Difference between Anologue and digital sound?
Mar 4, 2012 at 3:16 PM Post #46 of 78
DAC specs for dynamic signals use RMS - RMS value of 2 lsb pk-pk triangle pdf is not 2 bits
 
it actually simple enough to aproximate "on your fingers" - 1/2 the resulting rounded tpdf noise samples are 0, 1/4 +1, 1/4 -1 bit:  sqrt( (1^2 + (-1)^2)/4 ) = sqrt(1/2) ~ 3 dB
 
Mar 4, 2012 at 7:55 PM Post #47 of 78
Quote:
No, they're saying that nobody could hear the noise unless the playback volume was raised unnaturally high. Which obviously they did only during quiet or silent portions. If you raise the playback volume of 16-bit audio loud enough to hear the inherent noise of the medium, then playing music will blow out your speakers.
 
BTW, even though the original AES paper costs $20 for non-members, this site explains a lot about the tests:
 
http://www.bostonaudiosociety.org/explanation.htm
 
--Ethan


"[size=x-small]One of the authors, using a short repeated section of room tone on the Hartke disc mentioned above, obtained a positive result (15/15) at a gain of only 10 dB above our standard level. This setting produced sound levels clearly higher than those at the site, as the peak levels for this small vocal/percussion ensemble would have been 111 dB SPL on the loudest part of the disc. [/size]"
 
 
I see, thanks.
 
EDIT:  I think 85 dB was their standard level, so they're saying he used 95 dB, and the loudest part of the disc was 111 dB, that is actually pretty reasonable!
Reference: http://sound.westhost.com/articles/fadb-f1.gif
 
 
 
Mar 4, 2012 at 9:30 PM Post #48 of 78
Quote:
"[size=x-small]One of the authors, using a short repeated section of room tone on the Hartke disc mentioned above, obtained a positive result (15/15) at a gain of only 10 dB above our standard level. This setting produced sound levels clearly higher than those at the site, as the peak levels for this small vocal/percussion ensemble would have been 111 dB SPL on the loudest part of the disc. [/size]"
 
I see, thanks.
 
EDIT:  I think 85 dB was their standard level, so they're saying he used 95 dB, and the loudest part of the disc was 111 dB, that is actually pretty reasonable!
Reference: http://sound.westhost.com/articles/fadb-f1.gif


It was also in a room with an ambient volume of 19 dB. That's lower than normal. I think my room has an ambient volume of ~40 dB. So to hear what he heard, I'd have to play at 116 dB, with 132 dB peaks. Most rooms are probably around ~30 dB if they're not specifically designed to be quiet so say 106 dB with 122 dB peaks. No one should listen that loud.
 
Mar 4, 2012 at 11:26 PM Post #49 of 78


Quote:
Quote:
doesn't the discussion become redundant if the mediums respective faults are well beyond the tolerance of individual's hearing?


Not really, since so many are convinced vinyl is inherently superior and digital is inherently flawed. So long as people have the wrong idea, discussion should exist.


Even if digital is technically superior, if someone prefers the sound from their vinyl rig over their digital rig, then it's meaningless.
 
I recall reading a review of a $70k vinyl system where the reviewer stated he wouldn't buy it because it was too good and showed up the flaws in most of his records too easily. On the other hand, we now have a lot of very expensive SACD players that use disks that, by design, have a high number of actual errors on them, requiring included error correction data to play them back. That, to me, is crazy.
 
I reckon the key points were made in the early posts (before Kiteki interrupted with his "DACs draw straight lines between dots" stuff), the mastering is critical. While it's possible to argue technicalities of which format can do what, for various reasons, between a recording being made and put out on vinyl or CD, and more so when played back, various compromises are going to be made which can negate some of the technical benefits of either format.
 
Mar 5, 2012 at 2:31 AM Post #50 of 78


Quote:
 
84 dB does not seem to be correct, and using the two lowest bits may just mean a peak noise amplitude of +/- 2, rather than an RMS level of -84 dBFS. Furthermore, analog noise is usually specified as an A-weighted value, so the same should be done to dither noise as well for a fair comparison. Even a simple dither has better than 96 dB A-weighted SNR relative to a 1 kHz sine wave at 0 dBFS.
 



