ASR has made an attempt for a complete measurement suite that covers all reasonable aspects of an amp. They highlight one parameter. No summed score.
But I'm not quite sure their suite is complete. And I think a summed score would be interesting and valuable as opposed to concentrate on one. I have high hope that DIYAUDIO can show how its done ;-)
No one would blindly buy on a one figure score. But it would surely help weeding out the nasties and get a well balanced product. I assume all parameters is presented that led up to a score.
//
But I'm not quite sure their suite is complete. And I think a summed score would be interesting and valuable as opposed to concentrate on one. I have high hope that DIYAUDIO can show how its done ;-)
No one would blindly buy on a one figure score. But it would surely help weeding out the nasties and get a well balanced product. I assume all parameters is presented that led up to a score.
//
So we can't even say crossover distortion sounds bad? We don't have to define positives for sound quality, we just have to define obvious negatives. A metric that weeds out obvious distortions will be more successful than one which takes all distortions as acceptable. Keep in mind, most amps are far better in performance than the audio encoders which sound quality metrics were designed to measure. Most amplifiers will get a perfect score from metrics based on actual hearing threshold data. So you are arguing on behalf of a minority of amps which have egregious crossover distortion.I have to say that any idea of involving any notion of SQ is begging for fail.
There is a very wide spread in the build quality of DIY amplifiers. Be aware that the designs are usually incomplete. And how are you going to compare a reliable build to one that looks like it's found in hells sewers?Well, do we want to limit the scope to just reliability then? That isn't too hard, but it does have the problem that we can't necessarily determine reliability without destructive testing or taking it apart. I had written a post about test loads but didn't bother to send it.
I must also add that any metric that has any reference to the inside of a DUT is not recommended. As I see it, to be stringent, it must be a black-box approach. "Crossover distortion" is thereby notihing that should be mentioned. Distortion is distortion - irrespectively of its cause. The challenge is to define the right measurements.... What one could however do, is to use knowledge and experience when one participates in the peer review of the draft specification. I assume it will be open to all members and that taking care of the comments will be a really heavy job ;-)So we can't even say crossover distortion sounds bad? We don't have to define positives for sound quality, we just have to define obvious negatives. A metric that weeds out obvious distortions will be more successful than one which takes all distortions as acceptable. Keep in mind, most amps are far better in performance than the audio encoders which sound quality metrics were designed to measure. Most amplifiers will get a perfect score from metrics based on actual hearing threshold data. So you are arguing on behalf of a minority of amps which have egregious crossover distortion.
//
It seems difficult to find a distortion measurement technique that specifically weeds out cross over distortion (COD) and it’s associated high order harmonics. It might be that a a straight dBV measurement at 1V RMS out into 8 ohms will catch gross COD by showing higher than expected THD.
Separately, the assessment program should be ‘black box’ other than one or two specification items
Separately, the assessment program should be ‘black box’ other than one or two specification items
What about the phase of the distortion partials? Isn't that a significant part of what makes distortion to a greater or lessor extent objectionable? IOW, distortion has a crest factor, if you want to look at it that way. The problem is that typical audio FFTs discard potentially useful phase information.
Maybe consider that white noise can have a lot of different time domain waveforms and still be white (i.e. it still has a flat frequency distribution). But look at the time domain waveforms and you know the different types of white noise will sound different.
Same with distortion, isn't it?
Put on your thinking caps if you want to outperform ASR measurements.
Note: image taken from: https://www.edge.org/response-detail/11715
Maybe consider that white noise can have a lot of different time domain waveforms and still be white (i.e. it still has a flat frequency distribution). But look at the time domain waveforms and you know the different types of white noise will sound different.
Same with distortion, isn't it?
Put on your thinking caps if you want to outperform ASR measurements.
Note: image taken from: https://www.edge.org/response-detail/11715
Last edited:
If average hearing threshold of THD can be accepted, then amplifier with THD bigger than this hearing threshold should be measure the harmonic profile. If H4, H5, H6 bigger than H2 or H3, it is bad design. Maybe because of crossover distortion. Most class A have monotonic harmonic profile.
It should THD have maximum number, because it is an amplifier, not effect box.
It should THD have maximum number, because it is an amplifier, not effect box.
Here is what Earl Geddes said:
To be useful the metric must be consistent and reliable – the same number must mean the same thing in every context and there must be a close correlation between the metric and the response that it is intended to scale.
