peranders said:
On the contrary, 'truth' does not change with time....
i.e: 'Truth' is time-invariant....
Only what we accept to be the 'truth' changes with time....
viz:
....Lies become accepted as 'truth' as time tends to infinity....
Myths
Mikeks,
Very nice the 14th Myth debunk. The "origin" is great!!!
Congratulations for the tip.
regards,
Mikeks,
Very nice the 14th Myth debunk. The "origin" is great!!!
Congratulations for the tip.
regards,
i told myself I wouldn't enter this, but ...
what's the definition of "truth" ?...
everybody knows the rest of the questions, right?
what was the objective of this topic again ? ...
what's the definition of "truth" ?...
everybody knows the rest of the questions, right?
what was the objective of this topic again ? ...
mikeks said:
Re: Re: more reason to look up the Czerwinski article
Mikeks:
I am not quite sure about an answer to this. Nonlinear systems theory makes use of different methods (matrix measure, differential geometry) to address what are the main issues of control theory i.e. stability, bifurcation, chaos etc.
This does not imply frequency domain amplitude/phase response or its transform time domain impulse response are useless either, for they are very powerfull tools. The question - which probably someone else here may ilustrate us about - is whether single frequency THD+N characterization allows to determine reliably an upper bound to all nonlinear byproducts under real program input.
What I do personaly agree with anyway, is with the idea that there do not exist "ghost" perturbations that cannot be measured in the lab but do have an audible effect.
I want to stress again post #141 what I now learned is known as the "null test".
Not only this, but also to excercise the system not with a single sinusoidal excitation or combination of similar stimuli, but with real program source material as will be processed in real life.
Rodolfo
mikeks said:
Mikeks:
I am not quite sure about an answer to this. Nonlinear systems theory makes use of different methods (matrix measure, differential geometry) to address what are the main issues of control theory i.e. stability, bifurcation, chaos etc.
This does not imply frequency domain amplitude/phase response or its transform time domain impulse response are useless either, for they are very powerfull tools. The question - which probably someone else here may ilustrate us about - is whether single frequency THD+N characterization allows to determine reliably an upper bound to all nonlinear byproducts under real program input.
What I do personaly agree with anyway, is with the idea that there do not exist "ghost" perturbations that cannot be measured in the lab but do have an audible effect.
I want to stress again post #141 what I now learned is known as the "null test".
If you check again post #141, you will notice I propose in advance both a high resolution data aquisition and mathematical processing so as to correct for all known and characterized linear components present as deviations from the ideal signal.John Curl said:
Not only this, but also to excercise the system not with a single sinusoidal excitation or combination of similar stimuli, but with real program source material as will be processed in real life.
Rodolfo
mikeks said:...is that a 'yes'?
If you were wondering about THD+N, my answer is "don't know".
Was that?
Rodolfo
Re: Re: Re: more reason to look up the Czerwinski article
Hi Rudolfo
I certainly agree with you. If exist any perturbation at the output they must be measured in a null test.
About my way of making the null test see post #8 at.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=13415
Saludos!
ingrast said:
Hi Rudolfo
I certainly agree with you. If exist any perturbation at the output they must be measured in a null test.
About my way of making the null test see post #8 at.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=13415
Saludos!
john curl said:
Agreed. I guess I'm thinking power amps and you're thinking op-amps. This is a good argument for FET input stages in low-noise designs. And yes, I've had a course out of Motsenbacher and Fitchen back in 1980 so I am aware of these issues. Marshall teaches that course now but didn't back then.
Well, yeah. Since a simple harmonic distortion measurement usually notches out the fundamental (unless you're working with just a spectrum analyzer), it's obscuring what's going on with the fundamental, namely the variable phase shift with level. But the same mechanism that produces the variable phase shift with level also produces harmonics and IM. It's just that, given these harmonics and/or IM, you can't know by looking at them whether they came about as a result of AM-to-AM or AM-to-PM distortions.
