[nick_charles]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif Just in case anyone's using Wavosaur for any of this analysis - when I was testing RMAA, I noticed that THD increase by 300% after I did something with Wavosaur. So I did some more testing and found that simply opening the RMAA file and re-saving it increases RMAA's THD measurement by 300%. I tested with Adobe Audition and it didn't have this problem. Using Audacity, the THD increased by 100%. I assume/hope it's dithering causing this.

I am not using dither on my recordings as they are D-to-D and not (hopefully)going through DA/AD, dithering would increase noise yes ?

EDIT: In CEP I just trimmed my reference to 1.486 secs and trimmed a recording to the same length. Then I did the Invert/Add of the two files, ran a 64K fft and the difference file now shows levels of about -120db worst case and average on both channels of -146db , this is broadly in line with earlier attempts. Different window types do show sligthly different results but thay are all very close.

One final test: Digital silence of 1.486 and a low level -72db white noise add the two together = > white noise at -72db (diff test - all 000.000000)

And now for the bad news...

I took two recorded samples trimmed to 1.486 (65536 samples) did and invert/add and the difference file was averaging -144db with a peak at -115db. Tried this with a couple of different pairs getting similar results , thus the inter-sample variations are bigger than the differences between any 1 sample and the reference, thus it will be necessary to do many trials and average the results before any possibility of useful comparisons, but it does not bode well, now an average of -144db is small potatoes in real money but if differences do exist between optical cables in pure data transmission terms, they would have to be pretty sizeable to overcome this.

 

[nick_charles]

I just spent a few hours trimming all my samples down to sizes suitable for
64K and 32K FTT analyses. So I set up a CD in NERO to burn them and guess what Nero will not burn a track less than 2 seconds long ----- sigh

 

[leeperry]

make one track w/ lotsa sub-indexes?

I was kinda hoping you'd get some test tones through WaveSpectra to see what the THD/THD+N/SNR look like tbh, that could learn us quite a lot I think

 

[Prog Rock Man]

Will there be a Plain English, idiots guide version of this thread?

 

[haloxt]

Hehe jk. I need a layman explanation too, why can't you just make it a 2 second clip?

 

[nick_charles]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif make one track w/ lotsa sub-indexes? I was kinda hoping you'd get some test tones through WaveSpectra to see what the THD/THD+N/SNR look like tbh, that could learn us quite a lot I think

It's on my to-do list ! But I cannot do anything until I have suitable samples burned to CD as I use these for the recordings from which I can see what is happening...


Quote:

Originally Posted by Prog Rock Man /img/forum/go_quote.gif Will there be a Plain English, idiots guide version of this thread?

To correctly run an analysis of the frequencies of a sound clip the sound clip must be the same length (or no longer) in samples as the FFT (Fast Fourier Transform) used, so if I want to run a 65536 FFT analysis my clip must be no more than 65536 samples long. A 44.1khz signal (red book CD) that is 65536 samples long is 1.4861 seconds long.


Quote:

Originally Posted by haloxt /img/forum/go_quote.gif Hehe jk. I need a layman explanation too, why can't you just make it a 2 second clip?

Yes, I am padding the ends of the tracks with digital silence, this will make trimming and alignment easier.

 

[nick_charles]

First proper test run. Took reference file and burned it to CD. recorded sanple digitally from CD. Loaded up waveforms for reference and recording and did an invert/paste whih shows the difference between the files. Plotted it as below. Average digital grunty level is - 140.9db it aint by any means perfect but it is pretty good !

 

[nick_charles]

I took a 1K sine wave and ran it through Wavespectra and for the peak frequency 999.9hz the reference files measures

THD...............0.00037%
THD + Noise....0.00208%

The same signal recorded from CD to Edirol to PC (digitally)

THD...............0.00039%
THD + Noise....0.00401%

 

[leeperry]

very nice! but it might have been more meaningful w/ an original file that's 0% THD, and possibly 20kHz/20.5kHz to really push it

anyway...yes, bit-perfect, yeah right! this denomination is utterly meaningless...and makes me think twice about all those CD copies I did in coax on a standalone Philips recorder 15 years ago

good thing it had SCMS, so we couldn't make second gen copies..

 

[Pio2001]

Hello Nick Charles.

As I understand, your goal is to test digital cables by recording an S/Pdif output digitally.

It has been done several times (I did it many times), and the result always gave bit perfect results, even in the worse possible conditions (low end CD player, CDR burnt at max speed, shaky stand for the player, long runs of cable...)

I downloaded the files 07-130 chinese cymbal loop.wav and t6-v2.wav in order to investigate the difference. They are perfectly time-aligned. So I substracted the copy from the original and looked at the difference.

What remains is a constant noise spanning the three lowest digital levels -1, 0 and 1.

Zoomed part :

Larger part :

Statistics :

The spectrum is flat and the noise does not depend on the signal amplitude. It must be dither. The culprit are either the soundcard, or the OS, or the software. I remember having read that you tested the software and it was OK.
So it is maybe the soundcard that adds dither to anything it processes, maybe the operating system. You didn't tell what kind of audio drivers you used. In Windows XP, for example, you can only achieve bit exact playback or recording through special audio drivers, like ASIO, Kernel Streaming, or MME drivers.
Regular audio drivers (may they be DirectSound or not) do exactly what we see here : they alter the last bit of the data (sometimes from 0 to 1, sometimes from 1 to 0, sometimes not).

