I would alter TMMs post at the beginning, though, something that probably most here are aware of "10% THD will always sound awful". Other than when listening to unchanging periodic waveforms like sinewaves people don't listen to any percent THD (and actually 10% on a sinewave signal could be something like a triangle wave or clarinet tone, sounds nice).
We might listen to music that peaks up near where a sinewave at same peak would be 10%, but that could also sound good if the peaks are typically brief (or awful if peaks are plentiful). I worked a few years in hifi repair and occaisionally when repairs were done would foolishly try some sales - customers would too often ask "how many watts was that?" referring to how loud I'd played something. How do you answer? Same confusion between testing conditions and music playing conditions as with THD, the question doesn't apply without a lot of qualification and statistical considerations.
I don't think comments on audibility of %s of distortion mean much without a lot of qualification like how the amp gets to that % distortion producing level, what the amp does in the lower levels of output, how fast distortion at test levels changes. Single levelnTHD is too crude of a yardstick, and not just in the usual popular wisdom about 'harmonic structure'. I can tqke any .001% thd amp and find a sinwave level where it measures 10%, and music has levels only in statistical sense wherw thd is meaningless.
We might listen to music that peaks up near where a sinewave at same peak would be 10%, but that could also sound good if the peaks are typically brief (or awful if peaks are plentiful). I worked a few years in hifi repair and occaisionally when repairs were done would foolishly try some sales - customers would too often ask "how many watts was that?" referring to how loud I'd played something. How do you answer? Same confusion between testing conditions and music playing conditions as with THD, the question doesn't apply without a lot of qualification and statistical considerations.
I don't think comments on audibility of %s of distortion mean much without a lot of qualification like how the amp gets to that % distortion producing level, what the amp does in the lower levels of output, how fast distortion at test levels changes. Single levelnTHD is too crude of a yardstick, and not just in the usual popular wisdom about 'harmonic structure'. I can tqke any .001% thd amp and find a sinwave level where it measures 10%, and music has levels only in statistical sense wherw thd is meaningless.
effect of distortions on perceived levels
When Wolfgang Klippel was working for Harman, he demonstrated his mirror filter, an approach to reducing the nonlinear distortions of the loudspeaker. There was no doubt that it worked---the difference with and without was easy to hear, no need for DBT. He was able to generate some impressively high SPLs from relatively small transducers, even pushed close to x max.
However, the application to low-cost powered loudspeakers, the focus for the multimedia group, was at best unclear, because the distortions attendant on the unprocessed audio made the bass in particular sound louder, and the amount of bass is the first thing most listeners notice and what they tend to value. Provided a given system didn't have really annoying steady-state distortion, or misbehavior at clipping that produced a lot of high-order harmonics or worse, some clipping was perceived as desirable overall.
Ford determined something similar when it came to clipping distortion in automotive audio, which if symmetrical on positive and negative swings generates odd-order distortions. They requested that power amplifiers Harman then supplied have the ability to reach about 10% clipping before the limiters took effect.
When Wolfgang Klippel was working for Harman, he demonstrated his mirror filter, an approach to reducing the nonlinear distortions of the loudspeaker. There was no doubt that it worked---the difference with and without was easy to hear, no need for DBT. He was able to generate some impressively high SPLs from relatively small transducers, even pushed close to x max.
However, the application to low-cost powered loudspeakers, the focus for the multimedia group, was at best unclear, because the distortions attendant on the unprocessed audio made the bass in particular sound louder, and the amount of bass is the first thing most listeners notice and what they tend to value. Provided a given system didn't have really annoying steady-state distortion, or misbehavior at clipping that produced a lot of high-order harmonics or worse, some clipping was perceived as desirable overall.
Ford determined something similar when it came to clipping distortion in automotive audio, which if symmetrical on positive and negative swings generates odd-order distortions. They requested that power amplifiers Harman then supplied have the ability to reach about 10% clipping before the limiters took effect.
Member
Joined 2009
Paid Member
Reply to JC: I have found differences to be far smaller between amps with low distortion though
10% THD comprised purely of 2nd order distortion isn't all that objectionable either as far as things go. Sure you can hear it, but it wont make your ears melt off.