Ahhhhh, I think I'm seeing what your concern is now. Okay, how about this? Tomorrow, I'll go through Gilbert's equations, picking a test frequency that shows a significant variable phase shift with output level. I'll do a plot. On the x-axis I'll put either the input or output level, whichever makes more sense. On the left y-axis I'll put the deviation of the input-output phase shift from its ideal value assuming a linear system. On the right y-axis, I'll put the harmonic distortion (third harmonic only, since Gilbert's analysis neglects the others). Then, by drawing a horizontal line on the graph at a given phase modulation level, you'll be able to see what the corresponding harmonic distortion level is. I honestly don't know what this plot will look like, but I think it will be interesting. Or, as Beavis and Butthead would say, "this is going to be cool!".
Are you trying to tell me that the 741 isn't the be-all and end-all of audio op-amps? Geez, where have I been?
john curl said:
Valid point John. Have you seen a schematic of the late Deane
Jensens discrete 990 opamp. LM394 IP with degenerated
emitters bypassed with small inductors so at AF the R's are
out of the noise equation.
Cheers,
Terry
Andy C I got my copy of Motchenbacher and Fitchens' 'Low-Noise Electronic Design' when it first came out about 30 years ago. It was the first really good book on low noise design, and its only limitation today is that there are much better parts that were developed in Japan after the book was published.
The technique used by the late Deane Jensen is more easily done by using a low noise jfet pair in the input stage. It gives the same short circuit noise, but reduces input bias current and sensitivity to input loading.
The technique used by the late Deane Jensen is more easily done by using a low noise jfet pair in the input stage. It gives the same short circuit noise, but reduces input bias current and sensitivity to input loading.
Jorge:
Thanks for the April 2003 thread link regarding "null test" results. I was not around here by then.
It is amazing even in a basic hardware analog form it did nonetheless provide convincing results! Like the Bob Carver - Stereophile challenge.
Now things can be done in a much more sophisticated and precise way. A high end stereo sound card (the ones priced at a few hundred dollars) could be used to acquire input and output signals for postprocessing. Wish I had time to spare for this!
Saludos, Rodolfo
Re: Re: Re: more reason to look up the Czerwinski article
I don’t know what you guys want, I can only assume I somehow hid my point in my “hidden distortion” Post #298
http://www.diyaudio.com/forums/showthread.php?postid=492499#post492499
A single frequency sine sweep showed –74 dB distortion (230 ppm, 0.00023 %, 230 uV) in my test circuit – in my first graph V(out) is multiplied by 1000X to be visible at all
The second plot with a 2 tone excitation shows –14 dB distortion (240000 ppm, 24%, 240 mV) product V(out)
I showed that you can get 1000 times more distortion with the 2 tone excitation at the same Vpp input in a "simple" nonlinear model circuit, how much plainer can I get?
I also pointed out at least one mechanism in transistors where 2 paths for signal with different frequency responses are multiplied, ie the power dissipation caused temperature modulation of the transistor modulates the Vbe and Hfe of the transistor and the transistor die temperature is clearly low pass filtered by thermal mass and package thermal resistance – the transistor may be amplifying a signal increasing in amplitude with frequency (error voltage in a integrating feedback loop, current into a Miller Cdom or power mosfet gate) – exactly the condition for creating a nonlinearity you cannot evaluate with a single frequency test
Czerwinski clearly shows this with his example 2nd order Volterra kernel surface plot and justifies his whole multitone approach on the Volterra kernel property that a n-th order nonlinearity can only be fully characterized/represented by a n dimensional surface with n frequency axis – n independent frequencies have to be simultaneously present and independently swept to cover the whole distortion kernel surface
So mike, the answer is NO, the magnitude of the distortion/intermodulation products with real music signals cannot be derived from the THD of a single frequency sine sweep
mikeks said:
ingrast said:
I don’t know what you guys want, I can only assume I somehow hid my point in my “hidden distortion” Post #298
http://www.diyaudio.com/forums/showthread.php?postid=492499#post492499
A single frequency sine sweep showed –74 dB distortion (230 ppm, 0.00023 %, 230 uV) in my test circuit – in my first graph V(out) is multiplied by 1000X to be visible at all
The second plot with a 2 tone excitation shows –14 dB distortion (240000 ppm, 24%, 240 mV) product V(out)
I showed that you can get 1000 times more distortion with the 2 tone excitation at the same Vpp input in a "simple" nonlinear model circuit, how much plainer can I get?
