Distortions' dynamics and human perceptions-what to measure?

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
[quote author="mikep"]Distortion is any deviation from perfect reproduction. It can be time, amplitude, phase, waveshape (harmonics), and other things we haven't yet quantified. saying that distortion is a transfer function makes it sound like you think you figured out all the terms and are just searching for the coefficients. might lead to better understanding of the low-order effects, but if you are trying to judge the overall performance this approach is fundamentally flawed.

there is one uncommon measurement that takes *everything* into account, even the stuff we don't yet know we are looking for. The phase cancellation test. you take your output signal and mix it with an inverted copy of the input signal. the level must be set precisely (tune by hand) and the summing amp should be "free of distortion". I think Mr Hafler wrote about this technique for testing power amps. it works, you can listen to the residual or look at it on a scope. no notch flter is needed. Why is this test not used more? Why am I not using it? :roll:

I am guilty of using a traditional distortion analyzer. But I think there is alot more info in there than the meter/scope is telling you. just listen to the residual. you ear can do a pretty good job of ignoring the fundamental bleed-through, the analyzer's own distortiion AND noise. I can usualy hear some low level, higher order products that are down in the noise that don't show up on a scope. you listen to the looped-back residual for a few seconds and then switch to the input, even when the meter is at its measurement limits you can often hear a change. For what its worth, the frequencies I use are 400Hz and 200Hz. this way the harmonics are in a range where they are easy to pick out. I guess there is a bit of a learning curve to this method but for anyone with musical background or lots of critical listening experience should try it.

mike p[/quote]

I too am a fan of null testing. IIRC it was popular in DIY audiophile circles decades ago. I have used it before to convince people of similarity between properly working products... People are often influenced by brand names and expectations.

The good thing about null testing is it is a kitchen sink kind of test revealing the entire difference. One shortcoming is it doesn't reveal which DUT is causing the error in product to product comparisons.

I find the common practice of THD analyzers to ignore phase shift a little misleading. The common error term in a NF amplifier is a 90' phase shifted (leading) version of the output (but referenced to the input). This sinusoidal error term at the fundamental frequency is completely ignored by THD test sets. That same error term for a non-sinusoidal stimulus will be a far from harmless phase shifted version of the output. Instead it represents the derivative, or a rate of change of the output waveform (i.e. for a triangle wave the error is a square wave, for a square wave the error is two spikes).

Years ago when I was still trying to design for audio-fools I settled on a two-tone IMD test. I used 19 and 20kHz 1:1 as test tones, under the premise that they were difficult but still within the valid audio band. Further the distortion product at 1kHz was smack in then middle of human hearing so clearly audible. I found this especially applicable to phono preamp design where the RIAA playback Eq would boost that IM product another 20 dB relative to the 19-20kHz stimulous unlike simple THD products where higher harmonics were simply rolled off. In other words THD testing understated nonlinearity while my variant two-tone IMD overstated nonlinearity by 20 dB.

I found this IMD test pretty revealing of how designs could perform with real world audio. I never had much use for the alphabet soup of TIM, SID, whatever. Slew related distortions just don't happen in adequately fast, properly band passed circuitry. YMMV.


JR
 
[quote author="JohnRoberts"]
I found this IMD test pretty revealing of how designs could perform with real world audio. I never had much use for the alphabet soup of TIM, SID, whatever. Slew related distortions just don't happen in adequately fast, properly band passed circuitry. YMMV.
[/quote]

If the measurement costs $1, interpretation may cost $999,999, right?
:grin:

How level of distortions depend on the signal level? I believe, the amp with high level of IMD on a big level may be favoured by audiofiles. I don't mean audiofools that hear "coloration" and think it is the sound, I mean real audiofiles who prefer when level of harmonics and their order decreases with fading out sounds. Anothed type of audiofiles who listens to a modern music would prefer the amp with low level of IMD on high signal levels despite it distorts fading out sounds, such as fine percussions, guitar and piano strings, reverberation in great concert halls... Modern music don't have such neuances, everything was passed through gates and compressors, so opamps and rap music love each other.

