Chance to link that again: http://www.aip.org/pt/nov99/locsound.html
Above 1.5K we are binaurally 'death' to synchronous detection. But, we still can detect waveform *envelope* binaural timings. Thus, unequal phase shifts in channels can result in imaging issues, even non-flat group delay response alone can, due to changes to waveform and envelope.
I have no idea how much acoustic output envelope changes does backemf damping incur, but I'd guess that could be part of explanation why amp-speaker interface has high impact on imaging, and often timbre perception. But these are dynamic linear distortions..
I believe we can pick recorded from real thing thanks to another matter. There are almost no recordings these days that do not apply dynamic compression. That is what makes it sound unnatural. But for natural sounding, we'd need to listen at very high SPL levels. One piano produces 0.5 watts of acoustic power. To reproduce that with speakers of typical 0.5-2% efficiency, you'd need 50-100W listening levels. To reproduce 60W acoustic of orchestra, you'd need few KW. Not for consumers, thus the compression.
When you go to live events, there is adequate sound reinforcement and little to none compression used. And it does sound more natural, despite all the horror stories of PA quality. When CD's are played through same systems, they sound like ducks singing.
Above 1.5K we are binaurally 'death' to synchronous detection. But, we still can detect waveform *envelope* binaural timings. Thus, unequal phase shifts in channels can result in imaging issues, even non-flat group delay response alone can, due to changes to waveform and envelope.
I have no idea how much acoustic output envelope changes does backemf damping incur, but I'd guess that could be part of explanation why amp-speaker interface has high impact on imaging, and often timbre perception. But these are dynamic linear distortions..
Human spectral resolution isn't very narrow when it comes to SPL and loudness. That is very related to masking effects. When it comes to loudness, I believe it has to do with total acoustic energy delivered. There is special mechanism in ears to supress excessively loud sounds, for adaptive dynamic range. Whenever that mechanism is triggered, it is perceived as loudness. It can be triggered by single narrowband stimulus, or equivalent wider band stimulus. I'd actually imagine that there are conditions when sounds bathed in loud widerband noise sounds actually weaker than lower SPL sounds in silent surroundings thanks to our ears adapting to background noise.
I believe we can pick recorded from real thing thanks to another matter. There are almost no recordings these days that do not apply dynamic compression. That is what makes it sound unnatural. But for natural sounding, we'd need to listen at very high SPL levels. One piano produces 0.5 watts of acoustic power. To reproduce that with speakers of typical 0.5-2% efficiency, you'd need 50-100W listening levels. To reproduce 60W acoustic of orchestra, you'd need few KW. Not for consumers, thus the compression.
When you go to live events, there is adequate sound reinforcement and little to none compression used. And it does sound more natural, despite all the horror stories of PA quality. When CD's are played through same systems, they sound like ducks singing.
I've understood that this is more of effect due to current drive vs. voltage drive. Not necessarily tubes or harmonics, perhaps more of speaker's efficiency in electric to acoustic conversion.
amplifierguru said:
then try reading Czerwinski! - SMTE IM single point measurement is nowhere near the full IMD story
the correlation you refer to is dependent on many assumptions, the 1st and most confusing to people trying to understand "distortion" in the larger context is referring to these narrowly defined single point measurements and implying they measure audible distortion - simply not true – I had hoped this discussion had already got beyond that
a thought experiment can show how far off even the correlation between these specific narrowly applicable measurements you refer to could be:
common audio power amplifier loop gain shapes include flat loop gain (as recommended by Otala/PIM devotees) over the audio range, 1st order "dominant pole" and 2-pole compensation with rising loop gains below the assumed 20 KHz reference
suppose we have only 3rd order nonlinearity from input bjt diff pair - then loop gain @ 6 KHz will vary - in the 2-pole case >10x less diff error voltage vs the flat gain case leads to 100x less 3rd order distortion of the 6 KHz component, you can do the math for ( 60 Hz / 20 KHz )**2 and tell us how much SMTE IMD is likely to differ in 2-pole compensation vs the flat loop gain case
it really tells us very little to know that the 20 KHz THD is "bigger"
Czerwinski should be the minimum basis for discussing distortion in the audio reproduction chain, Moore and Tan are moving forward with linking auditory models to distortion perception - by including all of the distortion products (where in general IM products dominate)
Amen - I have noticed this over and over: a CD playing over the same sound system after the live band at a club sounds flat, uninvolving, and downright ridiculous. So, we may blame dynamic compression up to a point.
