New approach to audio measurement and correlation to sound.

[Robert F]

With the closure of the deeply frustrating and ultimately pointless thread "I don't believe cables make a difference", driven by years of experimenting and hearing the non existant differences in such things, I came accross a fascinating article on new approaches to measurement. The article, whilst part of a manufacturer's marketing material, does appear to actually break new ground and also attempts to correlate the audibility of such non controversial (joke) steps as changing of power cords and equipment support. Sorry there is no DBT backup for those besotted with this, but I hope the measurements and techniques used might provide fruit for a worthwhile discussion. See here:

http://www.nordost.com/downloads/New...easurement.pdf

[bwaslo]

You can do this kind of testing yourself:

Audio DiffMaker

It does the time and level alignment of two similar recorded WAV tracks, extracts the difference for listening. If you want to look at it in time domain instead, you'll have to load the resulting WAV to a viewer such as in Audacity.

[Robert F]

Do these tools allow the testing and visual representations of the results reported from page 6 onwards of the pdf?

[tomchr]

My fundamental problem with the first paper (linked to in post #1), is that they change multiple variables between tests. It's clear that there is a change in the frequency spectra, but not clear what caused it. It could be any of the three components they add between test A and B. The other thing that is not explored is if the difference is audible.

One could argue that the method described in the link in post #2 will expose differences that are audible. But games could be played with the gain of the difference signal to make any difference seem huge. Further, it assumes that the test environment remains constant between two tests. I'm also concerned with the accuracy of the time alignment between the recordings made in test A vs B.

In either case, I think the tests will be very difficult to repeat with consistent results. Ideally, if you ran the same test twice without making any changes, you'd get a difference signal of zero (plus some quantization noise from the sampler). But I bet that won't be the case in real life.

Interesting idea, though.

~Tom

[Robert F]

Originally Posted by tomchr

My fundamental problem with the first paper (linked to in post #1), is that they change multiple variables between tests. It's clear that there is a change in the frequency spectra, but not clear what caused it. It could be any of the three components they add between test A and B. The other thing that is not explored is if the difference is audible.

One could argue that the method described in the link in post #2 will expose differences that are audible. But games could be played with the gain of the difference signal to make any difference seem huge. Further, it assumes that the test environment remains constant between two tests. I'm also concerned with the accuracy of the time alignment between the recordings made in test A vs B.

In either case, I think the tests will be very difficult to repeat with consistent results. Ideally, if you ran the same test twice without making any changes, you'd get a difference signal of zero (plus some quantization noise from the sampler). But I bet that won't be the case in real life.

Interesting idea, though.

~Tom

Yes it is interesting, though you will note that they did measure, in the frequency domain, the effects of the three components individually; see the results at pg 4.

[Don Hills]

Some observations:
- The DiffMaker program can itself introduce differences, depending on how well it is coded. It attempts to process the input to match as closely as possible the reference signal. The processes used (equalisation, level matching, delay compensation, sample rate conversion) can introduce quantisation errors. The program iteratively processes to minimise the differences, but this factors in the errors caused by the program. This means that the reference is compared against a convolution of the input and the program errors. Also, by trying to minimise the difference between reference and input, it will minimise some of the wanted differences. For example, if a tweak causes an overall slight level shift or equalisation tilt, this will be cancelled whereas in an actual listening test you would want to hear these differences.

- They claim that that the "jitter like" difference signal noted in one of the tests was signal dependent, not random, and thus not actually caused by clock jitter. Clock jitter doesn't have to be random. In the majority of the tests I have seen where CD player clock jitter has been analysed, the largest component of it has been signal correlated. It's not hard to see why. Logic circuits consume different amounts of power as they switch. Anyone who has listened to a poor quality soundcard in a PC as the PC operates will have heard the wide variety of different squawks, whistles and buzzes caused by logic switching causing fluctuations on the supply rails. In a CD player, logic switching is going to be somewhat signal related. The resulting supply rail noise, ground rail level shifts, and inter-conductor coupling will perturb the clocking in a signal-dependent manner.

[Robert F]

Yes I noted the misstatement about jitter myself, supposedly signal correlated jitter is the more audible than random jitter too. I suspect the problem may be that the copy was not written by the testing engineer but rather by Roy Gregory who while presenting the findings, is not sufficiently technically literate to be making such statements.

[cbdb]

A cable maker trying to sell more cables, no wonder there are so many holes in there paper.

When can we expect this paper in an AES journal?

Never, it wouldnt cut the mustard. Its psuedo scientific at best.

As stated, too many variables changing. (and I dont think any body will argue that Rf filtering of the power can effect SQ, but that is not a cable issue its a filter issue).

Redbook audio CDs are not bit perfect storage media (poor error correction), did they keep tack of the uncorrectable errors?

Do they specify what kind of power they are using or if its consistant. Ive seen reidential/commercial AC vary from 110 to 130 volts, and line noise vary 20 db from day to night.

Why do they deliberately hide the numbers on there graphs? (making them pretty much useless),the text magnifies beautifully while the graphs stay unreadable.

The freq. domain graphs show that the "good power cable" system has made the test tones louder? How does that work?

The time domain graphs show that the ordinary system has slowed down the output (akin to low pass filtering)? How does that work?

Scientific experimental data can very easily be biased, and its pretty obvious this experiment is used to sell cables. They offer no explanation for the "measured" differences just that its been measured and they dont even show you the measurements until those graphs have numbers. So take their word for it and go spend too much money on there cables.

[Robert F]

Of course this is not a paper worthy of academic publication, but that is not to say a properly written paper demonstrating this research would not be.

The points you make about varying mains quality are well made, one could only record meaningful data if the mains quality were identical test to test. However, given that Nordost say they used a reputable, nay state of the art testing facility, I assume that they took care of these variables before submitting the information to Nordost. Unless of course Nordost are falsifying the results...not something that would be worth the risk in my opinion.

Secondly you haven't actually read the article correctly in at least one area it seems. I don't read the article to state that the "good power cable system" (note that there was equipment support and mains filtering components too) made the test tones louder, rather the peaks on the treated system vs the untreated system were louder. Quite a different assertion, no? On page 4 they separate out the effects of the mains cable, support and mains filter respectively, though sadly this is the only place the results are not lumped together.

How does a the player introduce a time lag? I don't know but if the results are correctly reported it would make a fascinating study don't you think?

The article doesn't address why there were changes just the fact there were, I don't see why it should be criticised for that.

Finally, if the "go and spend too much money on their cables" (once again there were 3 components here not just cabling) was directed at me, then I can tell you I build all my own stuff and haven't bought a commercial product in many years. I find the literature interesting but it will not change my spending habits except perhaps in pursuit of things I can build myself.

[cbdb]

Of course this is not a paper worthy of academic publication, but that is not to say a properly written paper demonstrating this research would not be.

Then why don't they write one? I'ld guess its because they couldnt, and not because they don't know how.

I don't read the article to state that the "good power cable system" made the test tones louder, rather the peaks on the treated system vs the untreated system were louder

Then you need to read it again. The peaks in the freq domain are the test tones levels.

How does a the player introduce a time lag?

Its not a time lag, but a freq. dependent phase (time) lag like you see in a low pass filter.

Finally, if the "go and spend too much money on their cables" (once again there were 3 components here not just cabling) was directed at me

No, it was directed at Nordost. (indirectly)

And still the biggest problem I have with the paper is that the graphs that support there assumptions that the 2 systems sound different have no numbers and are therefor useless, which kinda implies there paper is also useless. IMHO.