The question is, is it possible to build the track from microphone to speakers that will satisfy both camps, i.e. when both a loud thunder and a light brease sound equally real? :razz: The system that will be equally good on concerts, for studio monitoring, and for home video theater? :razz:
 
thanks Mike
and
thanks John

I could go through point by point but it's not worth it now
I think both of you have brought up excellent points and it is from here that the discussion could start

a quick google on distortion and you get the wiki
http://en.wikipedia.org/wiki/Distortion

and there is some simple info and from here some definitions etc can be set so a further discussion could start

from the wiki page you can find
http://www.mindspring.com/~j.blackstone/dist101.htm
it has a couple of pictures that we have looked at before

F+2nd.gif

this is a representation of a simple harmonic where it is phase sync'd
for situations where the harmonic is randomly phased the current crop of test equipment will returnn the same results
... usually based on power only

both Mike and John have mentioned phase and that current methods of quantifying distortion have no regard for phase and as mike suggests
" ... things we haven't yet quantified. "

Just because someone uses the words Transfer Function or
FFT - Fast Fourier Transform, does make it any more accurate or reliable.
FFT implies the use of a specific algorithm

I'm guessing here but I think most people would only have DFT equipment ... Digital Fourier Transform
and so
FFT - Fast Fourier Transform ... is the use of the specific algorithm in the context of a Digital Fourier Transform

DFT requires skill and knowledge to use the mathematics
and the collection system
and the sampling rate to interpret the results


:cool:
Did you ask him?
no
I wasn't conducting the interview
the interviewer's name is Lynn Fuston

:wink:
 
[quote author="Kev"]
I think both of you have brought up excellent points and it is from here that the discussion could start
[/quote]

Really? :grin:
I thought it started from the beginning of the topic you are trying to steal. :cool:

Kev, do you have something to add to the discussion except a difinition of distortions that don't require mathematical skills? :roll:
 
Distortions' dynamics and human perceptions-what to measure?

without adequate definitions how can you measure anything ?

... the topic you are trying to steal.

Kev, do you have something to add to the discussion except a difinition of distortions that don't require mathematical skills?

oh yes I do and it was to be a great deal indeed
but stealing the thread as you have suggested is not my intention nor was it my intention in the SSL Dist thread and I urged you a number of times to start a new specific thread.

I though this one was it.

But
I am happy to leave the thread for good.

Just before I do I need to clarify something

DFT
I used the modern refernce Digital but in the distant past it was used as Discrete
in the context of Continuous/Discrete Fourier Transform
:cool:


:roll:
ohh ?
I can't just yet
Ok, let's return back to the topic. Vocal, acoustic instruments, sound better with a single ended tube class A amp than with a modern AB class one designed like a typical operational amplifier. Why?
Contrary, modern music sounds better through a transistor AB class opamp, and horrible through a single ended tube class A amp.
Why?

I don't think that generalisation fits at all and I don't think those situations are better.



Kev has left the thread
:cool:
thank you very much
 
[quote author="Kev"]Distortions' dynamics and human perceptions-what to measure?

without adequate definitions how can you measure anything ?

[/quote]

Sure. I can measure "How many that things containing that thing fit in that thing" and be satisfied: all bottless of beer I bought are in the cooler. :grin:

Ok, let's return back to the topic. Vocal, acoustic instruments, sound better with a single ended tube class A amp than with a modern AB class one designed like a typical operational amplifier. Why?
Contrary, modern music sounds better through a transistor AB class opamp, and horrible through a single ended tube class A amp.
Why?

I don't think that generalisation fits at all and I don't think those situations are better.

It may mean also that you don't have enough of subjective experience for such generalizations, that's why distortions and deletions prevail. :grin:


DELETION

Deletion occurs when we selectively pay attention to certain aspects of your experience and not others. Deletion means we overlook or omit certain sensory information. Without deletion, we would be faced with much too much information to handle in our conscious minds.



DISTORTION

Distortion occurs when we make shifts in our experience of sensory data by making misrepresentations of reality. There's a well known story of distortion in Easter philosophy. It is called the story of the rope and the snake. A man walking along a road saw what he believed to be a snake and yelled, "SNAKE". However, upon closer investigation he is relieved to discover that it really was only a piece of rope.

Distortion also helps us in the process of motivating ourselves. Motivation occurs when we actually distort the material that has come to us and that has already been changed by one of our filtering systems. Distortion is also helpful in planning. We distort to plan when we construct imaginary futures.



GENERALIZATION

The third process is generalization, where we draw global conclusions based on one, two or more experiences. At its best, generalization is one of the ways that we learn, whereby we take the information we have and draw broad conclusions about the world based on one or more experiences. at its worst, generalization is how we take a single event and make it into a lifetime of experience.

