Thanks SY, that's excellent. So there are tests which aim to view the characteristics of the amp in the time domain?
Another thought crosses my mind:
If I were to take a series resistor and follow it with a couple of back-to-back parallel diodes to ground, I would create a simple clipping characteristic. Now, if I tested this for THD and IMD using test tones at several levels I might be able to make some statements about which harmonics are prevalent, and how audible or offensive it was likely to be, but it would be a very indirect proxy for simply plotting the device's basic transfer function, which might tell me far more about what was going on.
Presumably real world transfer functions for amps or speaker drivers are more subtle and complex than this, involving further dimensions of time, loading and so on, but might there be some advantage in attempting to visualise the device in terms of N-dimensional 'transfer functions', or breaking it down into virtual 'functional blocks' which represent the device's characteristics? A crossover distortion characteristic might be an obvious candidate for visualising this way, or a speaker driver nearing its Xmax limits.
If an amplifier was found to be the equivalent of a basic linear-ish transfer function, but in parallel had a peculiar block involving a gated oscillator, we might immediately understand something about it that wasn't obvious from more conventional testing.
Plot of IR measurement in time domain is view of test system's characteristics. As is spectrogram of IR, and other interpretive methods of IR display.
Exponential swept sine measurements break result into linear IR component and harmonic components. IR of harmonics may be summed, normalized and convolved with music as an interpretation.
With pure tone testing, fundamental may be filtered out, the residual amplified and listened to. Harmonic distortion from soft clipping is vastly different sounding than crossover distortion of poorly adjusted class AB amplifier. Amplitude and phase of harmonics comes into play too.
Farina's paper:
NON-LINEAR CONVOLUTION: A NEW APPROACH FOR THE AURALIZATION OF DISTORTING SYSTEMS
is a good reference, as well most of Farina's work.
Klippel's papers are also great reference. He also has nifty Listening Test, the underlying methods are built on measurements, and lead to quantification of perceptual abilities.
Exponential swept sine measurements break result into linear IR component and harmonic components. IR of harmonics may be summed, normalized and convolved with music as an interpretation.
With pure tone testing, fundamental may be filtered out, the residual amplified and listened to. Harmonic distortion from soft clipping is vastly different sounding than crossover distortion of poorly adjusted class AB amplifier. Amplitude and phase of harmonics comes into play too.
Farina's paper:
NON-LINEAR CONVOLUTION: A NEW APPROACH FOR THE AURALIZATION OF DISTORTING SYSTEMS
is a good reference, as well most of Farina's work.
Klippel's papers are also great reference. He also has nifty Listening Test, the underlying methods are built on measurements, and lead to quantification of perceptual abilities.
Ho ho.Yes, we call this an "oscilloscope."
The frequency domain allows you to discern details that are -100dB down from the main signal using analogue equipment if that's all you have, and conveniently allows you to ignore the DUT's bandwidth and phase shift. But the criticism is that it is a "cartoon" measurement that doesn't necessarily tell you what the DUT's underlying characteristics are (e.g. the diode clipper) and hardly tests the entire 'signal space'. An oscilloscope won't let you discern details that are 100dB from the main signal, but a computerised gizmo working in the time domain could let you see wrinkes, nonlinearities and irregularities that a frequency domain test might miss, or might not be able to present in a meaningful way.
How else to bridge the divide between measurements and ears? I'm not prepared to believe that there is anything that ears can hear that can't be measured, but I see a huge vacant space where measurements simply don't go, at the moment. I suspect that partly this is because the frequency domain testing from 70 years ago has simply been carried through to the digital age without any attempt to do anything more interesting.
But the impulse response as recorded with a swept sine wave is really just an aggregation of the DUT's response to a fixed amplitude signal. It doesn't tell you if a peculiar combination of signal and load 3ms after a particularly hefty transient gives rise to a slight wobble in the output. That's what I want to see. Of course the sort of person who believes in measurements is also the sort of person who believes he knows exactly how an amplifier works - for him there's hardly any point testing it because he already knows what the answer will be. But this leaves a vacant space for the New Age subjectivists to move into.
Last edited:
Putting aside buzzwords, oscilloscopes are incredibly useful for spotting squegging, which is what you're talking about- if it's very low amplitude, you look at the residual from a distortion analyzer, which nulls out the fundamental.
