This is illogical if we assume that the cause of a supposed audible difference between HCS and FCS, or to put it in simple words, low definition vs high definition, comes from an intermodulation between two neighborous ultrasonic frequencies.
But this is not sure. Other experiments, like Griesinger's one ( http://world.std.com/~griesngr/intermod.ppt ) that can be easily reproduced (but can easily fry your tweeters), show that even between an audible and an inaudible frequency, no intermodulation is audible. So there may be another process at work. The most likely that I can thing of is a distorded impulse response. Distorded in a way that can't be modelized in terms of harmonic or intermodulation distortion. The ear would thus react differently to a lowpassed impulse than to a full range impulse. But it does not imply that it should react in any way to a high passed impulse.
Your point is interesting, but this apparent inconsistency doesn't surprise me. Actually, all this stuff is inconsistent to begin with : we know that past a given frequency, pure tones can't be heard, and it seems proven that these frequency don't intermodulate with lower ones in our ears. Thus it would be illogical that high definition audio formats can sound any different than low definition ones (talking about sample rate only).
So dismissing this result just because of this is quite the same as dismissing it just because it is successful.




No I was not talking about this at all.
I considered the following argument :

1-The spectrum analysis shows that musical instruments have weak high frequency content.
2-Tests have showed that even strong high frequency content don't intermodulate with audible frequency
Conclusion : higher sample rate can't improve audio quality.

And I object that the spectrum analysis of musical instruments only show the average level of their high frequency content, while the instant level, especially in percussive instruments, during transients (=impulses) might be much higher.

It should be shown using spectrograms (3D graphs with vertical = frequency, horizontal = time, color = intensity) instead of spectrum analysis (2D graph with vertical = intensity and horizontal = frequency), but there again, I wonder if changing the analyser setting (using a shorter window analysis) would not show even higher HF content.

Thank you for the reference. Unfortunately, I could not open/view this link.

Assuming(for this discussion) this is a valid experiment, and the conlcusion is accurate.....



This, perhaps, has some validity. I remember a couple of years ago, I submitted myself to ABX trials, comparing simulated phase distortions applied to a signal to mimich a 3 way 4th order LR crossover vs. the unmodified signal. The effects were not audible/discernible in normal program material clips. Howver, in transient clips(rim shots, etc.) I consistently scored significant results. I don't mean to draw direct correlation with the results of thes bandwidth tests. However, I suppose it is possible I could detect the different signal in that case due to the total percieved amplitude on the peaks in a limited time event. The heavily phase distorted signal would have spread the total spectral conent of the transient acoss a slightly wider time space, meaning less total percieved amplitude on the peak? I am only guessing. Perhaps, since the high frequency content that is in the audible range in these tests, MUST be severely phase distorted at the higher frequencies as you approach 22khz cutoff, that in the event that the band is widened, thus allowing a more linear response, tht the peaks in certain cases are audibly different due to the percieved amplitude of that peak? Just a crude idea.

-Chris

Try from the html webpage : http://world.std.com/~griesngr/

The link is near the bottom. It is called Slides from the AES convention in Banff on intermodulation distortion in loudspeakers and its relationship to "high definition" audio.

In French, a positive ABX result between castanet2-1644.wav and castanets2-2496.wav from the PCABX samples, by GBo :

http://www.homecinema-fr.com/forum/viewtop...76126#168172020

ABX 11/12

Beware of the slippery snake oil everywhere else in that forum.

I can not read French and a translator makes a mess.

Taking into account the known IM problems with a single transducer attempting to cover both sonic and ultrasonic ranges -- I wonder the point at all of this ABX test -- when performed with hardware that may cause audible artifacts not directly relevant to the intention of the test(audibility of bandwidth variable only). I believe controls on the actual audio performance need to be established at this point.

