When producing open source designs, or engineering designs in general, one would like to compromise between good performance and that which can be built within the budget and skill set of one reasonably skilled. So the question I ask myself is: how do I know if a level of performance can be not be so difficult to achieve in practice as to be practically unattainable, and if the level of performance is going to be perceptible or provide value to the consumer?
Now this may be heresy to ask this question on a forum dedicated to achieving the highest levels of audio perfection, but it is good for such evaluations to exist so that designs can be out there that can provide satisfactory or even excellent performance without extreme expense or difficulty.
As a wise old engineer once said: anyone can design a bridge to stand up, but only a good civil engineer can design a bridge to barely stand up.
After doing some searching out on the net, I was trying to find how much information there is on perceptible maximum permissible THD or other measures of distortion or noise (odd or even harmonics), amplifier slew rates, perceptibility of distortion in loud signals (sine wave or music), how well speakers can reproduce sound from voltage source inputs and current source inputs, what effect does speaker resonance have on audio perception, etc.
I realize there are probably very few objective ways to answer such questions, perhaps using psychophysical tests, for example A/B testing, and a lot of the audio business is marketing. Still, there are quantifiable measures of system performance, its just that it may not be that well understood how those measures relate to perceptual differences.
Is there some kind of compilation of research or bibliography of such research? How do other engineers go about obtaining answers to these questions? I would like to build a bridge that perhaps stands up perhaps more than BARELY but not for a hundred times the required load either.
Now this may be heresy to ask this question on a forum dedicated to achieving the highest levels of audio perfection, but it is good for such evaluations to exist so that designs can be out there that can provide satisfactory or even excellent performance without extreme expense or difficulty.
As a wise old engineer once said: anyone can design a bridge to stand up, but only a good civil engineer can design a bridge to barely stand up.
After doing some searching out on the net, I was trying to find how much information there is on perceptible maximum permissible THD or other measures of distortion or noise (odd or even harmonics), amplifier slew rates, perceptibility of distortion in loud signals (sine wave or music), how well speakers can reproduce sound from voltage source inputs and current source inputs, what effect does speaker resonance have on audio perception, etc.
I realize there are probably very few objective ways to answer such questions, perhaps using psychophysical tests, for example A/B testing, and a lot of the audio business is marketing. Still, there are quantifiable measures of system performance, its just that it may not be that well understood how those measures relate to perceptual differences.
Is there some kind of compilation of research or bibliography of such research? How do other engineers go about obtaining answers to these questions? I would like to build a bridge that perhaps stands up perhaps more than BARELY but not for a hundred times the required load either.
I would expect to start with setting your engineering GOALS. It is easy to toss around terms like better or efficiency or whatever, but without a context it is like asking a chef, how much oregano should I put in something.
The needs of a high end stereo design are not like those of a high end pro audio design.
The needs of a high end stereo design are not like those of a high end pro audio design.
Well that is so true. Though I have to admit, I have learned a lot by letting out the magic smoke, so there can be a benefit as well. It just gets frustrating when you want to enjoy something and it becomes a science project.
A lot of work on perception of audio quality was done by Floyd O'Toole and Sean Olive at the Canadian National Research Centre, particularly in relation to loudspeakers. Their books make very informative reading and provide insight that can be more generally applied. Both people were later employed by Harman International for ongoing research.
"Sound Reproduction The Acoustics and Psychoacoustics of Loudspeakers and Rooms" by Floyd Toole is a fascinating read. It's really interesting to find out the results of some psychoacoustic testing. For example here is an excerpt from page 418:
This makes sense to me and kind of is consistent with something I've observed, though admittedly not as systematically as Toole. Flat frequency responses sound better on the whole, and a lower amplifier output impedance drives the speaker with a flatter response. This suggests to me that probably one of the easiest (but perhaps not most cost-effective) ways of getting flatter performance is to use more output transistor devices to achieve a lower output impedance. Of course, this generally increases the available power output, which is why probably "headroom" for power is considered desirable, but it may not just be a matter of distortion, but output impedance.
On page 452:
Thanks for the tip, it looks likes an excellent source of information.