A - 90db signal does indeed use 2 bit, 1 positive & 1 negative giving 3 distinct levels & dither actually uses 4 bits 2 positive & 2 negative giving 5 distint levels. My bad but it still limits you to 84db S/N ratio in real practice. You can still record into that noise floor like you could with analog which without dither you couldn't. As such you can hear sounds recorded into the noise floor to about - 105db with flat dither & - 115db with noise shaped dither. This type of behavior is avalable with analog as well but it still speced to the noise floor even though you can hear past the noise floor. The noise floor is the point where the noise becomes audible, not where you can no longer hear the encoded signal. By the way various recording programs I've had determined the noise level of the dithered signal to be --84db that displayed recording level that low & I don't think these meters lie
 
 
Mar 5, 2012 at 2:50 AM Post #51 of 78
Quote:
Even if digital is technically superior, if someone prefers the sound from their vinyl rig over their digital rig, then it's meaningless.
 
I recall reading a review of a $70k vinyl system where the reviewer stated he wouldn't buy it because it was too good and showed up the flaws in most of his records too easily. On the other hand, we now have a lot of very expensive SACD players that use disks that, by design, have a high number of actual errors on them, requiring included error correction data to play them back. That, to me, is crazy.
 
I reckon the key points were made in the early posts (before Kiteki interrupted with his "DACs draw straight lines between dots" stuff), the mastering is critical. While it's possible to argue technicalities of which format can do what, for various reasons, between a recording being made and put out on vinyl or CD, and more so when played back, various compromises are going to be made which can negate some of the technical benefits of either format.

 
I'm not particularly familiar with the SACD format, but even normal CDs have error correction data.  When you save a file and open it again later, do you think it's crazy that you usually get the same thing that you had before?  Or that if you send a message over Wi-Fi, that it (1) requires error correction data and (2) the server gets your message intact?
 
Part of the discussion is about the compromises that are made during playback of each format, which is relevant as well.  With digital, the D/A stage of whatever you're using is inherently lossy to some degree (which may be incredibly small, or somewhat problematic in some cases).  Playing from an analog record, we have peculiarities of the needle, the physical medium, and so on, contributing to what are generally much greater losses.
 
But of course I'm not going to argue about what sounds one person likes or dislikes.  There are a lot of side factors as well--the experience is not all about the sound quality itself, though that's the focus of this topic.  For example, you could even argue that the act of interacting with vinyl to start the thing may prepare one to be in a better state of mind to listen to music. 
 
Mar 5, 2012 at 8:28 AM Post #52 of 78
Quote:
dither actually uses 4 bits 2 positive & 2 negative giving 5 distint levels.

 
5 distinct levels is actually 2.322 bits.
 
I did a simple test to compare a few dithers, and calculated the average RMS and A-weighted noise levels. The test signal was a 440 Hz sine wave that increased from -120 dB to -60 dB over a duration of one minute (so it is louder by 1 dB every second). After the dithering and quantization, I amplified the signal by a factor of 750 to make the differences easier to hear. The signal at around -96 to -84 dB (24 to 36s) is perhaps the most revealing of the dither quality. For the noise calculation, I used a notch filter at 440 Hz (Q = 100) to remove the original signal, and the levels are referenced to a 1 kHz sine wave at 0 dBFS. The results are:
Code:
 dither1: -95.21 dB (RMS), -97.61 dB (A-weighted) dither2: -86.66 dB (RMS), -91.54 dB (A-weighted) dither3: -91.72 dB (RMS), -96.09 dB (A-weighted) dither4: -94.78 dB (RMS), -98.46 dB (A-weighted) ditherns: -75.82 dB (RMS), -84.81 dB (A-weighted) (LPF): -84.29 dB (RMS), -93.03 dB (A-weighted) noise: -85.96 dB (RMS), -90.00 dB (A-weighted)
"LPF" means that I used a lowpass filter (3 dB attenuation at 18 kHz, "infinite" at 22050 Hz) for a perceptually more realistic result. The files compared are:
  - dither1: simple triangular dither
  - dither2, dither3, dither4: these are my attempts at simple dithering, the noise is intended to be uniform but with more energy in the high frequency range, and is used in various amounts to trade distortion vs. noise level; dither4 sounds reasonably clean, and is better than dither1
  - ditherns: noise shaping; this sounds by far the best perceptually, but has large amounts of high frequency noise, which makes the measured noise levels look bad
  - noise: this is not actually dithering or quantization, only pink noise mixed to the signal to (crudely) simulate "analog" noise
The test files are available here.
 
