Michael, we have tried a number of different tests, including: Two, 3, and multitone IM, noise loading, and other tests. These are sensitive tests for static distortion, better than harmonic distortion, BUT difficult to get extremely high resolution.
There is one important component that we have not completely addressed. This is FM distortion, due to the working amplifier bandwidth being modulated by dynamic changes in open loop gain, due to open loop distortion. The only reasonable cures for this, at this time, appear to be high open loop bandwidth and moderate feedback, or little or no global feedback. Duh!
This has been called IIM distortion by Matti Otala about 20 years ago, and fairly recently was considered as an important distortion factor by Barrie Gilbert of Analog Devices.
There is one important component that we have not completely addressed. This is FM distortion, due to the working amplifier bandwidth being modulated by dynamic changes in open loop gain, due to open loop distortion. The only reasonable cures for this, at this time, appear to be high open loop bandwidth and moderate feedback, or little or no global feedback. Duh!
This has been called IIM distortion by Matti Otala about 20 years ago, and fairly recently was considered as an important distortion factor by Barrie Gilbert of Analog Devices.
john curl said:
That's almost funny since Gilbert almost ridiculed Otala and the
audio community about discovering TIM, which Gilbert claimed
was already well-known to EEs except in audio.
On the main question, although it may be difficult to know what
would be meaningful ways of measuring distorsion, it seems that
with todays high-precision DACs and processing power, it should
be possible to do various types of measurements hardly possible
10 or 20 years ago. Just one idea that just struck me (having no
idea how relevant it would be). Suppose we repeatedly feed an
amp with one cycle of high amplitude and short cycle time,
immediately followed by lower-frequency sines of lower amplitude.
It should be no problem to postprocess the result so these
high-amplitude cycles are removed and the rest is concatenated
together as if it were a just a continuous sine wave. Doing an
FFT on this signal might, perhaps, reveal something on how
large transients affect what comes immediately after? This is
just a rough basic idea, and should probably be refined in
one way or another.
Something I would like to see is a test using real music. There is a marked tendency among people who have been at this long enough to disregard the traditional distortion measurement techniques. They're missing something. What? Don't know yet.
Okay, so let's drop steady tones or even step functions. Use music as the test signal. Feed a real-world complex signal into the circuit under test. Sample and record this in the same manner that digital oscilloscopes capture transient events for prolonged study. At this point, I see two possible avenues: Overlay the two signals and look for differences (blunted peaks might indicate dynamic compression, for instance--perhaps a slow power supply regulator?). Or use the original signal to null the output signal (level-adjusted, obviously) and see what kind of residual you've got.
The main problem I see is getting the sampling to correspond between the test signal and the output. There's going to be some delay going through the circuit and the input signal is going to have to be offset by some arbitrary number of milli- or microseconds in order to match the output signal. This will need to be an adjustable parameter, since different circuits will have different delays. Given that I'm assuming that this will take place in the digital domain, the two signals can be slid back and forth until you achieve closest match. Lock that and go from there.
Grey
Okay, so let's drop steady tones or even step functions. Use music as the test signal. Feed a real-world complex signal into the circuit under test. Sample and record this in the same manner that digital oscilloscopes capture transient events for prolonged study. At this point, I see two possible avenues: Overlay the two signals and look for differences (blunted peaks might indicate dynamic compression, for instance--perhaps a slow power supply regulator?). Or use the original signal to null the output signal (level-adjusted, obviously) and see what kind of residual you've got.
The main problem I see is getting the sampling to correspond between the test signal and the output. There's going to be some delay going through the circuit and the input signal is going to have to be offset by some arbitrary number of milli- or microseconds in order to match the output signal. This will need to be an adjustable parameter, since different circuits will have different delays. Given that I'm assuming that this will take place in the digital domain, the two signals can be slid back and forth until you achieve closest match. Lock that and go from there.
Grey
Please point me to an avenue of more education if my suggestion is off base, but how would a pulse, followed by a music signal work for aligning the signals? If the pulse was a known entity, then aligning the pulse would align the rest of the sample, correct. If this is not the right approach or (more than likely) is obvious, but not the correct method, please help me to understand the fallacy. I truly am trying to learn more so I can understand what is being discussed.