I don't believe this applies to the piano case, though. I would assume that the recording of a solo piano does not involve significant use of compressors. Yes, realistic SPL even of a single piano demands quite some power, but achievable power in a home environment (unlike the complete orchestra - but the real orchestra would likely not play in your living room, while the real piano just might). So, why does it still sound so relatively unrealistic on any music system I have heard?
I believe part of the answer lies in the physical size of instrument vs. speaker - radiative surface and room interaction etc. In addition to that, with multiple complicated harmonics the piano would have a multitone structure at any point, and here massive contamination of the harmonics through wideband amp and speaker distortion could simply mask the natural timbre of the piano.
The distortion story must go beyond sheer SPL issues because I at least can easily pick out a real piano from a recorded one even from the street, if someone plays a piano inside a neighbour's house. I have never even wondered whether a sound came from a real piano or a recorded one. It just sounds so obviously different, and IMO, the recorded one sounds less transparent, less striking, and less rich. The recorded one *does* sound masked/muffled.
Regarding tubes
Not sure if I follow. Even typical tube amps AFAIK use voltage drive, albeit at a slightly higher output impedance. And I can't see how any of this would affect the speaker's efficiency. But I can well imagine how a more distorted signal with more wideband elements, specifically, can sound louder.
In fact from my experience, I can play my system at a very wide range of actual SPL without having my system sound "loud". I can only tell the actual SPL when I try to speak... Things start to sound "loud" , subjectively, when some elements go nonlinear (clipping of amp, speaker leaves linear xmax). This change from "OK" to "loud" IME occurs abruptly within a few dB change.
I think we're on a bit of a witch hunt here for distortions that simply don't exist!
jcx
Agreed the correlation depends on many assumptions but it's significantly reliable to be a 'rule of thumb'. Of course I'm also assuming that 20KHz THD is the highest and it decreases monotonically at lower frequencies. MOST CASES
Your example of the dominant THD being due to the input stage nonlinearity and as such varies with the diffl signal level in accordance with the higher (2pole) or lower (1pole) open loop gain and also varies when closed loop gain is changed due to altered CM range at the input. This is often observed in IC op amps where the inverting mode is used for THD critical applications.
IM distortion products within the loop will be reduced according to where they are generated within the loop. i.e gain before vs gain after just like PS intrusion.
Phase distortion, secret IM distortions, Spanish Inquisitions
Seems as though some are looking for excuses???
jcx
Agreed the correlation depends on many assumptions but it's significantly reliable to be a 'rule of thumb'. Of course I'm also assuming that 20KHz THD is the highest and it decreases monotonically at lower frequencies. MOST CASES
Your example of the dominant THD being due to the input stage nonlinearity and as such varies with the diffl signal level in accordance with the higher (2pole) or lower (1pole) open loop gain and also varies when closed loop gain is changed due to altered CM range at the input. This is often observed in IC op amps where the inverting mode is used for THD critical applications.
IM distortion products within the loop will be reduced according to where they are generated within the loop. i.e gain before vs gain after just like PS intrusion.
Phase distortion, secret IM distortions, Spanish Inquisitions
Seems as though some are looking for excuses???
Indeed.
I'm afraid that when known distortions are orders of magnitude greater and containable with good design (and PSRR and CMRR are all about THD), while secret and nefarious distortions at least an order of magnitude less are discussed at length and given undeserved notoriety, I ...
Have a nice day.
I'm afraid that when known distortions are orders of magnitude greater and containable with good design (and PSRR and CMRR are all about THD), while secret and nefarious distortions at least an order of magnitude less are discussed at length and given undeserved notoriety, I ...
Have a nice day.