I have no idea what the diode thing brings to the table- that's not representative of any actual load. Reactive loads (e.g., speaker simulators made from resistors, capacitors, and inductors to simulate lumpy impedance curves) are very, very useful and DO represent real world loads. Clipping and overload is tested by using a test signal that alternates between a few dB under clipping to a few dB over clipping- there's articles going back to the 1950s which show how to do it and how to interpret it. I use a piece of software called DACtester to generate the input signal, then use a scope to look at the output, using different loads. This is the basic way I was able to determine that my tube power amp didn't have the blocking issues common to most RC-coupled tube amps.
If you have bigger bucks than I, the Power Cube is even better for power amp testing.
Of course it does.
I have no idea what the diode thing brings to the table- that's not representative of any actual load. Reactive loads (e.g., speaker simulators made from resistors, capacitors, and inductors to simulate lumpy impedance curves) are very, very useful and DO represent real world loads. Clipping and overload is tested by using a test signal that alternates between a few dB under clipping to a few dB over clipping- there's articles going back to the 1950s which show how to do it and how to interpret it. I use a piece of software called DACtester to generate the input signal, then use a scope to look at the output, using different loads. This is the basic way I was able to determine that my tube power amp didn't have the blocking issues common to most RC-coupled tube amps.
If you have bigger bucks than I, the Power Cube is even better for power amp testing.
Of course it does.
But squegging (nice buzzword) assumes some knowledge of how the amplifier works i.e. that a common characteristic of amplifiers is squegging and therefore we can devise a special test for it. But maybe in this case a steady sine wave isn't kicking off the instability.
The principle of nulling out the test signal, however, is a good one. But why should not the test signal be a much more complex, arbitrary waveform - even an passage of music that a Golden Ears claims to be able to hear a problem with. A THD test isn't going to satisfy him. You could still do repetitive averaging, and aim to detect tiny deviations from the 'straight wire with gain'. Basic gain, phase and bandwidth roll off in the DUT (which affects time domain measurements but not the frequency domain) could be nulled out, too, with frequent measurements throughout the test for just that purpose - if it varied during the test then that would be interesting in itself.
I was using the diode thing not as an example of a load, but as an example of an exaggerated characteristic that is very easy to understand in the time domain, but fairly opaque in the frequency domain. If an amplifier had a mild characteristic of the 'diode clipper', then frequency domain tests would hint at it, but you'd have to do a lot of tests and use considerable intuition to piece together what was a very simple characteristic in the time domain.
Again, this is assuming some prior knowledge of the device under test i.e. that it has a simple clipping threshold. It may also have several other thresholds, knees etc. and some that are time-dependent. A big transient may momentarily heat up the power transistors, kick the power supply down for a moment and approach the speaker cone's limits, changing the load. If it provokes some brief instability, then these are surely the things that maybe the Golden Ears can hear.
No, we live in a world where audiophiles routinely claim to be able to hear distortion and noise that is -100dB down. With a 'scope you might see something -40dB down with averaging turned on and a simple test signal, but it would be the hardest labour to use a 'scope for what I have in mind. A computerised gizmo could be left overnight to probe the DUT with every signal and load imaginable, and detect deviations -100dB down in the time domain.
(It might turn out that amplifiers are very boring, and never do anything interesting, but at least you'd have proved it.)
If you're arguing that someone doing the testing should know how to use his test gear and have some basic understanding of the electronics he's testing, I would agree with that. Past that, I'm unclear as to the point you're trying to make.
Actually, your diode clipper will have huge issues in both time and frequency domain, so I'm still unclear on your point there as well.
Sure, there's lots of wacky claims out there. When someone demonstrates that they actually can, then those claims can be taken more seriously. As it is, they have as much evidence as the folks who claim that they've been kidnapped and anally probed by space aliens. Doing a test with Bill Waslo's infamous Sousa band signals can be highly amusing.
Anyone with a scope, distortion analyzer, and spectrum analyzer (and who knows how to use the equipment) will easily see any of the pathologies you've brought up at -100dB, not that it is likely to matter.