ALternative: Someone assembles a special headphone(as someone suggested earlier in the thread) using a normal band and ultrasonic band driver element, amplilified by two seperate amplifiers and an active crossover circuitry. A 24/96 USB sound device could be picked, tested and confirmed to function adequately for the test. After complete system is tested for properly operation -- This system could be shipped between test subjects: they only need plug in the USB device, install software/samples/drivers and proceed to perform the ABX test using their computer.

Feasibility?

-Chris

I agree with you. If we want to prove that ultrasonic frequencies can have an audible effect, we need bi-amplification, or another way to get rid of IMD.
However, ABX success seem to show that after all, there might be an audible difference between a 44.1/16 bits recording and a 96/24 bits one, whatever causes this.



This seems out of reach, unless you have the money and time to make the headphones.
It is not difficult to set two speakers in one headphones. It was already done with the Superex PRo-B VI, for example. The passive filter was inside the shells (and the headphones weighted 480 g !). But the frequency junction between the drivers had a wavelenght larger that the size of the tweeter, which allowed to play the bass around it in a coaxial way in front of the ear. Maybe trying to set a super tweeter in the middle of headphones, and still try to reproduce high frequencies up to 18 kHz from another driver behind it would lead to big problems.
Also, it will become difficult to check with a microphone that there is no intermodulation at the output. One would have to use an artificial head to record the sound.
Once done, we still need to find some people willing to pass the test (but if you have the money to build the headphones, it should not cost much more to pay people for undergoing the test).

Anyway, I think we should investigate a completely different field first. The matter of ultrasonic frequencies has very little impact in my opinion, since most speakers and headphones can't play them anyway ! What's the use of extending SACD's response to 100 kHz while only plasma tweeters can reach this frequency ?
In that french forum, sort of an elite audiophile one, the discussion about high definition formats quickly took an interesting way : people agreed (though without any proper listening tests) that the main problem should not be the reproduction of high frequencies, but the audibility of the antialias filters.
They gave the example of HDCD. According to them, the HDCD information tells the DAC which filter to use in order to perform first order oversampling. This is true, the docs at http://www.hdcd.com confirms it.
I couldn't verify it but it seems very plausible, HDCD would use a steep filter when the music has much high frequency content, in order to preserve it, and a soft filter when music has much transients, in order to avoid ringing, and keep temporal accuracy.
If this is founded, the improvements given by high definition formats would be to simply get rid of the limitations of these two kinds of filters. A 24 bits 96 kHz PCM stream has a linear frequency response much above 20 kHz, and is free of ringing in the audio band.

We should first check if this is audible, which would confirm the superiority of DVD-A and maybe SACD over the audio CD, which suffers either from ringing either from reduces bandwidth, and check also if 48 kHz 24 bits would not be enough to solve all problems if they exist, which would show that SACD and DVD-A are nonetheless overkill.

The most direct way would be to use a high end audiophile DAC which gives the choice between different antialias filters. Some models do it.

Unfortunately, I don't plan to do it in the near future, because I don't have such a DAC, nor the money to buy one.

Yes, 'A' difference. I am not yet comfortable attribuitng this to the bandwidth -- I agree that other indirect issues are probably responsible.



I do not have access to the required headphone testing heads in order to do proper/consistent measurments of headphones. 2nd, the cost of the testing rig would be of concern -- I would not be comfortable purchasing these materials then shipping them to unknown people.




Audibility testing has been performed on anti-alias filters:

Perception of Phase Distortion in Anti-Alias Filters

AES Preprint Number: 2008 Convention: 74 (September 1983)
Authors: Preis, D.; Bloom, P. J.

EDIT: It should be noted that the above perceptual testing paper specifically dicusses an audibility test performed using pulsed test signals played over headphones. Pio, If you want a copy of the paper, e-mail me: wmax@linaeum.com

-Chris

I'd be curious to know the conclusions of that 1983 AES paper.

Also, if we consider the hypothesis that it's tha antialiasing filters that are audible,
would this be considered an intrinsic audible difference between Redbook 16/44 and higher-rez formats, or would it be considered an implementation issue? In other words, is it *inevitable* that 16/44 must employ audible antialiasing or is the putative problem separable from the format?