Well, obviously, it is an African Swallow instead of the European variety, but is it Laden or Unladen? From the context, I must assume heavily laden. . . but regretfully, my favorite color is blue, unless black anodized aluminum is a choice. . .
Sandy.
Nelson Pass said:
You are chasing the Holy Grail. This is a worthwhile quest, but
try to remember your favorite color and the ground speed of an
African Swallow.
Well, obviously, it is an African Swallow instead of the European variety, but is it Laden or Unladen? From the context, I must assume heavily laden. . . but regretfully, my favorite color is blue, unless black anodized aluminum is a choice. . .
Sandy.
Nelson Pass said:
You are chasing the Holy Grail. This is a worthwhile quest, but
try to remember your favorite color and the ground speed of an
African Swallow.
Are you sure you mean an African swallow?
I didn't mean to propose this particular measurement to be the
one to use, but rather hint at the possibilities of using modern
digital signal processing to measure things we couldn't measure
with traditional analog equipment.
Since most people seem to agree that distorsion figures in the
usual sense do not correlate well to how an amplifier sounds
(even PMA suggests there is only a one-way implication) the
question frequently arises: Are we measuring the wrong things?
What should we measure? These questions were recently
asked again in this thread. I personally think that if two amplifiers
sound different, then it must be possible to measure the difference,
even if we don't know today what to measure. Is it worth it to
find out how to measure it? For the individual DIYer, no, probably
not. For the audio industry as a whole, yes, I definitely think so,
provided these differences actually do exist and are not only
psychological.
GRollins said:
I certainly think this is worthwile. Baxandall, Walker, Hafler (I think) and others have used this approach. You need some passive amplitude and phase correction to get the best nulling, but this should not be a cause of additional distortion.
The nice thing is that you can use music as the source with all its dynamics, complex multitones, etc. Exactly the signal that is going to be used in normal life with an amplifier.
And what is important, it allows you to listen to the residual. The residual will tell you about the character of the distortion. Some amplifiers may create distortion that might be not even be annoying to listen to. I can imagine that low order harmonic distortion will create a residual in which the original music is very recognizable. While lower amounts of uncorrelated noise-like distortion could be very annoying. And TIM-like distortion will be very objectionable, since it has no correlation with the original music whatsoever.
Normally the distortion is masked by the original (amplified) signal. Removing that signal will act as a very nice magnifying glass.
Steven
Everybody's done it, and it tends to lead to the belief that
there's no difference in the sound of amps that measure
the same.
My setup in the 70's got down to -80 dB, and sure enough,
the residual sounded crappy, but I was never able to correlate
that to the subjective result of listening to the amp through
speakers.
there's no difference in the sound of amps that measure
the same.
My setup in the 70's got down to -80 dB, and sure enough,
the residual sounded crappy, but I was never able to correlate
that to the subjective result of listening to the amp through
speakers.
But, Nelson, that was to be expected, isn't it? Subjective testing, in the end, does mean that different subjects (testers) have different opinions (results).
Objective testing, or measuring, tries to reach a result that is repeatable and valid for different test locations by different testers.
There is no way that the two will ever correlate, except the occasional case where by pure coincidence some testers' "objective" test result correlates with someones "subjective" test, but by definition this will not be the case for all the other subjective testers.
So given that the two types of tests are fundamentally differently oriented, there is no way 'the twain shall ever meet'.
Jan Didden
Objective testing, or measuring, tries to reach a result that is repeatable and valid for different test locations by different testers.
There is no way that the two will ever correlate, except the occasional case where by pure coincidence some testers' "objective" test result correlates with someones "subjective" test, but by definition this will not be the case for all the other subjective testers.
So given that the two types of tests are fundamentally differently oriented, there is no way 'the twain shall ever meet'.
Jan Didden
Yeah, I seem to recall that Hafler used it as a showroom stunt.
It's not that cancellation of the input & ouput hasn't been done--it's what's been done with the result.