I also like TMMs post because it succinctly sums up that it doesn't matter what your harmonic progression looks like as long as all the harmonics are below the threshold of audibility.
For some reason the anti low THD brigade always seem to talk about this as if its a deal breaker, but if you cannot hear any of the harmonics then their progression matters zilch.
I also like TMMs post because it succinctly sums up that it doesn't matter what your harmonic progression looks like as long as all the harmonics are below the threshold of audibility.
For some reason the anti low THD brigade always seem to talk about this as if its a deal breaker, but if you cannot hear any of the harmonics then their progression matters zilch.
Excellent posts by Bonsai (a few pages back, about the state of high-end audio) and TMM, also a lot of good points from many contributors in the
last few pages, in particular SY mentioning the different sound of
measurably very close amplifiers being a combination of a lot of subtle
and varied mechanisms relating to frequency, time/phase and amplitude
related issues.
The fact is, amps with very close measurements that sound very accurate on
their own do have differences when compared. Even when measurements (and not the simple ones like raw THD either) would suggest otherwise. Although they can be subtle, they are percieved with repeatibility even when some of the rather large and not that uncommon shortcomings of the individual ear and/or adapted brain are eliminated. One could in fact put forth the hypothesis that as more obvious inacuracies get eliminated, the subtler
effects count more, but I will get to this later on.
First I will dare to slightly correct SY on one point. Feedback as a tool does not eliminate the quirks of the electronics, but rather reduces them by some ratio to a level which is deemed insignifficant compared to the signal of interest. I am pointing this out because it seems to me this is one of the main pitfalls in applying NFB. The only way the performance of a system with feedback applied is defined only by the feedback, is if it has infinite gain and bandwidth before feedback is applied (although in some cases one could argue that the bandwidth requirements could be relaxed somewhat 😛 ), and this is clearly only a mathematical concept, impossible in real life.
There is also another pitfall, and this is that our calculations and simulations are based on mathematical MODELS, not the real world. We describe the behavior of the amplifier to some order of accuracy, and at some point decide to disregard higher order contributions. In most cases this is indeed what makes simulation and calculation practical in the first place. However, this is again not the real world. In the real world, we build amplifiers and then measure to verify that the actual thing comes 'close enough' to what was envisioned as a simplified abstract. Even at this level of abstraction a competent engineer can see that applying NFB is the logical, simplest and cheapest - i.e. best bang for the buck/effort - approach to building some amplifiers, but also indeed, not all of them.
Yet we still get NFB applied around output transformers, even though feedback theory tells us that hysteresis plus time constant in a NFB system gets you an oscillator. And don't let me get started on negative resistances 😛
The common thing, and my point here, is: as soon as you are not dealing
with mathematical perfections but rather with approximations, the higher
order effects take place that you have 'disregarded', i.e. you have to
decide what the 'acceptable' in 'reduce problems to an acceptable level'
is. And while there is a lot of data on this, even many engineers remain
ignorant of it, let alone the general public - helped along nicely by
examples given in Bonsai's and TMM's posts.
But, there is also an open question here. While there are good guidelines
on how to build accurate amps as well as how to verify performance (and
today we can measure so much more than static THD and the like, lots of
which was not practical even 10 years ago let alone 40 or 50 years ago), I
am sure everyone can come up wih an example of an amplifier that sounds
remarkably accurate even though (some) of it's measurements do not suggest
it should. Many have said the reason for them being here is trying new
things because it is fun, and this would be one of my reasons for being
involved in audio, but also, one of the reasons are cases like I just described - and this reason for me is the question WHY this is. I will say it again - we can use well proven approximations to build amplifiers that sound accurate, by reducing some effects to the point where they can be disregarded, but then there are amplifiers where these effects do indeed exist far more than can ever be disregarded, yet still sound accurate - one of the reasons being, the ear-brain complex does not operate ONLY along our approximations, and it is a VERY complex system, with a lot more research to be done on how we perceive what.
