|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
206.255.200.146
In Reply to: RE: The "point" was in the subject header... posted by Steve O on November 12, 2015 at 18:44:00
If the converse is true, there is a correlation, a negative correlation, yes?
Yes.
Not necessarily THD alone since there are other measures of linearity and maybe even good sound.
To wit?
Follow Ups:
To wit!: IMD, SID, smoothly decending levels of high order distortion products (there's name for this condition but it escapes me at the moment), basic stability, rapid overload recovery and so on. Optimizing any one or two parameters is probably insufficient so it's really an exercise of judicious balance. Much of this stuff was known and effectively dealt with in the middle of the previous century by a few designers.
Geddes found that IMD also has no correlation (actually slightly negative but statistically insignificant at the 95% confidence).
An interesting paper by Geddes. Since you've cited Geddes on a couple of occasions here, I assume you're familiar with his paper.
While I'm not qualified to directly butt heads with Mr. Geddes, what he describes in his paper raises a number of questions.
1. The testing methodology employed a 15 second excerpt from ALW's "Phantom of the Opera" ripped from the redbook CD and converted to a .wav file. This was considered "the reference". Is 16/44.1 sufficiently resolving considering the potential subtleties involved? Is the music selection sufficiently revealing? Is a 15 sec sample of sufficient duration?
2. Twenty one different stimuli (distortions) were mathematically simulated (calculated). The exact nature of the distortions is unclear as is the correlation of them to real life distortion generating devices. The calculated distortions were then used to multiply the input reference wav file to produce the test sample or signal. All stimuli were digitally adapted and presented to the subjects via computer...as 16/44.1 wav files. Is a mathematically generated distortion a valid means of simulating the performance of a physical device? Has this methodology been validated by Geddes or independently by others?
3. Subjects rate the test signals as better or worse than the reference. Doesn't this predispose the results towards "preference" instead of "accuracy"?
4. Geddes finds a weak negative correlation between THD and IMD and preference but essentially dismisses this finding based on statistics and pursues it no further. He also finds a moderate positive correlation between the proposed Gedlee measure (Gm) and preference. He then chooses to improve the statistics of Gm by throwing out three unsupportive stimuli. Is this appropriate?
5. The paper is 12yo. Has there been any independent validation of the methodology or verification of the results? Have there been any additional developments or refinements of the Gm concept since 2003? Is any independent mfgr using Gm in place of or as a supplement to more traditional measures? A quick web search turned up nothing but not being a member of AES may be a limiting factor.
Overall, I found Geddes' findings and proposal to be interesting. But really isn't this just one relatively old data point and another dead end in the search for a means of correlating objective device performance with subjective preference or "goodness" using one or two measures?
1) I would assume it was chosen specifically it's revealing nature and i think 16/44 is sufficient.
2) There is another paper that discusses the distortion in more detail (it is a theory paper).
3) No, preference is a better way IMO as it takes off some of the pressure of "do I hear" vs. "what do I hear".
4) I think the findings of THD and IMD mirror what a lot of audiophiles out there complain about. I do not think that what Geddes does with his own metric is all that valid (one data point maybe if it fails a Q test but not three). His metric suggests he is on the right track but the fact of a weak correlation means there is something missing from his model to make it more complete. Maybe Cheever's model would fair better? Apparently, Shorter's simpler model was not that effective either or maybe more would use it.
5) I found a more recent pub from a Swedish (or Danish) group that reached similar, but more difficult to understand, results
Based on the synthesis of all recent research I have read I reached the conclusions that I posted some time ago on the amp/preamp forum.
http://www.audioasylum.com/forums/amp/messages/20/208108.html
I stand by my analysis until such time that I hear something or see research that falsifies what I wrote.
I posted a link where you can take the test yourself ,
Geddes findings and research is valid, what is not valid is how much of it actually correlates to us in actual use , when one hears their favorite song vs non familiar music , our sensory system will and usually does change gears...Short of directly taking and using amplifiers with the distortion characteristics suggested by Geddes research with golden ear audiophiles , what are we really proving....
Regards
Go Rossi ......
Edits: 11/14/15
Geddes found that IMD also has no correlation (actually slightly negative but statistically insignificant at the 95% confidence)- Morricab
Go Rossi ......
nt
try it! you know you want to!
Yep, that's the one
That wasn't difficult, was it?
Much of this stuff was known and effectively dealt with in the middle of the previous century by a few designers.
Then such got lost when SS first arrived and the arms race for "better" metrics and worse sound was prevalent. :)
"Then such got lost when SS first arrived and the arms race for "better" metrics and worse sound was prevalent. :)"
Seems that way although I don't believe "worse sound" was an actual design goal back then. I'd guess that the degraded sound was the result of the pursuit of unbalanced objectives with advertisable power and distortion specs taking precedence at a time when SS technology was not well developed. I have to believe the better designers knew what was going on but weren't in a position to resist.
The problem is the outcome of the techniques used to lower distortion, namely push/pull and negative feedback. The outcome gives the desired ENGINEERING solution of low distortion but doesn't achieve the SONIC soluton of having a proper harmonic content for good sound.
I'd guess that the degraded sound was the result of the pursuit of unbalanced objectives
Absolutely and to the point of this thread - chasing metrics that fail to correlate with qualitative aspects of the listening experience.
Which continues today with some designers.
Post a Followup:
FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
This post is made possible by the generous support of people like you and our sponsors: