you'll get a number, but not know why.
BTW I'd trade every waterfall display mode for a spectrogram mode in all the available measuring suites I've seen for PC based sound cards, and even professional instruments of some fame lack one - preferably with selectable false coloring. Instead I have to generate WAV files and then display them in spectrogram mode in an audio editor (which lacks proper control over FFT point number and false coloring).

 

[ilimzn]

audibility is complex...

Well, there's a whole sub-forum on this site proving it, called Tubes/Valves. It is sometimes all to easy to get caught up in a single technology, I try to look at the synthesis of all we have available for amplification (ok, perhaps not magnetic amplifiers 😛 ), to get a grip on how to design an amplifier.
Surely no-one here (keeping in mind some here also post there) is suggesting that all these people in the Tube sub-forum are 'deaf' or 'like distortion'?!
Yet the designs there will hardly approach some 'standards of inaudibility' discussed here, by at least one or two orders of magnitude. So it seems to me, there is a lot more to learn.

 

[ilimzn]

if SS industry production long time ago was not stopped development of those SIT devices today standard available SS amps sound will be way better I`m sure .

Don't get me started 😡
I'm sure Mr. Pass would have a lot to say about it too.
Even with older not so linear SIT designs employed in standard SS configurations for amplifiers there would be one thing less to fuss about, and that is crossover distortion. Yes, they are low Gm and relatively high output impedances and have drain resistance so may require regulation of the power supplies but their 'imperfection' makes them almost impossible to turn off hence there is no real crossover region. And that's not even going into equivalents of tube circuits...