I didn't have the time yet to read all this page, that seems interesting : http://www.zainea.com/firing.htm

...but it reports the results of Bloom and Preis : they didn't test the audibility of the antialias itself, but the audibility of group delay in the antialias. Also, they tested 4 kHz and 15 kHz antialias. So it's irrelevant for the audibility of anti alias in 44.1 kHz DACs

Edit : oops, sorry WmAx, I didn't see your edit. Are there some more info in this paper ?

Actually, the findins are relevent to 44.1. THe purpose of the paper was to test for sensitivity of a given filter at a determined frequency. At 15kHz, in this test, no one could detect a change with an extremely sharp 'brickwall' filter(as some call it - thought not truly a brick wall filter) in place. At higher frequencies, the sensitivity is logically further reduced. If you can not detect the filter at 15kHz, it's not logical to assume it can be heard at 20 or 21 kHz. The Optimal Bandwidth for Sound Transmission paper I referenced earlier in this thread also uses a sharp filter in order to accurately test for bandwidth sensitivity with all releveant factors(including the sharp anti-alias filter). Again, they were not even able to get to 20kHz before no one could identify a difference. As for aliasing errors becoming audible, that is anothe issue from the bandwidth/filter rate issues, though directly affeted by the filter. However, I can not imagine that aliasing errors being of a magnitude to be of any concern today. Upsampling/interpolation has been a standard feature of most cd player DACs for at least 2 decades - for the very purpose of decreasing errors by way of a higher precision anti-alias filter. But maybe I'm missing something?

If either you or krabapple need a paper that I reference, just email me. ( wmax@linaeum.com )

-Chris

I'm not aware of anyone demonstrating under properly controlled conditions, that a properly used anti-alias with filter as is required on 44.1kHz, has any audible effect for humans.

-Chris

I created a 96khz sample (a bad electric guitar doodle) that I was able to successfully ABX against a downsampled 44.1khz version. It wasn't easy to do, but there were differences (the 96khz sounded a bit "fuller" to me). Could others ABX this sample as well?

http://www.hydrogenaudio.org/forums/index....ost&id=4387

Interesting. How did you downsample it?

I ran 3 tests, first was the original 96khz vs 44.1khz (resampled with AA 1.5)
second test was original 96khz vs 44.1khz (resampled with r8brain)
third test was original 96khz vs 96khz (downsampled to 44.1khz with r8brain then upsampled again)

I was able to pass the ABX test on all of them.

Anybody able to ABX this? If needed, I could provide logs.

ABX log for 96khz vs 44.1khz resample (r8brain resample)


It was a major PITA to ABX.

You have compared 96-24 vs. 44.1-16.
There were changed 2 different parameters, bit resolution 16 vs. 24 bit, and 2nd, sampling 44.1 kHz vs. 96 kHz.

Could you please create your sample as 44.1 or 48 kHz with 24 bit and compare that against 44.1-16 and/or 96-24, to get an impression, if the differences are due to 16 vs. 24 bit resolution, or due to 44 vs. 96 kHz sampling.

Vice versa, you could create a sample with 16 bit as 44.1 (you have that) or 48 kHz and compare that against 16 bit 96 kHz.

These tests could give further informations, where transparency is possible.

I assumed Unfortunateson had left the material at 32 bit floating when dropping the sampling rate.

Anyway, that's what I did this evening when testing his guitar "doodle". I used Audacity for the conversion to 44.1Khz. This evening when I listened, I found that initially the unconverted version of the guitar did sound different. However, I was only able to get 3 correct results with ABX before the samples started sounding the same and I had to abandon the exercise.


An easier exercise!
A much easier exercise is to compare the 96KHz/24-bit sample of a musical triangle located on the PCABX test page, to a downconverted version. Here is a link to the test page: http://64.41.69.21/technical/sample_rates/index.htm

I chose the Triangle Reference Presented At 24/96 version, and downconverted it to 44.1KHz at 32 bit floating point, using Audigy.