I would like to slightly correct (I think) Jan, regarding Cleever. I would agree that concluding an amplifier must have the same distortion pattern as the human ear would be a logical fallacy. However, the data he bases his findings on does hold water and indeed the ear does produce distortions which the brain cancels, somewhat like a complex adaptive filter. Therefore it does not follow that an amplifier MUST have the same distortion characteristics, but it does stand to reason to expect that when the harmonics do indeed fall within limits of perceptibility, the closer the distortion pattern is to what the brain eliminates as a matter of course, the more likely the brain will remove it and perceive it less.
From my experience, which may well be limited compared to other's here, amplifiers with monotonously falling amplitude of harmonics with harmonic order, sound very clean - and in fact it seems that the even and odd harmonics are to a point treated independently. If they fall monotonously each on their own but start at different levels, the perception is between more relaxed (even low order dominant) and more analytical (odd low order dominant). Subtle violations do not seem to be a problem but, say, lower second, higher fourth, in presence of falling odd order is not per se objectionable but more along the lines of 'cant put a finger on it but something is not quite right' depending on amplitude of said harmonics. As I said ion another post, almost no second, dominant third, lower fourth, missing fifth, then present sixth at around 0.1% did not sound objectionable but just odd and wrong.
Transient behavior of harmonic distribution is a whole other matter, as TMM pointed out, but an important one. In any given system it is actually very likely some clipping will occur on transients, which means that even before that happens, one will get into a region where THD and especially IMD crosses into perceptibility (even accounting for the fact that the brain is far more forgiving on transients than for steady state distortion). IMHO the ability to hold a certain THD profile is important in these cases, and just behind clipping behavior (which I have experienced to often be the make or break of 'amplifier sound'). Consider that modern music production is almost as a matter of course compressed and is routinely pre-clipped (ever since digital) so operation in this region is even more likely. But then, I have also experienced 'audiophiles' who could not discern clipping until it occurred nearly 50% of the whole recording 😛
Then there is the problem of 'effects' - a well known thing in producing and recording music. A slight expansion on what SY said on this, one should be well aware that 'effects' are unavoidable and start at the very beginning even by choice and placement of only two microphones, let alone anything else - even if the rest of the chain is kept immaculate. This however is a whole different discussion, and at some point one has to use the actual recording as a reference rather than the performance it was recorded from (especially in case the recording is a part of the performance, i.e. the studio is in a way one of the instruments used to get what the author wants). There is a tool that can help here but it does require some education to use, and it's called headphones 😛
Finally there is the matter of perception of 'accurate' even for a well kept ear and educated brain. I am sure many here can name examples of systems that to the ear sound exceptionally accurate on their own yet different when compared - and, further, to the point where we come into the realm of subjective taste as much as we want to remain objective. I know of several such systems that I could happily live with and that I can't say do anything wrong whatsoever, and what they do do, comes down to preference - and not gross, but rather more fine. The thing to note is, not all of these have (all) figures by the guidelines talked about here, so we are back to the question of why. And - one of the goals of our designs is surely the matter of 'system A does not do this bad, but system B does it better, so let's design C to be the synthesis of what A nd B do best' 😛
So, is GNFB a huge benefit for audio? It certainly can be, but it's not the be all and end all of audio by any means. It can be improperly applied and even though there is a wealth of knowledge how to avoid this, there are applications where the rationale for it being applied is at best stretched if not flawed, where one could say there are alternatives but the designer reached for that tool not being acquainted to other, possibly better suited ones. And then, there are the examples where things appear to work just fine without it even though measurements do not suggest that (and let me reiterate that today we can measure a lot more, but correlation with perception is still a field where a lot more research is required, and many still use old methods shown not to be sufficiently relevant, be that out of ignorance or standards that have not been updated too long). The question here is, how is that possible, and I believe there is work to be done here, even though one could argue that making an accurate amp is a solved problem - for one, it offers the sneaking suspicion that there are more ways to make an accurate sounding amp, and it plugs the holes heavily exploited by salesmen and various spin artists, akin to the 'god of the gaps' reflex, where, when you don't know what and how it was done, it's god - in this case it's the devil of the GNFB, i.e. when it does not sound as expected, it must be because (G)NFB was used.
last few pages, in particular SY mentioning the different sound of
measurably very close amplifiers being a combination of a lot of subtle
and varied mechanisms relating to frequency, time/phase and amplitude
related issues.