This is precisely where the signal-based distortion metrics fail. In our next paper we will show that .01% THD of one type of nonlinear system can be perceived as unacceptable while 10% THD in another example is perceived as inaudible. Even one of these simple examples is sufficient to invalidate THD as a viable metric for discussion of the perception of distortion.
http://www.gedlee.com/Papers/Distortion_AES_I.pdf
Basically, we should forget THD. Sean Olive appears to be of the same opinion:
So, why do who should know better persist in dragging up THD?
To be useful the metric must be consistent and reliable – the same number must mean the same thing in every context and there must be a close correlation between the metric and the response that it is intended to scale.
This is precisely where the signal-based distortion metrics fail. In our next paper we will show that .01% THD of one type of nonlinear system can be perceived as unacceptable while 10% THD in another example is perceived as inaudible. Even one of these simple examples is sufficient to invalidate THD as a viable metric for discussion of the perception of distortion.
http://www.gedlee.com/Papers/Distortion_AES_I.pdf
Basically, we should forget THD. Sean Olive appears to be of the same opinion:
So, why do who should know better persist in dragging up THD?
A black box approach will end up with poorly derated amplifiers being highly ranked because they measure well.
Let me point out that discovering an amplifier is under rated by opening it up and looking at the output devices is LESS destructive than to discover it catches on fire during a reactive load test or later when playing music. We cannot open up or see inside all amplifiers but you can't say much about the reliability of an amplifier without doing so.
As for crossover distortion:
The harmonic profile of a sine wave with a disturbance is the same as the spectral profile as a single disturbance, but missing all the frequencies that are not multiples of the fundamental, because when you repeat a series of impulses the frequencies that are not multiples of the fundamental cancel out.
So the peaks of the harmonics of a disturbance on a sine wave follow the spectral response of that disturbance if it were a single isolated disturbance. The shape of the disturbance, usually a form of impulse or doublet or triplet, determines the shape of that spectrum, with the number of zero crossings and net positive signal determining the lower BW slope and the sharpness determining the upper BW. The peak in the spectrum is usually twice the wavelength of the disturbance. So crossover distortion is identifiable by broad peaks in the - trend - of the harmonic spectrum which increase in frequency as output current increases. The harmonic spectrum is usually tilted up by the loop gain slope.
The nulls in the response are related to the wavelength of the disturbance and may be more reliable markers when there is a wild loop gain curve. But then the achilles heel of spectral nulls is they are easily filled in by other spectral content.
The phase of the harmonics tells you where on the sine wave the disturbance falls, regardless of total disturbance energy, so would mainly be relevant if hearing is selective based on whether the disturbance is at the crest or the zero crossing. Note that crossover distortion will not occur at the zero crossing of the sound wave, if the speaker has a phase shifted load current.
Crossover distortion is a discrepancy of the wavelength of the residual which contains a disturbance with a smaller wavelength than the fundamental. You could say that an amp without crossover distortion is an amp where the wavelength of the residual is equal to the wavelength of the fundamental.
As output power increases, the spectral peaks of the disturbance increase in frequency because the wavelength of the crossover region becomes smaller. This would be identifiable as odd harmonics tumbling over each other in a power vs harmonic chart. The spectral peak of the disturbance is usually twice the wavelength of the disturbance itself, but loop gain slope shifts this to a higher frequency.
One would expect that crossover distortion becomes most audible when the spectral peak of the disturbance occurs in the 3KHz region where hearing is most sensitive.
With this in mind, every method I think about reduces to A-weighting the distortion residual plus masking of low orders, because ultimately the harmonic profile becomes most audible when the harmonics get concentrated at certain frequencies which are not correlated with the fundamental and sweep in tone in response to output level.
If you want something audible to contemplate you can listen to the sound of a square wave as the duty cycle is varied, shifting the harmonic profile peaks to tones unrelated to the fundamental.
Let me point out that discovering an amplifier is under rated by opening it up and looking at the output devices is LESS destructive than to discover it catches on fire during a reactive load test or later when playing music. We cannot open up or see inside all amplifiers but you can't say much about the reliability of an amplifier without doing so.
As for crossover distortion:
The harmonic profile of a sine wave with a disturbance is the same as the spectral profile as a single disturbance, but missing all the frequencies that are not multiples of the fundamental, because when you repeat a series of impulses the frequencies that are not multiples of the fundamental cancel out.