So... which of the distortions dicussed here, in your opinion, doesn't exist? And which known distortion, if any residual remaining and not contained, produces the result which we all know, i.e., that reproduced sound does not really sound like the real sound?
That sounds a lot more scientific to me: no such thing as a single parameter explaining everything. Measurements are not necessarily wrong per se, but one may be measuring an irrelevant parameter.
That sounds a lot more scientific to me: no such thing as a single parameter explaining everything. Measurements are not necessarily wrong per se, but one may be measuring an irrelevant parameter.
Very possible
But I think it is worthwhile breaking the discussion in to two components. Distortion mechanisms and distortion measurements. I think the two are often confused and confuted.
There is no shortage of new ideas for distortion mechanisms, there are more pet theories about how some hitherto unforeseen issue might give rise to non-linearity than I can count. And indeed the vast majority seem to be pretty much a non-issue.
But, one of the usual arguments to dismiss them is to measure THD. The issue of metrics of distortion is still wide open. The GedLee paper for one essentially presented evidence that there is almost no correlation between a conventional THD measurement and perceived quality. This should give serious pause for thought.
To reiterate my posting from before, there is actually little to recommend THD as a metric of quality except ease of implementation. Intrinsically in its implementation a great deal of information is discarded - this alone should make one very very suspicious. I would want to hear a very good story that shows that the discarded information is actually of no value - and this would eventually turn into an argument that the ear/brain cannot receive any of the information in the discarded component of the spectra. I very much doubt anyone would be willing to make that call (and it is trivially refuted anyway.)
A contrived example. Consider a system where a single tone is input and the same, quite undistorted tone is output, along with a swept tone that travels from 20Hz to 20kHz every two seconds. We could construct a THD measurement that simply displayed this swept tone as a (very) slightly raised noise floor.
I think this where I must disagree. They aren't ALL about THD. THD presents us with a steady state metric of the signal correlated components of the distortion resultant from PSRR and CMRR.
PSRR reflected into THD does not take into account rejection of any extraneous noise (for instance power supply ripple.) Similarly both PSRR and CMRR may change in a signal correlated manner, and may have other time varying mechanisms. THD will not show these effects. It intrinsically cannot.
THD+N does, where the other distortion products simply appears lumped under noise. And it is then quite commonly ignored. Which is a significant mistake. So often one sees an analysis where the noise floor is a bit higher, but discussion concentrates upon the harmonic peaks. Ignoring the fact that there may well be quite discernible energy under the noise floor shown by the FFT. This is because the FFT is run with a basis function set that works in harmonic space, a different basis set would show quite clear correlated energy in a form that can be heard.
It is crucial to realise this - the noise floor in a simple THD plot is not just AIWN. Analysis with a different basis function set can find other signals that we can hear. Indeed the masking properties of the ear seem to suggest that the distortion products we see in THD are actually the least intrusive, and thus those products smeared across the noise floor of an FFT are significantly more important.
amplifierguru said:
Hi amplifierguru,
I did manage to dig up a fairly legible copy of the 3rd article.
I don't really want to comment on the amp design although I do
think it is excellent and may have to look into it a bit more
in the future.
I'll focus on the distortion measurement techniques instead
because this is what is being used to judge whether
IM is significant or not. I think all would agree that it would be
easy to dismiss IM if the techniques used to measure IM
did not turn much of it up. I think this is roughly the case when
using these types of tests. They are not showing how IM does
damage. Fortunately the Czerwinski article shows how IM and HD
are very strongly linked. The mathematics aren't to bad (at least
in the middle) and probably helped guide the authors in devising
a test that can measure IM effectively and explain how it works.
First the SMPTE test uses a tone at 60Hz and 7Khz in a 4:1
ratio. This is meant to expose IM as sidebands around the upper
signal seperated in multiples of the lower signals frequency.
The problem with this is that a lot of IM products occur below the
two frequencies and 60Hz doesn't leave room for much bad
behavior.
Then there is the DIM-XX. This is a square wave with a smaller
hi frequency sine wave riding on top of it. The square wave is
bandwidth limited. This test is designed to cause the sine wave
to misbehave (produce distortion). As the square wave goes
from low to hi or hi to low, these sudden transitions will cause
slewing problems that appear as harmonic products of the sine
wave.
