Folks, before this thread drifts into the past, I would like to point out that most of you have no idea how the ear hears phase, and when humans are sensitive to it. If an op amp generated a changing phase with every different frequency and amplitude, even if it is small in amount, I suspect that it could be detected by the human ear.
john curl said:
John, do I detect some regret there? I'm glad you decided to stick it out. And as Chairman Barney (ask SY) once said, "you gotta start young if you're gonna stick it out"
. One thing that's been mentioned by MikeB is the possibility of programming WAV files with certain types of phase distortion. I'd be interested in hearing more about this from MikeB, especially any info, links etc he might have about WAV file programming. Since that type of file is uncompressed and lossless, it raises some really interesting possibilities of audibility testing.What's especially interesting, to me at least, is the possibility of programming a distortion mechanism into the file (that is, some kind of mathematical simulation of the circuit problem that causes the phase distortion in the first place) rather than just some mathematically produced distortion. The Czerwinski article that jcx mentioned earlier has an interesting quote about distortion simulations, something to the effect that "it's the rock thrown at your head that causes the damage, not the velocity of the rock. Although the velocity can be measured." In that qoute he was criticising some published articles that had audibility testing of mathematically produced distortion where that distortion didn't have the properties of "real" distortion, namely its variation with level and other properties of the physical nonlinearity that caused it. So if one could model the nonlinearity itself that's the root cause, (dynamic nonlinearity or otherwise), create some DSP model of the nonlinearity, run some music through it, and see if the difference could be detected (under suitable handcuffed, err, sorry, blind test conditions) then that would be cool.
Andy C, I doubt that you folks can generate anything with enough initial audio quality that you could detect small differences in phase, double blind or otherwise. Mostly, because most likely, the signal would pass through a number of marginal op amps, BEFORE listening. This would mask any potential differences.
Not at all, just realistic. Most of you apparently think that computer generated effects are OK. What op amps are passed through before listening can be attempted?
To me, it is like fine wine tasting using unwashed beer mugs. It is virtually useless.
This does not attack people's enjoyment of music, any more than drinking wine from dirty glasses keeps people from getting drunk. It is just that 'fine' comparisons require extra ordinary test conditions.
The bottom line is that high end audio has minimized the use of op amps, because we can hear what they do to the music. To do this, we have live music, or at least class A discrete designs, either tube or solid state, of the highest quality to compare to. So far, almost all op amps have come out second best.
To me, it is like fine wine tasting using unwashed beer mugs. It is virtually useless.
This does not attack people's enjoyment of music, any more than drinking wine from dirty glasses keeps people from getting drunk. It is just that 'fine' comparisons require extra ordinary test conditions.
The bottom line is that high end audio has minimized the use of op amps, because we can hear what they do to the music. To do this, we have live music, or at least class A discrete designs, either tube or solid state, of the highest quality to compare to. So far, almost all op amps have come out second best.
I don't get the point...
If i modify a wavefile mathematically, put it out through spdif or burn
it onto CD, where are all these opamps ?
Of course i exaggerate the effects, to be sure that distortions are
from the algorithm.
Using algorithms helped me a lot to understand distortionmechanisms,
or do you know a better way to apply specific distortionmechanism
to music (not sinewaves) and look at the effects via FFT or EAR ?
Using wavefiles on the computer gives a unique possibility to do
research with musicsignals, signals an amp should be designed for...
Also you can investigate which kind of unlinearity produces which
artefacts. I have a small program generating waves, apply unlinearity
to them and finally fft the result. So i am able to investigate IMD with
different types of distortion. (First results scared me, didn't know that
IMD is that evil...,may be IMD should be renamed to THD-explosion)
When i have more infos about this, i will post them...
For nonpro-hobbyists like me this is extremely helpful !
BTW, my measured dynamic groupdelay was introduced by inputcouplingcap...
MikeB
If i modify a wavefile mathematically, put it out through spdif or burn
it onto CD, where are all these opamps ?
Of course i exaggerate the effects, to be sure that distortions are
from the algorithm.
Using algorithms helped me a lot to understand distortionmechanisms,
or do you know a better way to apply specific distortionmechanism
to music (not sinewaves) and look at the effects via FFT or EAR ?
Using wavefiles on the computer gives a unique possibility to do
research with musicsignals, signals an amp should be designed for...
Also you can investigate which kind of unlinearity produces which
artefacts. I have a small program generating waves, apply unlinearity
to them and finally fft the result. So i am able to investigate IMD with
different types of distortion. (First results scared me, didn't know that
IMD is that evil...,may be IMD should be renamed to THD-explosion)
When i have more infos about this, i will post them...
For nonpro-hobbyists like me this is extremely helpful !
BTW, my measured dynamic groupdelay was introduced by inputcouplingcap...
MikeB
john curl said:
When you look at the antomy and physiology of the inner ear and how it actually operates as a transducer it appears to work more like a spectrum analiser than an oscilloscope. It detects the amplitudes at continuous frequencies. It is sensetive to changes in the time of arrival of sounds at each frequency which may be caused by a significant phase shift. In my opinion a frequency dependant time shift due to a significant phase shift may convice the brain that a particular sound is a reflected, diffracted or otherwise distorted sound and not a direct one.
Having said this I wouldn't think that a phase shift of less than 20 degrees at 20KHz would be picked up
http://privatewww.essex.ac.uk/~cplack/PDF_Papers/Plack_White_00b.pdf
http://privatewww.essex.ac.uk/~cplack/Pitch_Meeting/presentations/Watkinson_Plack.ppt
http://scitation.aip.org/getabs/ser...00108000004001826000001&idtype=cvips&gifs=yes
http://scitation.aip.org/getabs/ser...00113000005002801000001&idtype=cvips&gifs=yes
http://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf
http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=2808909&dopt=Abstract
http://www.mp3-tech.org/programmer/docs/non-linear_human_hearing.pdf
http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=8015725&dopt=Abstract
http://www.staff.ncl.ac.uk/t.d.griffiths/griffiths_clinsci_1999.pdf
Journal Aud Eng Soc , Vol 52, No 7/8 2004 July/August, Predicting the Perceived Quality of Nonlinearly Distorted Music and Speech Signals
The JAES article above responds to Czerwinski's criticisms. You might find it interesting.