The fact is, amps with very close measurements that sound very accurate on
their own do have differences when compared. Even when measurements (and not the simple ones like raw THD either) would suggest otherwise. Although they can be subtle, they are percieved with repeatibility even when some of the rather large and not that uncommon shortcomings of the individual ear and/or adapted brain are eliminated. One could in fact put forth the hypothesis that as more obvious inacuracies get eliminated, the subtler
effects count more, but I will get to this later on.
First I will dare to slightly correct SY on one point. Feedback as a tool does not eliminate the quirks of the electronics, but rather reduces them by some ratio to a level which is deemed insignifficant compared to the signal of interest. I am pointing this out because it seems to me this is one of the main pitfalls in applying NFB. The only way the performance of a system with feedback applied is defined only by the feedback, is if it has infinite gain and bandwidth before feedback is applied (although in some cases one could argue that the bandwidth requirements could be relaxed somewhat 😛 ), and this is clearly only a mathematical concept, impossible in real life.
There is also another pitfall, and this is that our calculations and simulations are based on mathematical MODELS, not the real world. We describe the behavior of the amplifier to some order of accuracy, and at some point decide to disregard higher order contributions. In most cases this is indeed what makes simulation and calculation practical in the first place. However, this is again not the real world. In the real world, we build amplifiers and then measure to verify that the actual thing comes 'close enough' to what was envisioned as a simplified abstract. Even at this level of abstraction a competent engineer can see that applying NFB is the logical, simplest and cheapest - i.e. best bang for the buck/effort - approach to building some amplifiers, but also indeed, not all of them.
Yet we still get NFB applied around output transformers, even though feedback theory tells us that hysteresis plus time constant in a NFB system gets you an oscillator. And don't let me get started on negative resistances 😛
The common thing, and my point here, is: as soon as you are not dealing
with mathematical perfections but rather with approximations, the higher
order effects take place that you have 'disregarded', i.e. you have to
decide what the 'acceptable' in 'reduce problems to an acceptable level'
is. And while there is a lot of data on this, even many engineers remain
ignorant of it, let alone the general public - helped along nicely by
examples given in Bonsai's and TMM's posts.
But, there is also an open question here. While there are good guidelines
on how to build accurate amps as well as how to verify performance (and
today we can measure so much more than static THD and the like, lots of
which was not practical even 10 years ago let alone 40 or 50 years ago), I
am sure everyone can come up wih an example of an amplifier that sounds
remarkably accurate even though (some) of it's measurements do not suggest
it should. Many have said the reason for them being here is trying new
things because it is fun, and this would be one of my reasons for being
involved in audio, but also, one of the reasons are cases like I just described - and this reason for me is the question WHY this is. I will say it again - we can use well proven approximations to build amplifiers that sound accurate, by reducing some effects to the point where they can be disregarded, but then there are amplifiers where these effects do indeed exist far more than can ever be disregarded, yet still sound accurate - one of the reasons being, the ear-brain complex does not operate ONLY along our approximations, and it is a VERY complex system, with a lot more research to be done on how we perceive what.
I would like to slightly correct (I think) Jan, regarding Cleever. I would agree that concluding an amplifier must have the same distortion pattern as the human ear would be a logical fallacy. However, the data he bases his findings on does hold water and indeed the ear does produce distortions which the brain cancels, somewhat like a complex adaptive filter. Therefore it does not follow that an amplifier MUST have the same distortion characteristics, but it does stand to reason to expect that when the harmonics do indeed fall within limits of perceptibility, the closer the distortion pattern is to what the brain eliminates as a matter of course, the more likely the brain will remove it and perceive it less.
From my experience, which may well be limited compared to other's here, amplifiers with monotonously falling amplitude of harmonics with harmonic order, sound very clean - and in fact it seems that the even and odd harmonics are to a point treated independently. If they fall monotonously each on their own but start at different levels, the perception is between more relaxed (even low order dominant) and more analytical (odd low order dominant). Subtle violations do not seem to be a problem but, say, lower second, higher fourth, in presence of falling odd order is not per se objectionable but more along the lines of 'cant put a finger on it but something is not quite right' depending on amplitude of said harmonics. As I said ion another post, almost no second, dominant third, lower fourth, missing fifth, then present sixth at around 0.1% did not sound objectionable but just odd and wrong.