Normally the conscious mind can only handle seven, plus or minus two, items of information at any given time. Of course, many people cannot even handle this number. Try this: Can you name more than seven products in a given product category: for example, cigarettes? Most people will be able to name two, maybe three products in a category of low interest, and usually no more than nine in a category of high interest. There is a reason for this. If we did not actively delete information all the time, we would end up with excessive amounts of information coming in. In fact, you may have even heard that psychologist say that if we were simultaneously aware of all of the sensory information that was coming in, we would go crazy. That is why we filter the information.

So the question is, "When two people have the same stimulus, why don't they have the same response?" the answer is: because we delete, distort, and generalize the information from the outside in different ways.

We delete, distort and generalize the information that comes in from our sensed by using certain internal processing filters. The filters are: Meta Programs, Values, Beliefs, Attitudes, Decisions and Memories.


Taken from here: http://www.execonn.com/matt/Docs/NLP.htm
 
[quote author="Wavebourn"][quote author="JohnRoberts"]
I found this IMD test pretty revealing of how designs could perform with real world audio. I never had much use for the alphabet soup of TIM, SID, whatever. Slew related distortions just don't happen in adequately fast, properly band passed circuitry. YMMV.
[/quote]

If the measurement costs $1, interpretation may cost $999,999, right?
:grin:

How level of distortions depend on the signal level? I believe, the amp with high level of IMD on a big level may be favoured by audiofiles. I don't mean audiofools that hear "coloration" and think it is the sound, I mean real audiofiles who prefer when level of harmonics and their order decreases with fading out sounds. Anothed type of audiofiles who listens to a modern music would prefer the amp with low level of IMD on high signal levels despite it distorts fading out sounds, such as fine percussions, guitar and piano strings, reverberation in great concert halls... Modern music don't have such neuances, everything was passed through gates and compressors, so opamps and rap music love each other.

The question is, is it possible to build the track from microphone to speakers that will satisfy both camps? :razz:[/quote]

The rationale for complexity in tests is to make interpretation more straightforward (less is better). The simpler the test the more experience and skill plays into evaluation with the opportunity for subjective bias (to wit interpreting square wave response). IMD commonly sounds like LF mud (FM tuned off station frequency, etc.). I am not aware of any euphonious benefit of IMD. With most solid state designs I found a pretty close mathematical correlation between THD and IMD. THD at 20 kHz was often 10 DB higher than SMPTE IMD (which correlates to ratio between SMPTE's 7 kHz stressor vs. THD @ 20 kHz). SMPTE testing is archaic and obsolete for modern gear as it was designed to detect problems in movies projectors. Some smarter folks than me have proposed more than two tones for IMD testing but that's a little beyond this discussion.

It is difficult to reduce all distortion mechanisms into simple generalities. For some like the noise component in THD+N it only appears to rise in falling signal but that is an artifact of the measurement while crossover or digital LSB quantization will indeed rise. Alternately there are some distortion mechanisms in power amps that increase as signal level rises, some due to increased current demand but others due to being less volts away from clipping (output stage device and driver characteristics). Another confounder is nonlinearities in tube circuits that not only have level dependance but can have time constants such as associated with recovery from overloads due to internal charge redistribution. The sonic signature of tube overload is well studied and loved by many guitarists.

Distortion is a very complex, multi-variant mechanism that is IMO impossible to simply characterize. For reproduction (IMO) less is always better. For making music a nasty diode clipper in the hands of the right guitarists can make good sounds, just not in my reference path please. Making music is different from capturing an acoustic event. If you choose to then enhance that captured acoustic reality to make it subjectively sound more musical, that'd OK too just let's call a spade a spade. That subjective modification is an effect and IMO needs to be in a suitably labeled black box..not in a reference path. I find it difficult to expect that every sonic event will benefit from the exact same coloration/modification. I am not opposed to psycho acoustic tricks, stereo reproduction is based on suspending perception of the actual (two sources) to listen through that reality and hear a larger performance in a complex sound space. Sudry tricks improve that experience. Lets not confuse that with the need for accurate signal blocks in reference paths.