So the peaks of the harmonics of a disturbance on a sine wave follow the spectral response of that disturbance if it were a single isolated disturbance. The shape of the disturbance, usually a form of impulse or doublet or triplet, determines the shape of that spectrum, with the number of zero crossings and net positive signal determining the lower BW slope and the sharpness determining the upper BW. The peak in the spectrum is usually twice the wavelength of the disturbance. So crossover distortion is identifiable by broad peaks in the - trend - of the harmonic spectrum which increase in frequency as output current increases. The harmonic spectrum is usually tilted up by the loop gain slope.
The nulls in the response are related to the wavelength of the disturbance and may be more reliable markers when there is a wild loop gain curve. But then the achilles heel of spectral nulls is they are easily filled in by other spectral content.
The phase of the harmonics tells you where on the sine wave the disturbance falls, regardless of total disturbance energy, so would mainly be relevant if hearing is selective based on whether the disturbance is at the crest or the zero crossing. Note that crossover distortion will not occur at the zero crossing of the sound wave, if the speaker has a phase shifted load current.
Crossover distortion is a discrepancy of the wavelength of the residual which contains a disturbance with a smaller wavelength than the fundamental. You could say that an amp without crossover distortion is an amp where the wavelength of the residual is equal to the wavelength of the fundamental.
As output power increases, the spectral peaks of the disturbance increase in frequency because the wavelength of the crossover region becomes smaller. This would be identifiable as odd harmonics tumbling over each other in a power vs harmonic chart. The spectral peak of the disturbance is usually twice the wavelength of the disturbance itself, but loop gain slope shifts this to a higher frequency.
One would expect that crossover distortion becomes most audible when the spectral peak of the disturbance occurs in the 3KHz region where hearing is most sensitive.
With this in mind, every method I think about reduces to A-weighting the distortion residual plus masking of low orders, because ultimately the harmonic profile becomes most audible when the harmonics get concentrated at certain frequencies which are not correlated with the fundamental and sweep in tone in response to output level.
If you want something audible to contemplate you can listen to the sound of a square wave as the duty cycle is varied, shifting the harmonic profile peaks to tones unrelated to the fundamental.
Last edited:
Relative to the test signal, yes. How about relative to the harmonics themselves, e.g. their crest factor? It is the same for each and every amplifier, regardless of test tone frequency, and regardless of the commutation point if a reactive load? Would your position be that crest factor is never audible? What about differences in envelope modulation beyond the threshold of group delay? None of those types of factors distinguishes crossover distortion from other types of distortion? Reason I ask is that I would seriously like to know.The phase of the harmonics tells you where on the sine wave the disturbance falls...
I suppose another way of asking these sorts of question can be expressed in the time domain. Is it that all crossover distortion residuals look exactly the same in the time domain, or does waveshape distinguish them only in terms of harmonic magnitudes? And is that different from the distortion residual arising from a curved transfer function of an amplifier without crossover distortion?
Last edited:
I know Bonsai already gave the obvious method of identifying crossover distortion where THD decreases with frequency. I was looking for any other possible method.
It strikes me now that the main feature of crossover distortion which would defy all masking is the erratic movement of the spectral peaks with output level. Musical instruments with a natural decay for example would have descending HF tones as their level decreases. These are easily heard as distinct sounds. A tone sweep that doesn't follow any musical relationship with anything else is extremely easy to identify.
Monotonic distortion on the other hand concentrates a disturbance at only one point on the sine wave. Odd order distortions concentrate it at two points, IE the rising and falling edges of a square wave. This represents a cancellation of even harmonics, but if the duty cycle is varied the cancellations occur in other places giving the possibility of of many characters of tones.
It strikes me now that the main feature of crossover distortion which would defy all masking is the erratic movement of the spectral peaks with output level. Musical instruments with a natural decay for example would have descending HF tones as their level decreases. These are easily heard as distinct sounds. A tone sweep that doesn't follow any musical relationship with anything else is extremely easy to identify.
Monotonic distortion on the other hand concentrates a disturbance at only one point on the sine wave. Odd order distortions concentrate it at two points, IE the rising and falling edges of a square wave. This represents a cancellation of even harmonics, but if the duty cycle is varied the cancellations occur in other places giving the possibility of of many characters of tones.