Transient behavior of harmonic distribution is a whole other matter, as TMM pointed out, but an important one. In any given system it is actually very likely some clipping will occur on transients, which means that even before that happens, one will get into a region where THD and especially IMD crosses into perceptibility (even accounting for the fact that the brain is far more forgiving on transients than for steady state distortion). IMHO the ability to hold a certain THD profile is important in these cases, and just behind clipping behavior (which I have experienced to often be the make or break of 'amplifier sound'). Consider that modern music production is almost as a matter of course compressed and is routinely pre-clipped (ever since digital) so operation in this region is even more likely. But then, I have also experienced 'audiophiles' who could not discern clipping until it occurred nearly 50% of the whole recording 😛
Then there is the problem of 'effects' - a well known thing in producing and recording music. A slight expansion on what SY said on this, one should be well aware that 'effects' are unavoidable and start at the very beginning even by choice and placement of only two microphones, let alone anything else - even if the rest of the chain is kept immaculate. This however is a whole different discussion, and at some point one has to use the actual recording as a reference rather than the performance it was recorded from (especially in case the recording is a part of the performance, i.e. the studio is in a way one of the instruments used to get what the author wants). There is a tool that can help here but it does require some education to use, and it's called headphones 😛
Finally there is the matter of perception of 'accurate' even for a well kept ear and educated brain. I am sure many here can name examples of systems that to the ear sound exceptionally accurate on their own yet different when compared - and, further, to the point where we come into the realm of subjective taste as much as we want to remain objective. I know of several such systems that I could happily live with and that I can't say do anything wrong whatsoever, and what they do do, comes down to preference - and not gross, but rather more fine. The thing to note is, not all of these have (all) figures by the guidelines talked about here, so we are back to the question of why. And - one of the goals of our designs is surely the matter of 'system A does not do this bad, but system B does it better, so let's design C to be the synthesis of what A nd B do best' 😛
So, is GNFB a huge benefit for audio? It certainly can be, but it's not the be all and end all of audio by any means. It can be improperly applied and even though there is a wealth of knowledge how to avoid this, there are applications where the rationale for it being applied is at best stretched if not flawed, where one could say there are alternatives but the designer reached for that tool not being acquainted to other, possibly better suited ones. And then, there are the examples where things appear to work just fine without it even though measurements do not suggest that (and let me reiterate that today we can measure a lot more, but correlation with perception is still a field where a lot more research is required, and many still use old methods shown not to be sufficiently relevant, be that out of ignorance or standards that have not been updated too long). The question here is, how is that possible, and I believe there is work to be done here, even though one could argue that making an accurate amp is a solved problem - for one, it offers the sneaking suspicion that there are more ways to make an accurate sounding amp, and it plugs the holes heavily exploited by salesmen and various spin artists, akin to the 'god of the gaps' reflex, where, when you don't know what and how it was done, it's god - in this case it's the devil of the GNFB, i.e. when it does not sound as expected, it must be because (G)NFB was used.
Last edited:
+1 however:
This is not a trivial matter - we do not listen to single or dual or triple sines, but a jumble of tones superimposed on each other, with an ear that varies in sensitivity both with frequency and amplitude and while our amps etc work linearly, the ear works more like fixed resolution mantissa floating point.
So consider what happens with a subtle 3kHz fundamental riding on 20-30dB or so higher bass note - the transfer characteristic of the amp is not the same when this combo is right below clipping and the 3kHz plus harmonics is distorted differently than around zero to begin with, then add intermodulation. Then get just slightly into clipping so the 3k is clipped but the bass note is not. And then go further until both are.
There is a whole jumble of stuff going on there. In real life amps WILL clip and indeed WILL work at levels where imperceptibility of harmonics cannot be guaranteed, even if it's 'only' transients. Consider that what we hear as transients looks rather more transient on a linear scale and this is what the amp does. I remember once talking to an electronics engineer working in industrial control, that said, surely it's not that difficult to get adequate performance for that sort of puny bandwidth, so I said, sure, then make your systems perform accurately within that bandwidth to 0.1% or less across 6-7 orders of signal magnitude because although most of the time it does not matter, sometimes it does so your system has to be able to do it all the time, and we'll talk 😛
Yet, just as an example, more often than not our amps are simulated and indeed measured on a resistor load, knowing full well it's not going to see anything like that resistor on it's output. Erroneous assumptions are made well before GNFB is applied 😛
Food for thought.