 
Mar 5, 2012 at 8:28 PM Post #53 of 78
 
Audacity is a open source sound editor
 
Audacity: Signal Gen create silence, (default project rez 32 bit float)
 
> edit>preferences>quality>high quality conversion dither: triangle
 
export 16 bit .wav
 
repeat with Signal Gen Sine 441 Hz 1.0 amplitude
 
 
keeping with the free software tools Scilab is a MatLab workalike, heavy duty math software for numerical/matrix/engineering calcs
 
SciLab:
-->y=wavread("C:\Users\jcox\tpdf.wav",1000);
 
-->x=wavread("C:\Users\jcox\sin441.wav",1000);
 
 
-->10*log10(variance(x)/variance(y))
 ans  =
 
    92.906146 

 
 
 
for my quick calc of tpdf dither RMS I didn't use enough of my fingers though:
 
 
Scillab normalizes wavread digital full scale to +/1.0, so to get unit lsb we multiply by 2^15
 
 
-->tabul(2^15*y)
 ans  =
 
    1.    134. 
    0.    725. 
  - 1.    141. 

 
so Audacity's triangle dither is +/- 1 lsb for a 2 lsb pk-pk as I said above, but the RMS is less than my calc
 
but the error just works out with the dynamic measure of sqrt(1/2) for the RMS digital full scale signal
 
 
-->variance(x)
 ans  =
 
    0.5005005 

 
 
variance is RMS squared for the sine, with +/-1.0 range full scale sine RMS ~= 0.707
 
Mar 6, 2012 at 11:03 AM Post #54 of 78
Audacity uses a different triangular dither that is noisier but has less distortion. By the way, in the table above, the RMS noise for dither1.wav was wrong (now corrected), because the file has a small DC offset. Filtering that out improved the noise by a few dB. For the Audacity dithered version of the file, I got RMS = -93.30 dB, A-weighted = -95.72 dB, but these are not perfectly accurate (there may be some error, but well below a dB). It may also depend somewhat on the input signal. I also tried the "shaped" dither of Audacity, but it did not seem to work well at all; maybe it is buggy. I was able to make some improvements compared to dither4.wav by shifting the noise energy more into the high frequency range, but still nowhere near the sound of the noise shaped version. Although I guess at realistic listening levels all would sound the same, with the exception of some very noisy "experimental" dithers.
 
 
Mar 6, 2012 at 4:45 PM Post #55 of 78
Wonder how bad a sweep on an LP can look like?
 

 
Only the single big peak between 2 kHz and 3 kHz is the actual signal.
 
 
 
Mar 6, 2012 at 5:12 PM Post #57 of 78
Ideally, something like this:
 

 
 
Mar 6, 2012 at 5:15 PM Post #58 of 78
just about any stereophile CD player review
 
http://www.stereophile.com/content/marantz-cd5004-cd-player-marantz-cd5004-cd-player-measurements
 
there are some required extra info needed for reading spectral plots - the record time/bin size changes the plotted "noise floor" -- unlabeled audio graphs should be 1 second/ 1 Hz bins
 
 
on dither - I don't think anyone today considers uniform 1 lsb as an acceptable "baseline" for audio applications 
 
the 2 lsb p-p triangle probability distribution dither doesn't have audible noise modulation with signal level - a audibly unacceptable flaw of the 1 lsb uniform noise dither
 
2 lsb p-p triangle dither gives ~ 93 dB  S/N, is used when triangle is selected in Audacity - TPDF dither is pretty universal in DAW, preferred over noise shaped dither when more processing after dither is anticipated
 
advanced noise shaping dither schemes are only appropriate as the final step for distribution
 
Mar 6, 2012 at 8:41 PM Post #59 of 78
As much as I enjoy hearing this scrap play out, as it has elsewhere many times, let's face it. It's all relative. When cd's first came out, I was pretty impressed with the obvious advantages. It took me a long time to get around to committing to a high end table and MM (happy I saved my vinyl). I was thrilled that I had "surpassed" the cd! Now the worm has turned again, perhaps. I'm surprised that there hasn't been mention of transistor vs tube here, as this is a valid part of the discussion.
I suspect that for extreme digital fans, the need for the final conversion from zero's and ones back to analogue (speakers, ears). may be a source of  frustration. Despite the very best in electronics, a cheap pair of speakers still will sound like .......Yes if you spend 5 figures plus on your system, it will sound great, analogue or digital. So coming up with balanced compromises is what makes the result worth the effort. It's not about digital or analogue, it's about enjoyment. After the sound has faded, was it good for you??
 
Mar 6, 2012 at 9:12 PM Post #60 of 78
Regarding straight lines between the sampling points, aren't "sharp" points in the dot-to-dot comprised of infinite frequency components? After filtering, won't the waveform go back to the original?
 

Users who are viewing this thread

Back
Top