JR
 
[quote author="JohnRoberts"]Lets not confuse that with the need for accurate signal blocks in reference paths.
[/quote]

It is exactly what I mean from the beginning. If transfer function is smooth and represents the 2'nd order mostly, on high level of signal it will give more IMD than straight line sharply (high order) broken near the zero. But it does not mean that the sound modified by the 2'nd one will be heard as more clean than modified by the 1'st one if it is the sound that represents acoustical instruments, but it will sound more "clean" with the matherial with narrow DR. The 1'st case is the case of single ended amplification by vacuum tubes, the 2'nd one is the case of amplification by opamps.
 
[quote author="Wavebourn"][quote author="JohnRoberts"]Lets not confuse that with the need for accurate signal blocks in reference paths.
[/quote]

It is exactly what I mean from the beginning. If transfer function is smooth and represents the 2'nd order mostly, on high level of signal it will give more IMD than straight line sharply (high order) broken near the zero. But it does not mean that the sound modified by the 2'nd one will be heard as more clean than modified by the 1'st one if it is the sound that represents acoustical instruments, but it will sound more "clean" with the matherial with narrow DR. The 1'st case is the case of single ended amplification by vacuum tubes, the 2'nd one is the case of amplification by opamps.[/quote]

Sorry, I don't follow your logic. You seem to be debating the relative merits of different types of distortion. I dislike them all, and don't see it as an either or choice. How about arbitrarily small amounts of whatever remaining non linearity you can't completely squash?

JR
 
What happens after the info enters the ear?

How are you going to measure that?

From person to person?

Are you going to implant a distortion analyzer inside someones brain?

We probably hear stuff differently by a much larger percentage than distortion specs cover.
 
[quote author="CJ"]What happens after the info enters the ear?

How are you going to measure that?

From person to person?

Are you going to implant a distortion analyzer inside someones brain?

We probably hear stuff differently by a much larger percentage than distortion specs cover.[/quote]

The link I posted the other day to Diana Deutsch's website may offer some insight into that. The study of human perception works to clarify how much we hear alike and how much our perceptions vary. More importantly how our perception can be confused or tricked.

I don't wish to suggest the path from the musician's mind to the listener's mind is a simple straight line. There is a certain amount of "magic" in the front end music making process, and the final interface between stored music and simulated recreation of a sound event involves a certain amount of subterfuge. The vast middle can and needs to be linear.

JR
 
I see this topic becomong more complicated with todays surround sound and different systems that you see at AES, etc.

It's hard to quantify sound, you liaten to two different stereo systems with the same specs and they sound totaly different.
 
[quote author="CJ"]What happens after the info enters the ear?

How are you going to measure that?

From person to person?

Are you going to implant a distortion analyzer inside someones brain?

We probably hear stuff differently by a much larger percentage than distortion specs cover.[/quote]

I paid attention to many opinions of people about sound quality, with presupposition that all of them have a point and nobody is "audiofool", neither people with huge opamps, nor with single ended glass chimneys. And found that indeed there are many information about their preferences that may be justified both from psychological and technical points of view. Also, I've experimented with synthesizers, amplifiers, speakers, microphones: in laboratory, on stage, at home... And realized for myself what kind of parameters are more critical for sound reproduction.
 
[quote author="JohnRoberts"][quote author="Wavebourn"][quote author="JohnRoberts"]Lets not confuse that with the need for accurate signal blocks in reference paths.
[/quote]

It is exactly what I mean from the beginning. If transfer function is smooth and represents the 2'nd order mostly, on high level of signal it will give more IMD than straight line sharply (high order) broken near the zero. But it does not mean that the sound modified by the 2'nd one will be heard as more clean than modified by the 1'st one if it is the sound that represents acoustical instruments, but it will sound more "clean" with the matherial with narrow DR. The 1'st case is the case of single ended amplification by vacuum tubes, the 2'nd one is the case of amplification by opamps.[/quote]

Sorry, I don't follow your logic. You seem to be debating the relative merits of different types of distortion. I dislike them all, and don't see it as an either or choice. How about arbitrarily small amounts of whatever remaining non linearity you can't completely squash?

JR[/quote]

I was debating two major approaches of two camps. Speaking of arbitrary small amounts of whatever, I've found that it is possible, but trade in is a heat generation. My answer in terms of amplification is very fast single ended class A amplifier with deep feedback. The problem is, I was taught how to develop radio and electronics equipment, how to organize and run manufacturing on existing plants, but nobody taught me how to start from zero and build own business... I have some good prototypes that I believe would satisfy audiofiles from both objectivist and subjectivist camps...
 