Last edited:
what a morass of fuzzy and wrong stuff we have been trying to debunk
Cheever?, simple sines, resistive loads only...
clipping behavior may be important - but input limiting is already discussed in Self, Cordell books, and real speaker loads current peak estimates are there too
lets at least start from the level of their books, the settled debates in JAES of the 1970s-80s
toss in a read of Audio Precision's whitepapers, user manuals, review articles by Hofer, Cabot, Dunn...
Cheever?, simple sines, resistive loads only...
clipping behavior may be important - but input limiting is already discussed in Self, Cordell books, and real speaker loads current peak estimates are there too
lets at least start from the level of their books, the settled debates in JAES of the 1970s-80s
toss in a read of Audio Precision's whitepapers, user manuals, review articles by Hofer, Cabot, Dunn...
Last edited:
Actually I think 2nd, as it manifests dynamically, is more perceptible than 3rd, and the reason may to some extent account for why some people seem to like it. It is not only because it makes some material sound pleasingly different in timbre, in the case of very sparse textures like solo vocals with minimal accompaniment. That is certainly one effect, and it can be approximated by a tone control (but without the direct dependence on level of a nonlinearity-generated distortion).
But usually the nonlinearities in a system that give rise to the 2nd harmonic also generate a low-frequency signal that, were there no highpass filtering, would manifest as a d.c. offset---a change in air pressure. Of course there always is something that highpasses, even if one d.c. couples all the way to the woofer, by virtue of a less-than-perfectly sealed listening room, or just the low-frequency cutoff of a ported system. But I think the "oomph" effect may explain why some listeners sense some sort of greater realism with significant second. It also may explain some of the absolute polarity sensitivities people claim to have, the so-called Wood Effect (no relation).
I have a friend who was proud that he designed a woofer that had second but minimized third. But second can be eliminated fairly well in the driver by proper motor design, and third is inevitable at sufficiently high levels as one approaches x max. I don't believe he had the Hiraga-esque structure of distortion products in mind---he just thought second was automatically less objectionable than third.
And yes---clipping can be avoided readily, or at least done cleanly and with very fast recovery---although with audiophiles showing off their systems, it probably rarely is. A clipping indicator on things makes sense, but it would probably appall many people to see how frequently it went off.
My amps do not clip under normal operation and have been sized accordingly to suit my needs, then again that's the luxury of DIYaudio.
Subjectivists use the word "dry" to describe an accurate, low distortion amp like the Self and Carver designs. The falsely lauded Mr Hiraga has made a "warm sounding" (meaning distorted) class A amp where the VAS stage has drastically low gain. Warm ?, yes the room certainly is after you waste massive amounts of power from the mains. The Pass A40 and the Linsley Hood class A designs were better IMO.
I use a Hugh Dean Naksa 125 which is class AB at 120 mA Iq . A good and sane compromise.
You may like the way Hiraga has internal rather than global feedback. Whatever.
My LTSPICE simulations of Hiraga also show poor damping factor.
Linsley Hood likes singleton over diff amp and this may have merit too for prevalence of second harmonic distortion. Minimal distortion should be the goal, as pro audio wants.
You can bet mastering suites DO NOT use Hiraga's power hungry and esoteric designs.
I am satisfied my Naksa 125 is a more than adequate well-performing, reliable amp.
I use a Hugh Dean Naksa 125 which is class AB at 120 mA Iq . A good and sane compromise.
You may like the way Hiraga has internal rather than global feedback. Whatever.
My LTSPICE simulations of Hiraga also show poor damping factor.
Linsley Hood likes singleton over diff amp and this may have merit too for prevalence of second harmonic distortion. Minimal distortion should be the goal, as pro audio wants.
You can bet mastering suites DO NOT use Hiraga's power hungry and esoteric designs.
