Bob
Thanks for the note. All your tricks seem quite reasonable and effective and for an amp designer I would expect such things. As I said, my interest was a simple test using only a sound card and software - both of which I already had on hand. Your tests require quite a bit more.
As I talk about this subject I hear people say more and more "Yea, its the low level stuff that matters", but then why is it that we never see this data?
Perhaps its like the reaction that I got when I showed that THD and IMD were completely uncorrelated with perception -
"Yea, we know, but they are easy to do."
One major manufacturer of test equipment commented that "No one is asking for new distortion tests." DAA!!
"Its dark in here!"
"Well turn on the light!"
"No thanks, I like it dark."
Thanks for the note. All your tricks seem quite reasonable and effective and for an amp designer I would expect such things. As I said, my interest was a simple test using only a sound card and software - both of which I already had on hand. Your tests require quite a bit more.
As I talk about this subject I hear people say more and more "Yea, its the low level stuff that matters", but then why is it that we never see this data?
Perhaps its like the reaction that I got when I showed that THD and IMD were completely uncorrelated with perception -
"Yea, we know, but they are easy to do."
One major manufacturer of test equipment commented that "No one is asking for new distortion tests." DAA!!
"Its dark in here!"
"Well turn on the light!"
"No thanks, I like it dark."
Bob Cordell said:
Hello M Cordell
I did seen your text about your DM in your web page sometime ago, I've look again today, it's a very interting concept, will you say more about it soon in your web page ?
Thank
Gaetan
I like it dark. Every time I stick my head out from under the blanket, I see scary stuff.
Making these measurements is wonderful and illuminating from a design standpoint, but I remain unconvinced on audibility. Is there any signal you can think of, correlated, uncorrelated, a recording of a brass band in a parade, whatever, that, mixed in with a music signal at -120dB, is audible under any condition? My living room on the quietest night still has a noise floor (is that a carpet with too busy a pattern?) that would drown it out. If the level is set reasonably, I'm not sure my ears have that much dynamic range- remember, I can't come close to hearing my electronic noise floor right now. We're talking about stuff below the LSB of a CD here, so does that mean that, with a CD source, again set to a reasonable level, all competently designed amplifiers, operated within their limits, sound the same?
Making these measurements is wonderful and illuminating from a design standpoint, but I remain unconvinced on audibility. Is there any signal you can think of, correlated, uncorrelated, a recording of a brass band in a parade, whatever, that, mixed in with a music signal at -120dB, is audible under any condition? My living room on the quietest night still has a noise floor (is that a carpet with too busy a pattern?) that would drown it out. If the level is set reasonably, I'm not sure my ears have that much dynamic range- remember, I can't come close to hearing my electronic noise floor right now. We're talking about stuff below the LSB of a CD here, so does that mean that, with a CD source, again set to a reasonable level, all competently designed amplifiers, operated within their limits, sound the same?
Conrad Hoffman said:
I would agree with the last part of your last statement, but not the first.
I am talking about stuff thats maybe 40-60 dB down below a signal level at about 70 dB SPL. These are reasonable numbers, not the extremes that you are mentioning. At 70 dB SPL we could detect a coherent signal modification well below the acoustic noise floor of your room. Thats because your noise floor is incoherent and the signal distortion is coherent. Recal what I said before, the ear is NOT an FFT analyzer and as such many of the conecpts that are so familiar with are not true of hearing. Our ears are very acute at hearing very small deviations in a signal at very low levels.
Now transfer these numbers to the amp and you will see that we need to be measuring well down to the noise floor of the amp and of "typical" measuring equipment. As Bob points out its not hard to do with specialized equipment, but the point is that no one does it, or at least they don't publish it.
You can't hear your amps noise because it is masked by the rooms acoustic noise - but only because both are incoherent and the ear has nothing to "lock" onto to differentiate these sounds in detail.
No question that something coherent (the reason I mentioned the brass band), and probably something higher in frequency, wil be easiest to hear. Also, we have to be very careful about our reference levels when talking dB. If my noise floor is 100dB down from full output of my Gigawatt amp, that's not too quiet, but from 1Vrms, it's not so bad. This is a system problem, and if the artifacts are in the -40@70 area you mention, then they might well be audible. That seems like a pessimistic scenireo, but maybe not. Now I need to think about my levels, though I still think that if everything is properly matched, the noise floor should be so far down that even if it's replaced with a brass band at the same level, I shouldn't hear it. OTOH, given the gain of a typical power amp, it might explain why removing the final gain stage from the preamp and running right off the vol pot is often so beneficial- all that low level junk from the gain stage gets a constant 10X or so boost with no real benefit to the audio signal.
Distortion measurements are very interesting, I have noticed different speaker cables and passive components can have different effects on results.
gaetan8888 said:
Yes, I do hope to get more information about the DM up on my web page soon.
Cheers,
Bob
soongsc said:
Sir - I sometimes think that you post just to stir up flame wars. Please desist as we are having a discussion here that has some merit.
Hi
Earl, below is what my soundcard measures when averaging is set to 10. Not exactly comparable to your method but certainly the hardware seems to be good enough for your trick, no ?
The stimulus peak at 1 kHz is set to be automatically 0dB at every measurement which means that the noise floor for every trace stays roughly below 120 dB.
I am not so familiar with MathCad and frankly I DO LIKE a simple and intuitive graphic user interface for any computer work.
Do you see any chance to make a piece of software for easy use ?
Maybe you can cooperate with one of the existing software developers and adding a special "Geddes" measurement module ?