[quote author="Wavebourn"]


I was debating two major approaches of two camps. Speaking of arbitrary small amounts of whatever, I've found that it is possible, but trade in is a heat generation. My answer in terms of amplification is very fast single ended class A amplifier with deep feedback. The problem is, I was taught how to develop radio and electronics equipment, how to organize and run manufacturing on existing plants, but nobody taught me how to start from zero and build own business... I have some good prototypes that I believe would satisfy audiofiles from both objectivist and subjectivist camps...[/quote]
Since I ran (not walked) away from the audiophile market a couple of decades ago I can't give you any good advice on how to succeed there.

Perhaps stuff like using gold plated jacks and knobs? You will probably have no problem explaining your design philosophy to them. They love class A room heaters and such.. Is Krell still around?

Only downside from a market size perspective is that these days all those doctors and dentists who used to buy exotic hi-fis to impress their neighbors are now buying exotic home theater systems, so figure some way to make your class A amp 5 channel. :cool:

Good luck. Business is all about another reality. You can't grow your bank account with good theories, but the public loves a good story.

JR
 
I just read the thread title again and maybe now understand what anatoliy was getting at. the thing about birds threw me. we are used to hearing tones get more stressed sounding as they get louder in part because of the way the human voice changes its harmonic structure when you go from soft singing to yelling. brass instruments too, harmonics increase with level. there are many other examples in nature. this doesn't mean we want our circuitry to distort this way, but rather that we can tollerate a bit more of this type of distortion because it is familiar.

If you concede that no circuit can be absolutly perfect, then it makes sense to pick a topology with a type of nonlinearity that is least problematic. in high-feedback circuits distortion is very low right up to clipping, with low feedback the onset is much more gradual. presumably we could construct a circuit with a knee matching the "natural profile", while still minimizing distortion overall. there is probably more to it than finding some magic amount of feedback.

the thread subject mentions measurment. I think you need to measure the THD vs level, of natural sounds: human voice, sax, trumpet, etc. perhaps autotune would get the frequency stable enough to use a notch filter to remove the fundamental.
 
Mike, you got my point exactly! We hear as distortions what we don't expect to hear!

But unfortunately, amount of feedback sharpens a transfer function increasing it's order, so order of generated harmonics, not just their level. May be some kind of approximation and feed-forward error compensation would be the way to go (and I used it loading an output source follower on a voltage to current converter, also used "Tango" pair of degraded BJT follower up to 3 amps and a FET follower for the rest), but anyway it would be too complex to follow "mechanical rules" strictly, and I don't think it is profitable, that's why I decided to go with a deep feedback to straighten a transfer function as possible, and in order to eliminate possibility of transcients I used a diode limiter in input that also drives an opto compressor and a signal LED so as soon as clipping happens it happened on input diodes once, then the gain goes down below the clipping level, and a banner "HEY DUDE TOO LOUD TURN THE KNOB" will be seen.

Indeed, I was tempted to measure dependence of levels of harmonics of most common mechanical sounding and reflecting systems, but gave up this idea, the understanding of the main rule was enough. May be I am wrong and measurements would help a lot... Do you have a sponsor for such a scientific project? I'd participate in it with a pleasure! :grin:
 
Highly offtopic:

Funny thing is, I just noticed that John Roberts sells one gizmo for 400
bucks, that would probably make (if it works well) more diference
(in drum recording) than $20000 spent in signal path electronics

cheerz
urosh
 
[quote author="recnsci"]Highly offtopic:

Funny thing is, I just noticed that John Roberts sells one gizmo for 400
bucks, that would probably make (if it works well) more diference
(in drum recording) than $20000 spent in signal path electronics

cheerz
urosh[/quote]


Indeed, a well tuned kit is a thing of beauty and not that common. Calling it "tuning" ignores the harder part, making all the lugs the same, no matter what note they're tuned to.

These are the early days for precision drum tuning and there will surely be much debate about the value of tuning a percussion instrument to specific notes but the improvement in a well "cleared" head (lugs matched) is obvious to even non drummers.

Sorry to hijack the thread... I find the physics or mechanics of this more interesting than the electronics involved. I've patented the basic invention so I can talk freely about how it works with any who are interested, but it's clearly OT here.

JR
 
Back
Top