Then he goes on to say on page 421:
This makes sense to me and kind of is consistent with something I've observed, though admittedly not as systematically as Toole. Flat frequency responses sound better on the whole, and a lower amplifier output impedance drives the speaker with a flatter response. This suggests to me that probably one of the easiest (but perhaps not most cost-effective) ways of getting flatter performance is to use more output transistor devices to achieve a lower output impedance. Of course, this generally increases the available power output, which is why probably "headroom" for power is considered desirable, but it may not just be a matter of distortion, but output impedance.
On page 452:
Also:
I kind of suspected this too. Distortion is probably important but it doesn't seem like there's a good way to know when it's important, except probably in extreme cases when there is clipping behavior, so it's hard to know when a great deal of effort should be used to suppress small amounts of distortion. Luckily, relatively low levels of distortion (<0.1% THD) are generally easily achieved and seem to be sufficient not to be perceived, but as said above, the THD measure itself may not be that useful of a way of discussing the "subjective measurements" of perception of distortion as Toole describes it.
Thanks for the tip, it looks likes an excellent source of information.
Last edited:
It's instructive when you design a difficult device, and the compromises are so dense that you can only bring them down in effect a certain amount without giving in to the others. Like trying to squeeze a balloon in your hands. And despite nothing being brought down too far, you can hear the benefit of having found and managed every issue.
This should be placed in context. The metric used is damping factor. If an amp has 0.1ohms Rout and drives an 8ohms speaker, the damping factor is defined as 8/0.1=80. It is generally accepted that anything above 20 is enough to make the effect on freq response insignificant. It really is a waste of effort to increase the DF even more.
BTW Adding more output devices does not, as a rule, decrease Zout. Zout is more a function of the feedback around the amp.
Jan
I know that dynamic/feedback impedance depends on the feedback, however, the load has to be supplied through emitter resistors to achieve thermal stability. The voltage swing necessary to supply a given amount of current to the load through a single transistor is higher because a single output emitter resistor drops a higher voltage with all of the current flowing through it. Quoting Cordell from page 70 of "Audio Power Amplifiers":
There is a drop in distortion due to the lower necessary voltage swing (and other effects like reduced beta droop due to current sharing) but another benefit may be that higher peak currents may be delivered through a lower static (determined by RE+VCE/IC) output impedance dominated by the emitter resistors. Because the amplifier does not perfectly compensate for the voltage drop of the output devices or emitter resistors, even for a small signal, there is an effect on the frequency response due to the finite output impedance.
The lower output impedance results in a flatter frequency response for speakers with resonances and may contribute to perceptual preferences if speaker resonances are significant.
>how do I know if a level of performance can be not be so difficult to achieve in practice as to be practically unattainable, and if the level of performance is going to be perceptible or provide value to the consumer?
FWIW, I've always thought of this as following an asymptotic curve, where the asymptote is "perfection" - and the x-axis is the time, effort, design cleverness and money it takes to get there. Someone in another thread remarked that today, with the benefit of modern, cheap (ebay, AliExpress...) electronics, you're so far up the curve that it's mostly the speakers that are going to convey - or not - the information pretty much all there within the upstream electrical signal.
This I tend to agree with and with my 64 yr old hearing, I've let go of getting every nat's *ss of each clock tick in perfect alignment. 1st, it's costly to do so, as predicted by the asymptotic curve. 2nd, I dont even know if I'm F-F-F-foolin' myself that I can hear anything different, because I expect to so much, due to the effort / expense put forth.
Unfortunately, in this day and age you may have to put up with things like "Noticed some queries to Baidu from my WiFi audio amplifier. Is this normal to see in PiHole?". So Cheap may be paid for via nefarious back door activity. Sounds good, AFAICT though.
FWIW, I've always thought of this as following an asymptotic curve, where the asymptote is "perfection" - and the x-axis is the time, effort, design cleverness and money it takes to get there. Someone in another thread remarked that today, with the benefit of modern, cheap (ebay, AliExpress...) electronics, you're so far up the curve that it's mostly the speakers that are going to convey - or not - the information pretty much all there within the upstream electrical signal.
This I tend to agree with and with my 64 yr old hearing, I've let go of getting every nat's *ss of each clock tick in perfect alignment. 1st, it's costly to do so, as predicted by the asymptotic curve. 2nd, I dont even know if I'm F-F-F-foolin' myself that I can hear anything different, because I expect to so much, due to the effort / expense put forth.
Unfortunately, in this day and age you may have to put up with things like "Noticed some queries to Baidu from my WiFi audio amplifier. Is this normal to see in PiHole?". So Cheap may be paid for via nefarious back door activity. Sounds good, AFAICT though.