 

[nick_charles]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif very nice! but it could have been more meaningful w/ an original file that's 0% THD, and possibly 20kHz/20.5kHz to really push it

Okay I found a new (better) signal generator and will knock off a few more test files....


Quote:

anyway...yes, bit-perfect, yeah right! this denomination is utterly meaningless...and makes me think twice about all those CD copies I did in coax on a standalone recorder 15 years ago

Still, a worst case of 0.00401% Noise + THD is really not big bananas and this with a pretty cheap recorder, I daresay the EMU would better this...

 

[nick_charles]

Quote:

Originally Posted by Pio2001 /img/forum/go_quote.gif Hello Nick Charles. As I understand, your goal is to test digital cables by recording an S/Pdif output digitally. It has been done several times (I did it many times), and the result always gave bit perfect results, even in the worse possible conditions (low end CD player, CDR burnt at max speed, shaky stand for the player, long runs of cable...)

What did you use at the recording end ?


Quote:

I downloaded the files 07-130 chinese cymbal loop.wav and t6-v2.wav in order to investigate the difference. They are perfectly time-aligned. So I substracted the copy from the original and looked at the difference.

I use Invert and add but the effect is the same I guess..

Quote:

So it is maybe the soundcard that adds dither to anything it processes, maybe the operating system. You didn't tell what kind of audio drivers you used. In Windows XP, for example, you can only achieve bit exact playback or recording through special audio drivers, like ASIO, Kernel Streaming, or MME drivers.

The Edirol has it's own ASIO drivers, which are in use, I cannot get the recording software to work without using the card's ASIO drivers. I am running Win Vista 32bit (Sp2) and the recording software is set to 16/44.1 and no dither.

The CD player is a Marantz CC4300 which is a bitstream player not a multibit player.

My guess is that the cards is resampling, I have it set to 44.1 but it is a 24/96 card so it's native setting may be overriding the setting for input ??

 

[Pio2001]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif What did you use at the recording end ?

The source was a CD Player. Its digital output is sent to the following chain :

Sony DTC55ES (pass through) -> another optical cable -> Fostex COP-1 optical to coaxial converter -> Coaxial cable (5 meters) -> PC with Marian Marc 2 soundcard (MME drivers), Windows XP, Soundforge software.

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif The Edirol has it's own ASIO drivers, which are in use, I cannot get the recording software to work without using the card's ASIO drivers. I am running Win Vista 32bit (Sp2) and the recording software is set to 16/44.1 and no dither.

Oops ! That should work.

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif My guess is that the cards is resampling, I have it set to 44.1 but it is a 24/96 card so it's native setting may be overriding the setting for input ??

Maybe... or maybe its mixer is poorly designed, and the digital input goes through a 24 bits volume setting with built-in dither.

 

[Pio2001]

Quote:

Originally Posted by slim.a /img/forum/go_quote.gif If you read the links below, you will see that when listening tests are conducted with the proper equipment, the human sensitivity to temporal resolution is far greater than previous scientists have assumed before. Information for prospective students http://www.physics.sc.edu/kunchur/pa...rge-Foster.pdf

Thanks for the links.
Kunchur's papers are very well written and documented. All the usueful information is available.

I see however a weak point is his demonstrations. He assumes that the Just Noticeable level Difference for a 7 kHz sinewave at 69 dB is 0.7 dB (and cites the source of the information). I decided to test this myself. I generated two 7 kHz sinewaves with a volume difference of 0.25 dB, which is the level difference between the stimulii at 5.6 µs used in his first paper.
I could recognize them in ABX with a score of 8/8 (Foobar2000 ABX plugin, no DSP, no replaygain), at a listening level inferior to 80 dB (can't tell exactly how much).

This puts seriously into question the conclusion that the stimulii in his studies were discriminated by temporal detection only, for both the electronic lowpass experiment, and the speaker displacement experiment.
Moreover, I also found the louder sine to sound brighter, as the listeners said in the experiments. Except that in my experiment, there was no 21 kHz component, and no time difference at all !

The calculation about the 14 kHz non-linear component appearing in the ear of the listeners in the electronic lowpass experiment is interesting, however, the absolute level of this component is unknown. In an old discussion about high sample rates in old George Massenburg forum, Nika Aldrich and I noticed that if we play an audible sine together with an inaudible one (12 kHz + 18 kHz in my case) in a speaker, intermodulation is audible, while if we play them through two different speakers, intermodulation is not audible anymore.
Therefore the calculation, though providing an interesting hypothesis about the possible cause of audibility, does not prove it.

 

[nick_charles]

I spent some time altering settings in Cool Edit Pro and on the Edirol, which was designed pre-Vista. The only way I can get the system to work as a digital recorder is to use the ASIO drivers. I went through CEP to make sure that no random dithering setting was lurking. That done I did another test recording. Same result,so it looks like the card is the culprit and that said there does not seem much point carrying on, especially since you have managed bit-perfect digiital recordings with cheapo cables and several intervening stages...