--In my mind's eye, I see two traces on my PC screen. On top is the original source signal. On bottom is the residual from the output. It's one thing to listen to nasties as they go by and say,"Man, that sounds bad," but it's another to be able to freeze-frame and look at the input vs. residual and say,"Hmmm...you know, every time the amplitude increases past such-and-such a voltage, we get a spike in the residual." Scroll horizontally back and forth and compare what happened at T=5:32 with what happened at T=5:37.
--Another possibility is to coerce the ghost of Fourier to massage the data. People nearly always focus on higher harmonics. That's all well and fine. But what about sub harmonics? Beat frequencies and such. Comparatively rare to find someone who is interested. Suppose there were transient power supply problems that manifested themselves in a sort of transient motorboating. We're talking single-digit beats that would in turn modulate the signal. Might not show up in a conventional test--and obviously not in steady-state testing.
I had another thing in mind, but that one got away from me.
That still leaves the overlay avenue, which as far as I know, no one has ever attempted. Color code the parts of the signal that are the same blue. Code parts that deviate by 1%, 5%, etc. with other colors in order to make them stand out. Then we see that, for instance, the signal is fine below 500Hz, even at high amplitude, but lookie-here...5kHz goes nuts over 3.5V, and 10kHz at 1V. Some of this can be seen or deduced with currently available equipment, but not all of it.
We have technology now that we didn't have before. It may not turn up anything new, but then again, it might.
Grey
It's not that cancellation of the input & ouput hasn't been done--it's what's been done with the result.
--In my mind's eye, I see two traces on my PC screen. On top is the original source signal. On bottom is the residual from the output. It's one thing to listen to nasties as they go by and say,"Man, that sounds bad," but it's another to be able to freeze-frame and look at the input vs. residual and say,"Hmmm...you know, every time the amplitude increases past such-and-such a voltage, we get a spike in the residual." Scroll horizontally back and forth and compare what happened at T=5:32 with what happened at T=5:37.
--Another possibility is to coerce the ghost of Fourier to massage the data. People nearly always focus on higher harmonics. That's all well and fine. But what about sub harmonics? Beat frequencies and such. Comparatively rare to find someone who is interested. Suppose there were transient power supply problems that manifested themselves in a sort of transient motorboating. We're talking single-digit beats that would in turn modulate the signal. Might not show up in a conventional test--and obviously not in steady-state testing.
I had another thing in mind, but that one got away from me.
That still leaves the overlay avenue, which as far as I know, no one has ever attempted. Color code the parts of the signal that are the same blue. Code parts that deviate by 1%, 5%, etc. with other colors in order to make them stand out. Then we see that, for instance, the signal is fine below 500Hz, even at high amplitude, but lookie-here...5kHz goes nuts over 3.5V, and 10kHz at 1V. Some of this can be seen or deduced with currently available equipment, but not all of it.
We have technology now that we didn't have before. It may not turn up anything new, but then again, it might.
Grey
I agree with PMA, the group delay is spoiled with a "slow" amplifier and the subtraction test wont work fully for higher audio frequencies.
More else the input low pass capacitor must also be removed which should be a clear point, an ordinary RC filter with ft 100 kHz gives a delay (if I remember it right) at 20 kHz around 5-10 uS (correct me if I'm wrong, to tired to calculate in the middle of the night.
)
More else the input low pass capacitor must also be removed which should be a clear point, an ordinary RC filter with ft 100 kHz gives a delay (if I remember it right) at 20 kHz around 5-10 uS (correct me if I'm wrong, to tired to calculate in the middle of the night.
)Ultima Thule said:I agree with PMA, the group delay is spoiled with a "slow" amplifier and the subtraction test wont work fully for higher audio frequencies.
More else the input low pass capacitor must also be removed which should be a clear point, an ordinary RC filter with ft 100 kHz gives a delay (if I remember it right) at 20 kHz around 5-10 uS (correct me if I'm wrong, to tired to calculate in the middle of the night.
I'm tired too, but wouldn't an LP filter make the comparison more
fair. If som amps start to fall off to much already in the audio band
it won't help, but then, isn't that a problem with them that should
show up in the test?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.