John,
It's very easy these days to generate almost any sound mathematically, using numerous methods, from Matlab through to good wave editing programs.
These can be generated totally in the digital domain, and then burnt to CD via a PC, so no need for any op-amps to get in the way.
One does need to be careful though that the signals generated are legitimate ones; it's easy to generate and burn signals that should never appear on a CD, like perfect square waves since one isn't passing through any AA filters.
Andy.
It's very easy these days to generate almost any sound mathematically, using numerous methods, from Matlab through to good wave editing programs.
These can be generated totally in the digital domain, and then burnt to CD via a PC, so no need for any op-amps to get in the way.
One does need to be careful though that the signals generated are legitimate ones; it's easy to generate and burn signals that should never appear on a CD, like perfect square waves since one isn't passing through any AA filters.
Andy.
LTSpice can be easily used to generate "interesting" wav files. See attach. It generates 4 wavelets of 400Hz signal modulated by 100Hz so that there are 4 distinct wavelets with sinusoidal increase and decrease in amplitude, separated by 100Hz period (10ms) of silence.
The 4 wavelets are generated for each stereo channel, and saved out to 2 wav files at 48K sampling rate, 16-bit. One channel is delayed wrt to other by 1 sampling interval, approx 20us. Other wav file has channels swapped. Thus comparing the two there is 40us difference. The wav files can be replayed by your soundcard, and if you like, using http://www.foobar2000.com/ ABX'ed. Expect 100% confidence ABX'ing.
Such wavelet modulation was picked to minimise spectral content. Single wavelet in isolation contains only 3 tones: 400Hz and +/-100Hz sidebands as per AM. When whole wav is analysed, lots of spectrum is required to represent the signal. Try FFT on time ranges: 10ms-20ms, 0-80ms and compare with FFT of whole 100ms signal. Use no windowing or smoothing. Ponder what spectrum ear is hearing.
I made that "simulation" to compare audibility of phase delays vs delays of onset ("attack") of changing inphase signals. The latter delay can be changed separately from phase of the carrier. Interestingly, although onset delay can be quite large and signal waveforms very different, its not audible. Audibility of both types of delay depends on carrier frequency.
So its one example of generating sounds mathematically, by abusing spice. It can be more convenient than messing around with Matlab or wave editors. Besides, such signals could be used to feed simulated circuits.
LTSpice can also be used to "replay" wav files through your simulated circuit, so one could even play with simulations to estimate audible results.
The 4 wavelets are generated for each stereo channel, and saved out to 2 wav files at 48K sampling rate, 16-bit. One channel is delayed wrt to other by 1 sampling interval, approx 20us. Other wav file has channels swapped. Thus comparing the two there is 40us difference. The wav files can be replayed by your soundcard, and if you like, using http://www.foobar2000.com/ ABX'ed. Expect 100% confidence ABX'ing.
Such wavelet modulation was picked to minimise spectral content. Single wavelet in isolation contains only 3 tones: 400Hz and +/-100Hz sidebands as per AM. When whole wav is analysed, lots of spectrum is required to represent the signal. Try FFT on time ranges: 10ms-20ms, 0-80ms and compare with FFT of whole 100ms signal. Use no windowing or smoothing. Ponder what spectrum ear is hearing.
I made that "simulation" to compare audibility of phase delays vs delays of onset ("attack") of changing inphase signals. The latter delay can be changed separately from phase of the carrier. Interestingly, although onset delay can be quite large and signal waveforms very different, its not audible. Audibility of both types of delay depends on carrier frequency.
So its one example of generating sounds mathematically, by abusing spice. It can be more convenient than messing around with Matlab or wave editors. Besides, such signals could be used to feed simulated circuits.
LTSpice can also be used to "replay" wav files through your simulated circuit, so one could even play with simulations to estimate audible results.
Now this thread is getting somewhere 
MikeB, cudos to you for standing up and admitting such a simple mistake. Not everyone would do it...
You might find the article by Audio Precision interesting as a comparison to your experiments. Also, multitone IMD seem to have quite good correlation to perceived sound quality according to many recent articles. Remember though that any nonlinearity will produce both THD and IMD. They are just two different ways of looking at it.
/Magnus
MikeB, cudos to you for standing up and admitting such a simple mistake. Not everyone would do it...
You might find the article by Audio Precision interesting as a comparison to your experiments. Also, multitone IMD seem to have quite good correlation to perceived sound quality according to many recent articles. Remember though that any nonlinearity will produce both THD and IMD. They are just two different ways of looking at it.
/Magnus
Swedish Chef said:
Yes, that's what i've "found out" with my little program, with multitone
any unlinearity creates IMD. This imd is far above the thd, even with
high openloopbandwidth. I will check this evening with music, what
the ratio THD<->IMD is aproximately. As far as i understood, with low
openloopbandwidth the frequencies "mirrored up" by the imd cannot
be compensated by feedback, means openloopdistortion gets
completely audible at higher freqs but caused by lower freq.
In my program a multitone (3+4khz) distorted for 3rd harmonic
created big harmonics at 10khz and 11khz, for 2nd harmonic these
freqs are 1khz+7khz.
Is this correct ? If yes, it explains why amps with low ol-bandwidth
tend to sound bright.
Mike
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.