With the original sample, the striking of the triangle appeared to emanate from about midway between the front speakers of my home theatre set-up. With the resampled version, the sound appeared to emanate more from the left. This was such a clear difference of itself, I could ABX with ten tests quite easily, as follows:-.

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2008/04/18 21:53:51

File A: C:\Users\Public\Downloads\triangle-2_2496.wav
File B: C:\Users\Public\Downloads\triangle-2_2496as44-1KHz.wav

21:53:51 : Test started.
21:55:55 : 01/01 50.0%
21:56:41 : 02/02 25.0%
21:56:56 : 03/03 12.5%
21:57:11 : 04/04 6.3%
21:57:22 : 05/05 3.1%
21:57:33 : 06/06 1.6%
21:57:45 : 07/07 0.8%
21:57:56 : 08/08 0.4%
21:58:07 : 09/09 0.2%
21:58:13 : 10/10 0.1%
21:58:21 : Test finished.

----------
Total: 10/10 (0.1%)

I tried other software (ntrack studio 4) and it also moved the sound of the struck triangle towards the left hand speaker when converting to 44.1KHz.

I also tried headphones: the effect was much less apparent.


I note that whereas (as mentioned in other threads) it has not been possible to establish that a reduction from 24-bits to 16-bits (using proper noise shaped dither) can be ABXd with 44.1KHz sampled music at realistic recording and listening levels, there does seem to be quite a demonstrable difference between a sample rate of 96KHz and a derived 44.1KHz version of the same recording, at least with some material.

Whether that is due to vagaries in filters in creating a 44.1KHz version I do not know. As I say, I used 'Audacity' and 'N-track' for my sample rate conversion. Perhaps other converters leave the sound closer to the original.* However even the 44.1KHz version on the PCABX webpage labelled "Triangle Down- sampled To 44 KHz Presented At 24/96" seems to emanate somewhat left of centre, when I play it back on my HTPC system.

I'd also mention that a distorted guitar sound, such as Unfortunateson provided us, may not be as prone to noticeable disturbance by a changed sample rate as other music, though I have not experimented with different types of music myself. [For me, the 96KHz version of the "guitar doodle" had a more solid sound, with more edge, but there was very little in it for my ears.]

_________________________

* Using another PC, and Adobe Audition, the movement in the stereo image after conversion to 44.1KHz of the triangle sample was much less, but there was still a discernable difference in the stereo image and in the sound quality, sufficent for ABX purposes.

Hi MLXXX,

can you also test some other ABX, with changing only 1 parameter each test ?
(In your abovely described tests, it is unclear to me, if you changed both parameters at same time, or only bitrate or only sampling.)

(Like I asked above the unfortunateson)

Ie. start sample has always 24 bit, 96 kHz, (= 96-24)

so, for ABX , both of you succeeded in discerning 96-24 to 44.1-16.

it will be very interesting,
if you can ABX (or not) the source 96-24 vs. 96-16, and next: 96-24 vs. 44.1-24 bits !

I don't believe I specified that the original was at 24bits, the original is the 16bits/96khz sample that I posted here.

OT: Thanks for that sample, it unearthed a bug in lossyWAV....

Hi user,
I had no desire to introduce an additional variable, so did all my tests at 24 bits (or 32 bits floating actually). But I see Unfortunateson has just clarified that the guitar sample was only 16 bits to begin with, anyway.

On the question of 24 bits vs 16 bits, evidence to date suggests that dither can effectively extend the resolution of a nominal 16 bit format final mix to the point where music sounds exactly the same whether kept at 24 bits or dithered to a 16 bit format. There can be a difference in the noise level and spectral distribution of the 'noise' but unless listening is at a very high level of gain, and the source had a very low noise floor to begin with, even these differences will not be noticeable.

A relevant thread for the 16 bit vs 24 bit question is here: 16 bit vs 24 bit, any samples that work?