Sampling rate and bits per sample – a practical comparison

[xnor]

Hello, folks!

I've done a few comparisons the last 2 days and came up with this article: Sampling rate and bits per sample – a practical comparison
@skeptics: don't take the blog title or rant-ish parts too serious

I hope it's comprehensible and that there are no mistakes. If there are, please let me know.

Enjoy.

 

[Publius]

You're confusing spot noise with total noise when you make that "-120db is below the CD noise floor" comment. IIRC, with a 16k tap FFT, it's more like 10db above the noise floor - possibly more. The last time I evaluated the spot noise floor of 16/44 I came up with about -150db, but I forget the FFT block size I was using.

Your comments regarding plotting difference signals on a linear plot (as opposed to exponential) are just not meaningful. The human ear has a logarithmic response to stimumus in many ways - it's not a waveform reader. Some signals are audible 80db below other signals, but you would never notice that from a linear plot. The spectrum plots of the difference signals are far more meaningful here, and there isn't much need for them to be plotted linearly either.

The dynamic range of a piece of music has nothing to do with its spectral content evaluated along any particular section of it.

You claim that labels producing high res tracks master the low res files differently but provide absolutely no evidence of this. That sort of thing has been documented for SACD but IIRC never for high res downloaded music. That's a rather ugly smear to throw about without having, you know... facts.

I agree with your conclusion, and it's clear you spent a lot of effort on this which must be commended, but srsly?

 

[xnor]

Quote:

Originally Posted by Publius /img/forum/go_quote.gif You're confusing spot noise with total noise when you make that "-120db is below the CD noise floor" comment. IIRC, with a 16k tap FFT, it's more like 10db above the noise floor - possibly more.

All those graphs have a 16k tap FFT.

Quote:

The last time I evaluated the spot noise floor of 16/44 I came up with about -150db, but I forget the FFT block size I was using.

Got any screenshots and audio files?

Quote:

Your comments regarding plotting difference signals on a linear plot (as opposed to exponential) are just not meaningful. The human ear has a logarithmic response to stimumus in many ways - it's not a waveform reader. Some signals are audible 80db below other signals, but you would never notice that from a linear plot. The spectrum plots of the difference signals are far more meaningful here, and there isn't much need for them to be plotted linearly either.

Hence the magnification. JOKE!

That's just for the fun!

Quote:

You claim that labels producing high res tracks master the low res files differently but provide absolutely no evidence of this. That sort of thing has been documented for SACD but IIRC never for high res downloaded music. That's a rather ugly smear to throw about without having, you know... facts.

http://www.head-fi.org/forums/f133/2...ml#post5944934

Quote:

I agree with your conclusion, and it's clear you spent a lot of effort on this which must be commended, but srsly?

pwetty

 

[Publius]

See for yourself. :F I just generated silence in Audacity, exported it to a 16/44 WAV with TPDF, reimported, amplified it 90.3db and plotted the spectrum. Total spectrum peak is -59db at Nyquist and is -66db for 0-3khz. This is as a PSD measurement. So, 149-156db.

Or, if you prefer, arguing strictly from spot power, and treating 96db as an RMS noise figure (which is a little bit of a cheat): 96+20 log10(sqrt(22050)) = 139db.

 

[xnor]

You're right, that's a mistake and it's fixed now.
That paragraph is about comparing noise coming from the soundcard to dynamic range / signal-to-noise ratio (not noise floor, as you said) on a CD. Maybe I should rephrase that a bit, hmpf I'm no native speaker!

Btw, silence (which I'm not talking about in the article) should be silence. -Infinity dB if you ask me. I don't know what you're seeing there. Maybe dithered silence? xD No seriously, don't use Audacity! Even if you go to settings -> quality and set the sample format to 24 bit (non float), generate silence, set the track's format to 16/44.1 (you can do that directly in Audacity), save and reopen and zoom in you'll see it's not 'silence' anymore. That's quite amusing.

edit: Didn't see the other dynamic range comment - that sentence needs to be rephrased too.
edit2: rephrased.