Relative to the test signal, yes. How about relative to the harmonics themselves, e.g. their crest factor? It is the same for each and every amplifier, regardless of test tone frequency, and regardless of the commutation point if a reactive load? Would your position be that crest factor is never audible? What about differences in envelope modulation beyond the threshold of group delay? None of those types of factors distinguishes crossover distortion from other types of distortion? Reason I ask is that I would seriously like to know.
I suppose another way of asking these sorts of question can be expressed in the time domain. Is it that all crossover distortion residuals look exactly the same in the time domain, or does waveshape distinguish them only in terms of harmonic magnitudes? And is that different from the distortion residual arising from a curved transfer function of an amplifier without crossover distortion?
If you take an impulse which contains all frequencies equally below it's upper BW, and apply a hilbert transform to rotate all spectral components 90 degrees, It looks like it has become a doublet, except that it now has exponential tails whereas a doublet wouldn't. The exponential tails contain the rotated LF components which a doublet does not have. If you apply a hilbert transform to a square wave, it looks like an exaggerated triangle wave with much higher peak values. This is because all the sine waves are aligned at their peak, whereas with 0 phase they are aligned at their zero crossing so two peaks never align.
If you are looking for a correlation with crest factor then you are looking for how many harmonics have a phase which corresponds to a specific point in time, which in relation to the fundamental will be a different phase for each harmonic so will require some math. This sounds similar to determining the RMS level of an arbitrary waveform.
Answering these questions in the frequency domain is a bit convoluted, another approach would be to use DiAna to get the distortion residual, and simply calculate it's crest factor.
I think the practical takeaway from my last few posts is that individual harmonics are a bit of a red herring, a group of harmonics is audible as a spectral peak, and if this peak is frequency modulated then it becomes far, far more audible.
Great insights and posts keantoken. Now, please, describe a test for a black box to reveal the effects you describe e.i. stimuli and detection/measurement.
What would be very interesting to see is also an example of a very well measuring amp with traditional way of examination and a new proposed test which would show a different, more negative view, and perhaps even some SQ observations to go along. First part would be enough - just to make a test that pulls the pants off a very "well measuring" amp would be enough I think. This could forward the state of art in clinical characterisation and lead perhaps to better sounding amps.
//
What would be very interesting to see is also an example of a very well measuring amp with traditional way of examination and a new proposed test which would show a different, more negative view, and perhaps even some SQ observations to go along. First part would be enough - just to make a test that pulls the pants off a very "well measuring" amp would be enough I think. This could forward the state of art in clinical characterisation and lead perhaps to better sounding amps.
//
Last edited:
Great if there is any proof made by a valid DBT test. If it is missing, we have only more opinions, technical but subjective opinions, without a valid proof.
I think if we could define a test that would reveal distortion of considerable magnitude, that we have not measured before, I would vote to add it to the standard suite.
I dont think we can progress the SOTA if we require a DBT tests in order for a test to be made. We just need to go about it in a very clinical way and be sure, that if an amp pass all tests with highest marks, it will sound the finest.
A DBT test is still hampered by the involved gear to conduct it. If one hope to progress SOTA, one cant allow it to be hindered by using a test setup that wasn't developed with the new tests? ;-) It bites its own tail...an becomes a hinder - so here we see why one cant rely on DBT as a criteria for accepting a test to be part of a suite.
Lets trust tech and science rather than human 🙂
//
I dont think we can progress the SOTA if we require a DBT tests in order for a test to be made. We just need to go about it in a very clinical way and be sure, that if an amp pass all tests with highest marks, it will sound the finest.
A DBT test is still hampered by the involved gear to conduct it. If one hope to progress SOTA, one cant allow it to be hindered by using a test setup that wasn't developed with the new tests? ;-) It bites its own tail...an becomes a hinder - so here we see why one cant rely on DBT as a criteria for accepting a test to be part of a suite.
Lets trust tech and science rather than human 🙂
//
The simplest test would just be a sine wave that starts out quiet and gets louder. And you would look for tone sweeps in the harmonic profile as it does this. It's not magical in any way though. An amplifier with low THD would not fail this test.
I
It’s not about ‘outperforming’ the other place. It’s about coming up with an assessment program that more holistically measures amplifier performance that can be used to guide designers and also as a finished product assessment.
Quite frankly, I don’t see the value of a low powered Topping amplifier with 0.00005% distortion but nothing else is really known about it, yet is (was?) ranked #1 and the Benchmark ABH1 has been ‘downgraded’ to 4th or 5th in the list because of a marginal non-audible increase in THD upon retest. Lots examples on this site where distortion is chased as the holy grail to the point that questionable stability and overdrive recovery are subordinate to getting 1 ppm rather than 10 ppm. Both levels of distortion ant this level are completely inaudible no matter what the harmonic content.