I also pointed out at least one mechanism in transistors where 2 paths for signal with different frequency responses are multiplied, ie the power dissipation caused temperature modulation of the transistor modulates the Vbe and Hfe of the transistor and the transistor die temperature is clearly low pass filtered by thermal mass and package thermal resistance – the transistor may be amplifying a signal increasing in amplitude with frequency (error voltage in a integrating feedback loop, current into a Miller Cdom or power mosfet gate) – exactly the condition for creating a nonlinearity you cannot evaluate with a single frequency test
Czerwinski clearly shows this with his example 2nd order Volterra kernel surface plot and justifies his whole multitone approach on the Volterra kernel property that a n-th order nonlinearity can only be fully characterized/represented by a n dimensional surface with n frequency axis – n independent frequencies have to be simultaneously present and independently swept to cover the whole distortion kernel surface
So mike, the answer is NO, the magnitude of the distortion/intermodulation products with real music signals cannot be derived from the THD of a single frequency sine sweep
Re: Re: Re: Re: more reason to look up the Czerwinski article
Amen !
Hopefully everybody understood now that a sinewave is not representive at all...
a single frequency sine sweep is good for testing bandwidth and linearity, not quality.
jcx said:
Amen !
Hopefully everybody understood now that a sinewave is not representive at all...
a single frequency sine sweep is good for testing bandwidth and linearity, not quality.
As to theoretical limits on real music signal IMD derived from single frequency THD measurements I’m not optimistic that they can be useful, particularly when the handiest of today’s tools for measuring amplifiers are PC sound cards and multitone measurement is simply a software issue
But I think an approximate limit could be guessed at when considering only the 2 port 2nd order case, and (more realistic) 1st order slopes of opposing frequency selectivity over the 20 Hz –20 KHz audio frequency range - I find in sim a ~ 150X multiplier for the worst case 2 tone vs the peak single tone distortion amplitude
It remains to be shown (by multitone meausrements) that audio amplifiers have only such a low order distortion kernel surface curvature but I guess successful audio amplifier circuits must be pretty good in this respect
Certainly it would seem that amplifier circuits with truly gross distortion from “lumpy” higher order Volterra kernel surfaces would be known and consequently avoided
But I think an approximate limit could be guessed at when considering only the 2 port 2nd order case, and (more realistic) 1st order slopes of opposing frequency selectivity over the 20 Hz –20 KHz audio frequency range - I find in sim a ~ 150X multiplier for the worst case 2 tone vs the peak single tone distortion amplitude
It remains to be shown (by multitone meausrements) that audio amplifiers have only such a low order distortion kernel surface curvature but I guess successful audio amplifier circuits must be pretty good in this respect
Certainly it would seem that amplifier circuits with truly gross distortion from “lumpy” higher order Volterra kernel surfaces would be known and consequently avoided
Is it true that scope trace of piano sound is not a perfect sinusoidal, like the Upper and Lower traces are not the same? If musical tones are not sinusoidal, why we test amp with sinusoidal traces?
By Mr. JC directions, I found that Sound Card manufacturer are very concern about these distortions. They can demonstrate AM, FM.
Sound Cards have better specification than power amps?
Re: Re: Re: Re: more reason to look up the Czerwinski article
Hi jcx 🙂
Thanks for that....