I am satisfied my Naksa 125 is a more than adequate well-performing, reliable amp.
Reading and mulling over the last several posts, I come to think that we really understand what we need to do to build an amp that for all intends and purposes is perfect. So, why don't we do that? Why are not all amplifiers on the market perfect?
A few possible reasons suggest itself:
- It would be too expensive and the market wouldn't bear it. Possibly for some part of the market, but with lots of people routinely forking out 10k+ for an amp, at least the hi-end segment should be perfect, no exceptions. And we know that is not the case by a long shot.
- The amp being perfect or not is not a deciding factor in peoples' decision to buy a specific amp. This certainly is a factor for some part of the market, and is no different from the markets for cars, watches, smartphones and other 'fashion' items that are bought for many different personal reasons not necessarily because they are the 'perfect solution' to the need. In fact, some of those items are bought for no reason than that the owner wants one, independent of how well it does perform its function. So here is scope for amps that are not very good yet sell because of other attributes.
- The current crop of amps in the market place, while not perfect in this context, are 'good enough' to give their owner pleasure and satisfaction. It is only a small subset of humanity, us audiophreaks, that demand perfection where everybody else is happy with what they already have. That would put us in a position like the F1 driver complaining how all these other people can stand their regular cars with all their obvious flaws, while those car owners have no idea what's he griping about.
I am sure there are other possibilities you can come up with.
Jan
A few possible reasons suggest itself:
- It would be too expensive and the market wouldn't bear it. Possibly for some part of the market, but with lots of people routinely forking out 10k+ for an amp, at least the hi-end segment should be perfect, no exceptions. And we know that is not the case by a long shot.
- The amp being perfect or not is not a deciding factor in peoples' decision to buy a specific amp. This certainly is a factor for some part of the market, and is no different from the markets for cars, watches, smartphones and other 'fashion' items that are bought for many different personal reasons not necessarily because they are the 'perfect solution' to the need. In fact, some of those items are bought for no reason than that the owner wants one, independent of how well it does perform its function. So here is scope for amps that are not very good yet sell because of other attributes.
- The current crop of amps in the market place, while not perfect in this context, are 'good enough' to give their owner pleasure and satisfaction. It is only a small subset of humanity, us audiophreaks, that demand perfection where everybody else is happy with what they already have. That would put us in a position like the F1 driver complaining how all these other people can stand their regular cars with all their obvious flaws, while those car owners have no idea what's he griping about.
I am sure there are other possibilities you can come up with.
Jan
Jcdrisc,
I would classify the Self amp as both "dry" and "dull", not my favorite tone control 😛. I use a zero gnfb preamp that measures well but is neither dull nor dry. The balance seems to be between very high gnfb and vast amounts of compensation100pf and up which seems to lead to a dull sound, or very well conceived for each design amounts of gnfb and minimal compensation which just always seems to sound better to my ears. I've never used more than 10pf on a low cob vas stage these days and if that doesn't suffice I have too much open loop gain or insufficient bandwidth. There is a benefit to large amounts of local feedback, as it also linearizes each device by making its operation more dependant on the passive components used, it also further linearizes the closed loop stability leading to IMHO a better sound and a sound that is consistent over device temperature variations.
Colin
I would classify the Self amp as both "dry" and "dull", not my favorite tone control 😛. I use a zero gnfb preamp that measures well but is neither dull nor dry. The balance seems to be between very high gnfb and vast amounts of compensation100pf and up which seems to lead to a dull sound, or very well conceived for each design amounts of gnfb and minimal compensation which just always seems to sound better to my ears. I've never used more than 10pf on a low cob vas stage these days and if that doesn't suffice I have too much open loop gain or insufficient bandwidth. There is a benefit to large amounts of local feedback, as it also linearizes each device by making its operation more dependant on the passive components used, it also further linearizes the closed loop stability leading to IMHO a better sound and a sound that is consistent over device temperature variations.
Colin
Further to all the recent posts. Let us assume that an mp that has deviations from perfection that are less than -100dB under a set of conditions.
This is the set of conditions:
- Over the whole intended level and frequency range;
- Over the whole intended power range with all intended loads.
This would be some kind of a 4-dimensional 'power cube'.