Last edited:
My main interest in answering this question is because I have projects I have provided to build amplifiers on my github repositories. It would be good to identify the "knee" of the curve so that one can have a project that produces sound hopefully competitive with commercial offerings, yet be within the budget and experience of hobbyists. I am interested in these as both audiophile equipment and educational projects, as good projects provide motivation to build them and learn from them by having the product being actually useful and enjoyable. This is like the old Heathkit projects which were much loved for both their educational value and actual utility, as one could build radios, televisions, amateur radio transceivers, and other useful devices.
Yes but this is in the context of open loop, the output stage on itself. The complete amplifier output impedance, which is what determines the impact on speaker freq response linearity (i.e. the damping factor) is ultimately determined by the feedback factor. If you have a Zout with one pair of 0.1ohms (damping factor 80), and you halve it to 0.05 (damping factor 160) by doubling the output devices, the effect on the speaker freq response is pretty much unmeasurable.
Yes that was the point, but you get into diminishing returns (the subject of this thread) above about 20.
Jan
Yes, it would be good to identify the "knee" of the curve. I suspect that would be a place where, for a given population of those who like audio -
- where a distinct majority loses the ability to "hear the difference".
- AND
- where a distinct majority loses the motivation to pay for any slight difference they can hear.
I've also imagined the x-axis of such a curve to be log scale. $10 amp, $100 amp, $1000 amp, $10k amp, for example. The two above events will probably fall somewhere within such a range.
My thinking is with modern electronics, that may happen within the surprising decade. Of course it's all conjecture on my part and I'm sure exceptions abound. Is it a profitable strategy to market to such exceptions? They do, as we know; there's plenty of amplifier examples in the $1k - $10k range. Somebody's buying them, somebody thinks they can hear the next epsilon closer to musical perfection. Only costs 10X, but, WTH - go for it!
I rest my own case comfortably in the $100 - $1k decade. That's for DAC and power amp that goes plenty loud for me.
- where a distinct majority loses the ability to "hear the difference".
- AND
- where a distinct majority loses the motivation to pay for any slight difference they can hear.
I've also imagined the x-axis of such a curve to be log scale. $10 amp, $100 amp, $1000 amp, $10k amp, for example. The two above events will probably fall somewhere within such a range.
My thinking is with modern electronics, that may happen within the surprising decade. Of course it's all conjecture on my part and I'm sure exceptions abound. Is it a profitable strategy to market to such exceptions? They do, as we know; there's plenty of amplifier examples in the $1k - $10k range. Somebody's buying them, somebody thinks they can hear the next epsilon closer to musical perfection. Only costs 10X, but, WTH - go for it!
I rest my own case comfortably in the $100 - $1k decade. That's for DAC and power amp that goes plenty loud for me.
"As a wise old engineer once said: anyone can design a bridge to stand up, but only a good civil engineer can design a bridge to barely stand up."
that wise old engineer was moron
I would not want to walk on the bridge like that
that wise old engineer was moron
I would not want to walk on the bridge like that
🙄 You missed the point. You can overbuild a bridge with enormous budget but doing so won't produce many bridges. Good engineer would know what and how much is needed at where. Getting the most out of the least amount of means is what good engineering is.
Not only that, but overbuilding a bridge won't necessarily make it stronger or last longer, it will still fail when the weakest link gives in. The secret in value engineering is making all the links equally weak, which is to say each link is equally strong.
The same applies to consumer audio and other products. Ikea has this "not one cent more than necessary spent on each/any component part of a product" down to a fine art, which is why they can make things so inexpensively.
The same applies to consumer audio and other products. Ikea has this "not one cent more than necessary spent on each/any component part of a product" down to a fine art, which is why they can make things so inexpensively.
The elephant in the room is that many don't know all the places where attention is needed. Overbuilding has a tendency to overcome some of the unknowns.
The experienced and wise engineer knows just how much (and no more) to put in the bridge to satisfy the buyers/users.
The cable bridge where I used to work failed. So floppy that students would walk around campus rather than cross it. But it was very pretty so it stayed. Getting looser every year as the Old Men retired and nobody understood the bracing.
The Verona Island bridge in Maine failed. It felt fine all along (no Galloping Gertie), but had hidden rot which could not be reliably patched.