I have tried to duplicate your findings in Multisim with a 'real' circuit model, i.e input diff. stage second stage TIS...etc......, and i fear changes registered by the multitone method are also registered by ordinary THD analysis.....
Perhaps if you replace your non-linear model with a more practical and realistic, nominaly linear amp. circuit, (say one of rod elliot's designs), you may obtain greater insight?
Cheers.
mikeks said:
jcx said:
Hi jcx 🙂
Thanks for that....
I have tried to duplicate your findings in Multisim with a 'real' circuit model, i.e input diff. stage second stage TIS...etc......, and i fear changes registered by the multitone method are also registered by ordinary THD analysis.....
Perhaps if you replace your non-linear model with a more practical and realistic, nominaly linear amp. circuit, (say one of rod elliot's designs), you may obtain greater insight?
Cheers.
Well done, and said, JCX. I looked at the distortion plots, but did not analyze them fully. Thanks for the post analysis.
Actually the same thing can be shown in a direct radiator loudspeaker. As you know, the loudspeaker must move in and out, in order to produce sound. This causes the frequency of each not played by the loudspeaker to change frequency slightly as the cone is going forward, and then change in the other direction when the cone is going backward.
From what I remember, this distortion can be seen as 1'st order IM (2nd harmonic), but not measured as harmonic distortion.
This is most probably similar to what Barrie Gilbert's article is pointing to.
Actually the same thing can be shown in a direct radiator loudspeaker. As you know, the loudspeaker must move in and out, in order to produce sound. This causes the frequency of each not played by the loudspeaker to change frequency slightly as the cone is going forward, and then change in the other direction when the cone is going backward.
From what I remember, this distortion can be seen as 1'st order IM (2nd harmonic), but not measured as harmonic distortion.
This is most probably similar to what Barrie Gilbert's article is pointing to.
"And yes, i am talking of delays below 1uS.
For example, you have a mixture of 1Khz and 5khz. A bad amp now
for example has a delay of 200nS at 1Khz and 250nS at 5khz."
Assuming that this is true. How can it possibly be measured? At 1Khz a 200nS delay is less than 1/2 degree. No way a Thd meter is going to catch this.
For example, you have a mixture of 1Khz and 5khz. A bad amp now
for example has a delay of 200nS at 1Khz and 250nS at 5khz."
Assuming that this is true. How can it possibly be measured? At 1Khz a 200nS delay is less than 1/2 degree. No way a Thd meter is going to catch this.
Since I feel somewhat responsible for this "THD war" I will get into the food-fight again...😉
Let me first say that I did NOT suggest that simple THD + IMD measurements were nowhere near enough to desbribe the SQ of a system. My point was that I do not feel that audio distortion is a "black magic" subject but rather one that can be quantified by scientific means. But it has not been done yet and that is probably because there have been limited scientific interest in this area.
I have approached this from an academic point of view (THD is pretty useless for SQ quantification anyway). This means that I don't care for the available measurement equipment accuracy. Today it might be 0.000001 %, tomorrow something else. But I gave different distortion mechanisms a thought.
First, let me adress the distortion induced by thermal effects as investigated by jcx. Suppose the thermal time constant is in the order of 0.1-1 seconds. Will a THD measurement at 1 kHz be valid then? - No, for a fixed amplitude not. But an fq of 0.1-10 Hz will certainly discover this nonlinearity provided that this falls within the fq range of the amp. I am assuming that the fq response of the amp is flat and extends between DC and far beyond the highest fq of interest since we all know that THD is not valid if these conditions are not met.