It’s also not about ‘my amp is better than yours’. Any amp with a balanced assessment result ie no glaring shortcomings, that achieves 7or 8 out of 10 in the final score will be a very good amplifier.
One final point. The tests being proposed should ideally be able to be done with a simple set of tools:
Sound card
Scope
Dummy loads
Dummy reactive loads
Capacitors
This puts it within reach of most builders and designers.
🙂
Mark,What about the phase of the distortion partials? Isn't that a significant part of what makes distortion to a greater or lessor extent objectionable? IOW, distortion has a crest factor, if you want to look at it that way. The problem is that typical audio FFTs discard potentially useful phase information.
Maybe consider that white noise can have a lot of different time domain waveforms and still be white (i.e. it still has a flat frequency distribution). But look at the time domain waveforms and you know the different types of white noise will sound different. View attachment 1257505
Same with distortion, isn't it?
Put on your thinking caps if you want to outperform ASR measurements.
Note: image taken from: https://www.edge.org/response-detail/11715
It’s not about ‘outperforming’ the other place. It’s about coming up with an assessment program that more holistically measures amplifier performance that can be used to guide designers and also as a finished product assessment.
Quite frankly, I don’t see the value of a low powered Topping amplifier with 0.00005% distortion but nothing else is really known about it, yet is (was?) ranked #1 and the Benchmark ABH1 has been ‘downgraded’ to 4th or 5th in the list because of a marginal non-audible increase in THD upon retest. Lots examples on this site where distortion is chased as the holy grail to the point that questionable stability and overdrive recovery are subordinate to getting 1 ppm rather than 10 ppm. Both levels of distortion ant this level are completely inaudible no matter what the harmonic content.
It’s also not about ‘my amp is better than yours’. Any amp with a balanced assessment result ie no glaring shortcomings, that achieves 7or 8 out of 10 in the final score will be a very good amplifier.
One final point. The tests being proposed should ideally be able to be done with a simple set of tools:
Sound card
Scope
Dummy loads
Dummy reactive loads
Capacitors
This puts it within reach of most builders and designers.
🙂
Last edited:
You have to be able to standardise a test like that and get a metric from it that could be used. That is the difficult bit. Just writing down numbers or pasting up 10 plots isn’t going to be ideal if what we are aiming for is a simple 1 or 2 page document.The simplest test would just be a sine wave that starts out quiet and gets louder. And you would look for tone sweeps in the harmonic profile as it does this. It's not magical in any way though. An amplifier with low THD would not fail this test.
Might be the way to do this is to write some test script for the QA401/2/3 but then we would have to work out how to present the COD data as a single metric.
It is simply a measure of amplifier electrical linearity. That some folks take it as representative of perceived sound quality is where the issue lies - see rankings of amplifier quality based solely on THD as an example.Here is what Earl Geddes said:
To be useful the metric must be consistent and reliable – the same number must mean the same thing in every context and there must be a close correlation between the metric and the response that it is intended to scale.
This is precisely where the signal-based distortion metrics fail. In our next paper we will show that .01% THD of one type of nonlinear system can be perceived as unacceptable while 10% THD in another example is perceived as inaudible. Even one of these simple examples is sufficient to invalidate THD as a viable metric for discussion of the perception of distortion.
http://www.gedlee.com/Papers/Distortion_AES_I.pdf
Basically, we should forget THD. Sean Olive appears to be of the same opinion:
View attachment 1257520
So, why do who should know better persist in dragging up THD?
The challenge is how do a measurement that capture’s the harmonic profile and correctly weights it in favour of amplifiers that have a benign or euphonic sonic signature? There aren’t any standardised tests for this and my fear is any attempt to do so is fraught with subjectivity and opinions.
Maybe we just measure THD and then a weighted score of harmonic profile eg 2 and 3 below .005% and no other harmonics is 5/5, presence of other harmonics 4/5 etc etc
But, we are now getting into quite complex stuff and it’s probably better to get AFOM of the ground quickly and then at a later date update the metrics with a more fully thought out measurement.
- Home
- Amplifiers
- Solid State
- AFOM: An attempt at an objective assessment of overall amplifier quality