Now, assuming this is a good goal, how would we, with measurements, ascertain that we are there? I can see some measurement challenges; for instance, I would really be able to measure distortion over a signal cycle.
I can envision a graph showing on the x-axis phase from 0 to 360 degree, and on the y-axis distortion, for a specific signal level, frequency and load.
It could be as simple as showing (input * gain) - output over a cycle; should be a no-brainer with all our ADC-DAC/PC based equipment*.
Then add another dimension like frequency, or level, so you get a plane.
Then a family of planes for different expected load impedance and phase shifts.
If we do all this, and all of those data planes are below -100dB re: input signal, can we then conclude we have the perfect amplifier?
* I think Scott would call that differential gain - Scott how do you guys measure that? NWA?
Jan
This is the set of conditions:
- Over the whole intended level and frequency range;
- Over the whole intended power range with all intended loads.
This would be some kind of a 4-dimensional 'power cube'.
Now, assuming this is a good goal, how would we, with measurements, ascertain that we are there? I can see some measurement challenges; for instance, I would really be able to measure distortion over a signal cycle.
I can envision a graph showing on the x-axis phase from 0 to 360 degree, and on the y-axis distortion, for a specific signal level, frequency and load.
It could be as simple as showing (input * gain) - output over a cycle; should be a no-brainer with all our ADC-DAC/PC based equipment*.
Then add another dimension like frequency, or level, so you get a plane.
Then a family of planes for different expected load impedance and phase shifts.
If we do all this, and all of those data planes are below -100dB re: input signal, can we then conclude we have the perfect amplifier?
* I think Scott would call that differential gain - Scott how do you guys measure that? NWA?
Jan
Last edited:
I though this things yesterday and I wrote it on my facebook wall 😛
It time consuming for simulation and measuring. Why not we choose a compromise target, like THD as low as possible and harmonic profile that we like, ex. monotonic?
This 'target' is different for everyone, so can't be used as a 'target' to develop something.
Jan
Please be gentle with my overly simplistic analysis, but if one's goal is linearity and transparency, then it seems to me that mundane, sine wave measurements can at least show that a circuit has no obvious faults that could be a problem with more complex inputs.
This, coupled with all of the other, well known techniques for making a circuit work well, such as removing out of band signals from the input, isolating the output from similar out of band signals, and avoiding predictable problems due to signal induced currents in the ground system and the power supply system, should be enough to assure good overall performance.
Sure, simple sine wave tests aren't the be-all end-all of tests, but since they can identify a lot of nonlinearities, and the ones that might be tough to illuminate can be solved by good basic design as mentioned above, I think we can do this predictably now.
I've been working with an APx-555 for a few weeks, and given its ridiculously low distortion floor, a lot of problems that were relegated to 'folklore' or other tough to verify opinions/observations are now able to be measured directly. For example, some bad samples of film caps can be identified specifically and directly, and not just as "oh, it's a Mylar cap, so it's bad", as that's not what the instrument shows.
So, I think the dimensions of your N-space of 'goodness' could be as simple as mundane sine wave tests over level and frequency, and that, along with the basic prerequisites of making a proper device, could be sufficient to state whether a device sounds transparent or not.
This, coupled with all of the other, well known techniques for making a circuit work well, such as removing out of band signals from the input, isolating the output from similar out of band signals, and avoiding predictable problems due to signal induced currents in the ground system and the power supply system, should be enough to assure good overall performance.
Sure, simple sine wave tests aren't the be-all end-all of tests, but since they can identify a lot of nonlinearities, and the ones that might be tough to illuminate can be solved by good basic design as mentioned above, I think we can do this predictably now.
I've been working with an APx-555 for a few weeks, and given its ridiculously low distortion floor, a lot of problems that were relegated to 'folklore' or other tough to verify opinions/observations are now able to be measured directly. For example, some bad samples of film caps can be identified specifically and directly, and not just as "oh, it's a Mylar cap, so it's bad", as that's not what the instrument shows.
So, I think the dimensions of your N-space of 'goodness' could be as simple as mundane sine wave tests over level and frequency, and that, along with the basic prerequisites of making a proper device, could be sufficient to state whether a device sounds transparent or not.