What I've found in simulations of DIM-XX is that the sine wave
does produce harmonics, but they are like small needles in a
vast forest of square wave IM products. This test just measures
the tiny little distortion products of the sine wave and ignores the
rest. I do believe it does what it was designed to do. Show the
effects of slew rate induced distortion, but there is (at least in my
opinion) a very small amount of distortion caused by this type of
phenomena.
I think the CCIF measurement is good because it uses a pair of
closely spaced signals with ample room below them to produce
IM products. The only difference between this and Czerwinsky
is that the later uses more tones and uses a logarithmic spacing
of the tones so they are not harmonically related. BTW, Cordell
did not use the CCIF IM measurement technique in his article.
So I've attached a picture of a Czerwinsky simulation result
using 5 equal level tones spaced logarithmically apart in
frequencies of 100, 177.8279, 316.2278, 562.3413, 1000Hz.
You can see the fourier transform looks like a lot of grass from
all the IM products. The spacing is chosen so the products don't
overlap each other like the IM products from a square wave
do.
So I hope this at least opens the door to considering that
there is more going on then a 20Khz THD measurement. I'd
highly recommend looking at the 5 tone measurement. Who
knows, maybe your amplifier designs will work. You will never
know unless you look
Mike
Attachments
Hi MBK,
Yes your description - 'wall' of 'spectral contamination' - hits home, and it is only by our listening based investigations that we have made genuine progress.
That is; we have managed to get away from accepting designer's statements of measurement proven technical perfection.
With the likes of reproduced piano maybe it could be a loss of 'spectral contamination' due its single point of recording that has the major influence. Does the same detectability apply with headphone listening ?
Also, tube amps *do* go cleanly louder for a given sine/resistor rating due to their increased ability to voltage (anode headroom) follow when loudspeaker impedance increases, also due to their low grid charge storing characteristics with good/new space charging tubes during half cycles when load impedance falls and the anode is not overheated.
I would support Francis and JCX in relation to solid state IM distortion. (JCX, your words are not wasted, horses - water etc.)
It was the same with radio receivers - transistor communications receivers were incompetent beside valve types but the designers would not admit it! The solid state types could be good only when sharply tuning pre-selectors were used to filter out *other frequencies*, this until all signal voltages were held to minimum levels wrt circuit noise and using specialised filters with nested agc loops.
Hi Wimms.
Yes, Dolby and more recent algorithm exponents have much to answer for, especially Philips/Sony for our CD medium, but we do appear to agree that so many loudspeaker driving amplifiers still make the problems worse because of dynamically induced current flows through the cable linked amplifier-loudspeaker interface affecting both imaging and timbre perception.
The same dynamically induced current flows do not arise with full range electrostatics/Magneplanars etc.
Hi Francis.
I agree that 5kHz is a high pitch, but timebase whistle can still be aurally nulled by head movement as long as it is not fully finger attenuated.
Actually, I often listen for amplifier/loudspeaker reproduction errors by using my fingers in my ears as variable attenuators, this to get close to a tweeter or midrange without total aural masking (deafening), and/or to examine room reflections.
Surely it is tweeter output that most influences both our directional and timbral notions and thus the realism of our stereo imagery. Also some loudspeakers do not have their tweeters crossing over until 5kHz, or even circa 12kHz with the very expensive Townshend super-tweeter.
Is it not the extended phase coherence, or phase correction due to supertweeters, that so improves overall realism?
Since the Sixties I have found it most useful to listen to the 'quality' of tweeter output by absorbing/reflecting away the output from other drivers in a system.
Also regarding there being little to recommend thd, will someone please explain the relevance of thd when a 1% thd valve amplifier can sound better than a 0.001% thd solid state chassis.
Valve amps always used to provide the cleaner tweeter output, especially when running loudly, which again meant that they could be comfortably driven to greater output when driving electromagnetic loudspeakers. Generally, valve amplifiers do not at high audio frequencies and at high voltage amplitudes IM the same way as quickly slewing transistors do when attempting to drive transiently modified load currents.