You might check out audioTester at
http://www.audiotester.de/
to see if that is something you would like.
Ulrich Müller as the developer is a responsive and credible person.
Also audioTester already is a capable and well maintained software with good distortion analysis and worked flawless with every sound card I ever plugged
Greetings
Michael
Earl, below is what my soundcard measures when averaging is set to 10. Not exactly comparable to your method but certainly the hardware seems to be good enough for your trick, no ?
The stimulus peak at 1 kHz is set to be automatically 0dB at every measurement which means that the noise floor for every trace stays roughly below 120 dB.
I am not so familiar with MathCad and frankly I DO LIKE a simple and intuitive graphic user interface for any computer work.
Do you see any chance to make a piece of software for easy use ?
Maybe you can cooperate with one of the existing software developers and adding a special "Geddes" measurement module ?
You might check out audioTester at
http://www.audiotester.de/
to see if that is something you would like.
Ulrich Müller as the developer is a responsive and credible person.
Also audioTester already is a capable and well maintained software with good distortion analysis and worked flawless with every sound card I ever plugged
Greetings
Michael
mige0 said:
OOPs Hit "Submit" a bit too soon.
Thanks for that link. His stuff looks very interesting. I may try it out.
But, you could not be doing what I suggested because 1 kHz cannot exactly fit into any time base that I know of so your averaging cannot be sychronous. Spectral averaging is a well know technique and works well, but only down to a limit. My idea was to use an exact fir sine wave and sychronously averag the time signal NOT the spectrum. Theoretically this is capable of infinite S/N ratio (although one could never actually get there).
I am not sure that what I am suggesting is needed or not. I am sure that we need to look at the spectrum of the very lowest signal that we can send through a piece of equipment. We need to see how the harmonics of the signal are distributed at very low signal levels.
It could well be that this can be done without any fancy techniques. I don't have enough experience with amps to know at this point.
But I am trying to get people to realize that THD and THD versus level or THD versus frequency or IMD or any of the classic measurements don't really tell us a thing about how an amp will sound. Yet these measurements abound and useful ones are non-existant.
Does this make sense to you guys?
Lets turn on the lights and see what we are doing for a change.
there are 2 main approaches: "black box" and "white box"
in Black box measurement/modeling you don't make assumptions about the internal construction of the amp, Volterra Series modeling is the most general for smooth nonlinearities and requires multi tone frequency sweeps with increasing numbers of tones for the order of nonlinearity you want to identify
White Box modeling/measurement involves inspecting circuit properties with an eye towards comming up with specially structured signals to fit the topology/bias/load conditions - the Geddes low level crossover test presented here is an example based on the assumption of class B output stage, Self's ideas may be a place to start:
http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm
section 5.3 concerns output device behavior in Class B vs A/B
with a substantial AB bias, which I believe our commercial product designers here seem to favor, a more sensitive output stage commutation distortion test would be a 2 tone intermod test that just drew enough current to straddle the notches between the "gm doubling" and single device conduction curves - a low level test would only exercise the Class A "gm doubling" region (Self's fig 14-17)
in Black box measurement/modeling you don't make assumptions about the internal construction of the amp, Volterra Series modeling is the most general for smooth nonlinearities and requires multi tone frequency sweeps with increasing numbers of tones for the order of nonlinearity you want to identify
White Box modeling/measurement involves inspecting circuit properties with an eye towards comming up with specially structured signals to fit the topology/bias/load conditions - the Geddes low level crossover test presented here is an example based on the assumption of class B output stage, Self's ideas may be a place to start:
http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm
section 5.3 concerns output device behavior in Class B vs A/B
with a substantial AB bias, which I believe our commercial product designers here seem to favor, a more sensitive output stage commutation distortion test would be a 2 tone intermod test that just drew enough current to straddle the notches between the "gm doubling" and single device conduction curves - a low level test would only exercise the Class A "gm doubling" region (Self's fig 14-17)
jcx said:
Your last point is a very good one, and reflects my preference for the CCIF test with spectral analysis. Doing that test in simulation is also quite useful.
Bob
jcx said:
Just a point of clarity. Volterra series modeling is not the "most general" and is in fact much more limited in application than Weiners approach (See "The Volterra and Wiener Theories of Nonlinear Systems" by Schetzen) and there are a number of ways to find the coefficients of the series involved.
But having gone through all this, I found that it was not not an effective approach. The higher the order the more audible it is, but the more difficult the analysis and the more unstable the results. This means that we have the least accuracy where we have the most need.
This is readily apparent if one uses Klippels system to measure the orders of nonlinearity of a loudspeaker. A guage R&R study will show that orders above about the third are not reliably reported - but second and third order are not audible. This is a problem.
jcx said:
It matters not what the mechanism is for low level distortion only that it is the low level distortion that is audible. If you want to design for the numbers thats fine - you will get the kind of amplifiers that enumerate today and the discussions about which amplifier sounds best will continue.
But if you want to design amps that "sound good" then you have to design for audibility, and its the low level stuff that matters, regardless of where or how this occurs.
Until measurements can be proven to correlate with subjective impression - and I and others have clearly shown that they don't - its simply a numbers game with no relavence to reality.
PMA said:
I'd have to see the spectrum at several output levels along with a comparison with these other amps using the same test.
Your test appears to be at a pretty high output level since the noise floor is so far down.
No, it is at 1W/8ohm. You have to read 1st harmonic level, which is at about -17dB of Y axis. I can provide you set of mesurements from 1W to 25W, no problem.
- Status
- Not open for further replies.