But this will only be of academic interest since THD will never tell us
what caused the distortion. And what is more interesting is measurements for the whole sound system, that is cascaded nonlinear stages. For those unfamiliar with such analysis I can bring an example:
Assume we start with dual test tones at say 8 kHz and 5 kHz from the CD player. The preamp will have both THD & IMD so arriving at the power amp we have n*(5kHz) + n*(8kHz) + 3 kHz + 13 kHz. The power amp and speaker also have both THD & IMD (the speaker lots of it too!) so coming to our ears we have:
1 kHz (2*13 - 5*5)
2 kHz (2*13 - 3*8)
3 kHz (8-5)
4 kHz (2*8 - 4*3)
5 kHz (fundamental)
6 kHz (2*3)
7 kHz (3*5 - 8)
8 kHz (fundamental)
9 kHz (3*3)
10 kHz (2*5)
11 kHz (3*8-8+5)
12 kHz (4*3)
13 kHz (8+5)
... and on and on, I think you get the picture... Quite a little orchestra! 😱 So clearly, THD or IMD of one component alone is pretty useless. But with today's computers it would probably not be too difficult to devise a series of measurements which put together would tell the big picture pretty well.
One of my professors at school has used this method (multivariable data processing) to differentiate between different skin cell types with the use of a simple skin conductivity measurement (EIT). The system is in clinical use in Sweden today and has a >90% sensitivity and specificity of detecting malign skin cells. Inspection by a skin specialist has a specificity of < 10%... You can even tell the gender with good accuracy with this test!
But quite clearly, the "high-end" industry has nothing to gain from a scientific approach. I guess we have to wait until cochlea implants are standard issue.
/Magnus
Let me first say that I did NOT suggest that simple THD + IMD measurements were nowhere near enough to desbribe the SQ of a system. My point was that I do not feel that audio distortion is a "black magic" subject but rather one that can be quantified by scientific means. But it has not been done yet and that is probably because there have been limited scientific interest in this area.
I have approached this from an academic point of view (THD is pretty useless for SQ quantification anyway). This means that I don't care for the available measurement equipment accuracy. Today it might be 0.000001 %, tomorrow something else. But I gave different distortion mechanisms a thought.
First, let me adress the distortion induced by thermal effects as investigated by jcx. Suppose the thermal time constant is in the order of 0.1-1 seconds. Will a THD measurement at 1 kHz be valid then? - No, for a fixed amplitude not. But an fq of 0.1-10 Hz will certainly discover this nonlinearity provided that this falls within the fq range of the amp. I am assuming that the fq response of the amp is flat and extends between DC and far beyond the highest fq of interest since we all know that THD is not valid if these conditions are not met.
But this will only be of academic interest since THD will never tell us
what caused the distortion. And what is more interesting is measurements for the whole sound system, that is cascaded nonlinear stages. For those unfamiliar with such analysis I can bring an example:
Assume we start with dual test tones at say 8 kHz and 5 kHz from the CD player. The preamp will have both THD & IMD so arriving at the power amp we have n*(5kHz) + n*(8kHz) + 3 kHz + 13 kHz. The power amp and speaker also have both THD & IMD (the speaker lots of it too!) so coming to our ears we have:
1 kHz (2*13 - 5*5)
2 kHz (2*13 - 3*8)
3 kHz (8-5)
4 kHz (2*8 - 4*3)
5 kHz (fundamental)
6 kHz (2*3)
7 kHz (3*5 - 8)
8 kHz (fundamental)
9 kHz (3*3)
10 kHz (2*5)
11 kHz (3*8-8+5)
12 kHz (4*3)
13 kHz (8+5)
... and on and on, I think you get the picture... Quite a little orchestra! 😱 So clearly, THD or IMD of one component alone is pretty useless. But with today's computers it would probably not be too difficult to devise a series of measurements which put together would tell the big picture pretty well.
One of my professors at school has used this method (multivariable data processing) to differentiate between different skin cell types with the use of a simple skin conductivity measurement (EIT). The system is in clinical use in Sweden today and has a >90% sensitivity and specificity of detecting malign skin cells. Inspection by a skin specialist has a specificity of < 10%... You can even tell the gender with good accuracy with this test!
But quite clearly, the "high-end" industry has nothing to gain from a scientific approach. I guess we have to wait until cochlea implants are standard issue.
/Magnus
- Status
- Not open for further replies.