JCX, I'd love to see some of your distortion spectra of output potential with say two/three cycle toneburst and virtual loudspeaker loading, possibly using the same investigating waveform as recently illustrated by Wimms in his Distortion Microscope thread.
Cheers ............. Graham.
Yes your description - 'wall' of 'spectral contamination' - hits home, and it is only by our listening based investigations that we have made genuine progress.
That is; we have managed to get away from accepting designer's statements of measurement proven technical perfection.
With the likes of reproduced piano maybe it could be a loss of 'spectral contamination' due its single point of recording that has the major influence. Does the same detectability apply with headphone listening ?
Also, tube amps *do* go cleanly louder for a given sine/resistor rating due to their increased ability to voltage (anode headroom) follow when loudspeaker impedance increases, also due to their low grid charge storing characteristics with good/new space charging tubes during half cycles when load impedance falls and the anode is not overheated.
I would support Francis and JCX in relation to solid state IM distortion. (JCX, your words are not wasted, horses - water etc.)
It was the same with radio receivers - transistor communications receivers were incompetent beside valve types but the designers would not admit it! The solid state types could be good only when sharply tuning pre-selectors were used to filter out *other frequencies*, this until all signal voltages were held to minimum levels wrt circuit noise and using specialised filters with nested agc loops.
Hi Wimms.
Yes, Dolby and more recent algorithm exponents have much to answer for, especially Philips/Sony for our CD medium, but we do appear to agree that so many loudspeaker driving amplifiers still make the problems worse because of dynamically induced current flows through the cable linked amplifier-loudspeaker interface affecting both imaging and timbre perception.
The same dynamically induced current flows do not arise with full range electrostatics/Magneplanars etc.
Hi Francis.
I agree that 5kHz is a high pitch, but timebase whistle can still be aurally nulled by head movement as long as it is not fully finger attenuated.
Actually, I often listen for amplifier/loudspeaker reproduction errors by using my fingers in my ears as variable attenuators, this to get close to a tweeter or midrange without total aural masking (deafening), and/or to examine room reflections.
Surely it is tweeter output that most influences both our directional and timbral notions and thus the realism of our stereo imagery. Also some loudspeakers do not have their tweeters crossing over until 5kHz, or even circa 12kHz with the very expensive Townshend super-tweeter.
Is it not the extended phase coherence, or phase correction due to supertweeters, that so improves overall realism?
Since the Sixties I have found it most useful to listen to the 'quality' of tweeter output by absorbing/reflecting away the output from other drivers in a system.
Also regarding there being little to recommend thd, will someone please explain the relevance of thd when a 1% thd valve amplifier can sound better than a 0.001% thd solid state chassis.
Valve amps always used to provide the cleaner tweeter output, especially when running loudly, which again meant that they could be comfortably driven to greater output when driving electromagnetic loudspeakers. Generally, valve amplifiers do not at high audio frequencies and at high voltage amplitudes IM the same way as quickly slewing transistors do when attempting to drive transiently modified load currents.
JCX, I'd love to see some of your distortion spectra of output potential with say two/three cycle toneburst and virtual loudspeaker loading, possibly using the same investigating waveform as recently illustrated by Wimms in his Distortion Microscope thread.
Cheers ............. Graham.
Just read Cheever's thesis. Quite an interesting read. He essentially goes one step beyond GedLee, and makes a rather interesting assertion about the nature of harmonic distortion perception.
Whereas GedLee posit that the masking of the harmonics is useful and allows us to place greater importance upon those that are louder than the masking envelope, Cheever says something far more restrictive.
He says that the harmonics must adhere to the masking envelope. This is significant departure. Indeed I would have to say the hypothesis is far from proven, but is none-the-less very interesting.
It does of course essentially totally constrain the topology of an amplifier. Either you must be able to show that no distortion product of any kind is ever generated at a perceivable level, or your distortion products must fit the masking profile. His TAD metric, unlike the GedLee metric (which simply scales down the harmonic relative to its location in the masking profile), measures the deviation from the masking profile. Low is as bad as high.