Actually, your diode clipper will have huge issues in both time and frequency domain, so I'm still unclear on your point there as well.
Sure, there's lots of wacky claims out there. When someone demonstrates that they actually can, then those claims can be taken more seriously. As it is, they have as much evidence as the folks who claim that they've been kidnapped and anally probed by space aliens. Doing a test with Bill Waslo's infamous Sousa band signals can be highly amusing.
Anyone with a scope, distortion analyzer, and spectrum analyzer (and who knows how to use the equipment) will easily see any of the pathologies you've brought up at -100dB, not that it is likely to matter.
Simply that frequency domain testing doesn't tell you directly what that circuit does, whereas a time domain test and visualisation reveals it directly. An amplifier may have a hint of 'diode clipper' (at -80dB) about it. Is it remotely possible that a time domain test and representation of a device might be more informative than the frequency domain?
What's the big deal about the frequency domain anyway? That it is very sensitive to certain characteristics using primitive equipment, and it can ignore inconvenient factors that time domain testing would fall foul of (phase shift, gain, bandwidth rolloff). The problem is that it still leaves a gap for a would-be Golden Ears to claim to hear a difference on real music.
Well it seems to me that you can therefore write off most of the contributors to this forum! They all believe they can hear differences between different types of amplifier, mains cable, capacitor, resistor, switch; and that's before they've burned them in or whatever they do. Some of the most serious people around here will talk quite straight-faced about hearing the differences between DACs that measurements show to be -100dB down. At least one of the Big Beasts around here sells his own brand of cables with a special Celtic weave to add musicality to the sound - which, needless to say, is too elusive to be measured. I didn't realise you were so dismissive of them!
I'm dismissive of implausible claims offered without evidence. Some people react emotionally to that!
I think you're attributing an argument to me that I haven't made- I use both oscilloscopes and distortion analyzer residual outputs (time domain) and spectrum analyzers (frequency domain) when I characterize amplifiers. Each has its use.
Very interesting contemplations, CopperTop, this is the direction I also believe that investigation into audio needs to go, to get ... The Answers !!. Conventional testing is far too crude and shallow, I'm certainly sure that the SQ issues I routinely pick up and work on would be very hard to detect with standard measuring.
Amplifiers are anything but boring, that's where the major issues are, for me. Speakers on the other hand are very ho-hum, anything vaguely reasonable can be knocked into shape with the obvious tweaks, the bottlenecks are in the earlier electronics ...
They would probably be very hard to detect in a double blind test as well. Just sayin'!
'Tis a great shame that all the relevant information that is obviously out there, available in easily digestible form, to allow all the quality issues to be resolved is largely ignored. Otherwise, audio systems would sound far better than they typically do ...
Possibly the main reason that this unfortunate situation exists are the rampant egos, and unmitigated greed of the majority of participants in this unfortunate arena ...
Possibly the main reason that this unfortunate situation exists are the rampant egos, and unmitigated greed of the majority of participants in this unfortunate arena ...
Possibly ... this is where things get hairy, because the setup of the test environment can profoundly affect the very qualities you're trying to chase down -- the biggest struggle for me has been to isolate the playback system from external interference, and mutual interference of the components. I've seen comments on the difficulty of creating a true black box scenario for critical evaluation many times, I can only agree.
Optimum sound for me has always been very fragile, the slightest aberration can bring it crashing down, in the subjective sense. I'm certain that it's not intrinsically so; it's just that I haven't sufficient grasp, understanding, control of all the relevant parameters, and these days the raw energy to persevere in tracking down the "culprits" ...
CopperTop:
Your usage of "frequency domain test" and "time domain test" infers a misunderstanding. For any discrete time period, time domain and frequency domain represent same information. They transform back and forth symmetrically via complex Discrete Fourier Transform.
All depends on what one is listening for: the music teacher hears unfortunate intonation, the instrument technician hears a poor choice of wood to refurbish the instrument, the recording engineer hears that an alternative set of mikes would have been better, the acoustician hears a bad choice of hall was used, the speaker nut identifies the colourations that particular drivers add to the equation ... with luck, some might actually hear music ...
I unfortunately hear deficiencies introduced further back in the chain ...
I unfortunately hear deficiencies introduced further back in the chain ...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.