Cheever's hypothesis is quite intriguing. It fits with some of the template based models of harmonic perception. But, what it needs is some tests.
This brings me to something I have been thinking about for the last couple of days. I rather think it would be fun, and highly scientifically valuable to propose a standard aural test environment. The observation is that pretty well everyone has a reasonable quality digital sound capability on their computer. That coupled with a standard pair of headphones (and it seems that the common ones used in aural tests are the Etymotic 4S) plus the use of some of the common tools on our computers would allow us to generate synthetic distortion functions, apply them to know sounds, and actually test out ideas. Much as the GedLee work is presented on their web page, but with the full range of tests, and possibly with a common double blind or ABX driver program. Any interest? If so we probably need another thread.
Whereas GedLee posit that the masking of the harmonics is useful and allows us to place greater importance upon those that are louder than the masking envelope, Cheever says something far more restrictive.
He says that the harmonics must adhere to the masking envelope. This is significant departure. Indeed I would have to say the hypothesis is far from proven, but is none-the-less very interesting.
It does of course essentially totally constrain the topology of an amplifier. Either you must be able to show that no distortion product of any kind is ever generated at a perceivable level, or your distortion products must fit the masking profile. His TAD metric, unlike the GedLee metric (which simply scales down the harmonic relative to its location in the masking profile), measures the deviation from the masking profile. Low is as bad as high.
Cheever's hypothesis is quite intriguing. It fits with some of the template based models of harmonic perception. But, what it needs is some tests.
This brings me to something I have been thinking about for the last couple of days. I rather think it would be fun, and highly scientifically valuable to propose a standard aural test environment. The observation is that pretty well everyone has a reasonable quality digital sound capability on their computer. That coupled with a standard pair of headphones (and it seems that the common ones used in aural tests are the Etymotic 4S) plus the use of some of the common tools on our computers would allow us to generate synthetic distortion functions, apply them to know sounds, and actually test out ideas. Much as the GedLee work is presented on their web page, but with the full range of tests, and possibly with a common double blind or ABX driver program. Any interest? If so we probably need another thread.
Here is a two tone multi-tone at 177.8279, 316.2278Hz same
level and frequncy as the 5 tone for comparison purposes.
I posted this to show how much less IM is produced with
2 tones -vs- 5 tones. Of course YMMV because it also depends
on the particular circuits nonlinearity.
Hi Francis,
Some thoughts on Cheever:
1) He doesn't allow any tolerance for distortion as the overall
pattern approaches zero. Instead he argues that the distortion
pattern must adhere to the overall pattern. I think this is not
correct as the overall pattern approaches zero. More leeway
would open up IMHO as the overall pattern reduces to 0.
2) His figure 2-2 is the main diagram and the one he derives
the foundation of his thesis from. It is not in the book referenced
in his paper:
Olsen, Harry F. “Music, Physics and Engineering”
Dover Publications, Inc. N.Y., 2nd ed. : 1967
I contacted Dan awhile back to see if he could provide a proper
reference for this figure. He was unable to provide it and actually
was evasive about it.
So I think his thesis is interesting, but would need some real
work to flesh it out.
3) At this point I think that IM distortion produced by multitones
is the dominant causitive agent of unpleasant distortion.
Combining work on masking seems also to be a path forward.
Mike
level and frequncy as the 5 tone for comparison purposes.
I posted this to show how much less IM is produced with
2 tones -vs- 5 tones. Of course YMMV because it also depends
on the particular circuits nonlinearity.
Hi Francis,
Some thoughts on Cheever:
1) He doesn't allow any tolerance for distortion as the overall
pattern approaches zero. Instead he argues that the distortion
pattern must adhere to the overall pattern. I think this is not
correct as the overall pattern approaches zero. More leeway
would open up IMHO as the overall pattern reduces to 0.
2) His figure 2-2 is the main diagram and the one he derives
the foundation of his thesis from. It is not in the book referenced
in his paper:
Olsen, Harry F. “Music, Physics and Engineering”
Dover Publications, Inc. N.Y., 2nd ed. : 1967
I contacted Dan awhile back to see if he could provide a proper
reference for this figure. He was unable to provide it and actually
was evasive about it.
So I think his thesis is interesting, but would need some real
work to flesh it out.
3) At this point I think that IM distortion produced by multitones
is the dominant causitive agent of unpleasant distortion.
Combining work on masking seems also to be a path forward.
Mike
Attachments
Hmmmm.....
1. He does allow the harmonics to fall into the noise floor, but does not provide much other leeway. The manner they fall under the noise is probably reasonable.
His argument on IM, however, is curiously floppy. He claims IM arises from the same mechanism as HD, and can be covered by the same metric. Then goes on to assert that IM does not occur in non-feedback amplifiers. This line of reasoning should have his non-feedback amplifier scoring very poorly on TAD, since the argument would have it that although it has HD, a lack of IM would fail to match the ear's natural IM distortion, and thus a mismatch with the brains filtering would occur.
Once past the discussion on HD, the arguments seems to get less robust, and in places get a little strident.
2. Bad. Real bad. I will need to wait a couple days to get a copy of Olson's book. (Gosh! a real physical book!) But that should never happen.
On the other hand, even with a graph of dubious provenance, we can keep the remainder of the argument. The argument is essentially that harmonic distortion does occur in the ear, and that the brain is capable of reversing it. The exact shape of the curves can be subject to further study.
I will admit to being quite bemused by the testing of a feedback amplifier open loop. All he proves is that the open loop performance is poor. He does not convolve it with the earlier derivations of HD content from feedback, or in any other way use the results. Other to say that should he have calculated the TAD it would have been bad. (Surprise surprise.)
It is a funny thesis. It reads well mostly, but is somehow not fully developed. One is left wishing some of the threads had been more fully developed and a wider range of tests allowing for some statistically valid conclusions to be drawn. Then again, his wife was probably about to remove parts of his anatomy if he didn't finish up. Such is the life of a grad student.
1. He does allow the harmonics to fall into the noise floor, but does not provide much other leeway. The manner they fall under the noise is probably reasonable.
His argument on IM, however, is curiously floppy. He claims IM arises from the same mechanism as HD, and can be covered by the same metric. Then goes on to assert that IM does not occur in non-feedback amplifiers. This line of reasoning should have his non-feedback amplifier scoring very poorly on TAD, since the argument would have it that although it has HD, a lack of IM would fail to match the ear's natural IM distortion, and thus a mismatch with the brains filtering would occur.
Once past the discussion on HD, the arguments seems to get less robust, and in places get a little strident.
2. Bad. Real bad. I will need to wait a couple days to get a copy of Olson's book. (Gosh! a real physical book!) But that should never happen.
On the other hand, even with a graph of dubious provenance, we can keep the remainder of the argument. The argument is essentially that harmonic distortion does occur in the ear, and that the brain is capable of reversing it. The exact shape of the curves can be subject to further study.
I will admit to being quite bemused by the testing of a feedback amplifier open loop. All he proves is that the open loop performance is poor. He does not convolve it with the earlier derivations of HD content from feedback, or in any other way use the results. Other to say that should he have calculated the TAD it would have been bad. (Surprise surprise.)
It is a funny thesis. It reads well mostly, but is somehow not fully developed. One is left wishing some of the threads had been more fully developed and a wider range of tests allowing for some statistically valid conclusions to be drawn. Then again, his wife was probably about to remove parts of his anatomy if he didn't finish up. Such is the life of a grad student.
Worth reading.
http://www.its.caltech.edu/~musiclab/feedback-paper-acrobat.pdf
A rather nicely done paper. They are pretty careful with their conclusions too. A model of how to do it. The authors are an interesting pairing. This is the sort of stuff you can do when you are eminent in your own field
http://www.its.caltech.edu/~musiclab/feedback-paper-acrobat.pdf
A rather nicely done paper. They are pretty careful with their conclusions too. A model of how to do it. The authors are an interesting pairing. This is the sort of stuff you can do when you are eminent in your own field
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.