or Connect
Head-Fi.org › Forums › Equipment Forums › Sound Science › Sampling Rate
New Posts  All Forums:Forum Nav:

Sampling Rate

post #1 of 7
Thread Starter 

We have the 16 vs 24 bit thread going on here but I am interested in a 44.1khz vs 192khz discussion. I have read several threads both here and elsewhere that say either, there is absolutely no audible difference or there is a clearly superior sound to the higher rez file. To start off though I want to clarify something in my mind and that is what sampling rate is.

 Here is how I understand it, rounding off some numbers just to make it easier to understand. For the sake of argument let us say that we can actually hear 20khz, and we are sampling that signal at 40khz rate (close enuff to 44.1), doesn't that mean that each cycle of that audio is only going to get sampled twice?( 20khz/40khz) Where on the other hand if we bump the sample rate up to say 200khz, the same cycle of 'music' would get sampled 10 times? If so it sure seems like the 192khz files would sound better, in the treble area at least.

post #2 of 7

http://xiph.org/video/vid2.shtml

 

This excellent 24 minute video will answer your question, and give you a good picture in your mind as to what is really happening. 

 

If done correctly higher sample rates would only give additional info at higher frequencies, and identical waveforms as lower sample rates at those lower frequencies.  So if you cannot hear more than 20 khz, there no additional info perceptible to you.  Hi-rez in regards to sample rate is actually a misnomer.  It should be called broader bandwidth recording.  As you don't really need the extra bandwidth it isn't helpful. And yes, 20 khz waves would only be sampled twice per wave, and that is enough to reconstruct the sine wave at that frequency.

 

Now 24 bit vs 16 bit does theoretically give additional resolution at levels you might hear though practically at the playback end it may not matter.  Or matters very little and rarely.  But spend just the few minutes watching that video above it is very instructive.


Edited by esldude - 9/28/13 at 11:35am
post #3 of 7

there is extra information in more (independent, 16 bit) samples in the same time range

 

at higher sample rate the quantization noise (preferably decorrelated with dither) is spread over higher than audio bandwidth so there is less perceptible quantization noise or dither in the audible frequency range with 192k vs 44.1 with the same word size

 

the reconstruction filter is usually set at a lower fraction of Nyquist with the higher sample rate so even the full bandwidth analog output noise can be less with higher sample rate

 

while the perceptible noise floor reduction effect is relatively small for flat with frequency noise, noise shaped dither can take advantage of the inaudible higher frequency to deliver much higher perceptually weighted audio S/N, better than 20 dB lower perceived noise floor with higher sample rates

 

so with noise shaped dither taking advantage of 2-4x over sampling you can easily get higher than 16 bit ~96 dB resolution in S/N - over 120 dB dynamic range ~ 20 bit effective audible resolution even with (more) 16 bit words

Quote:
Originally Posted by jcx View Post

actually dither can be used to give even greater psychoacoustic perceived S/N when you increase sample rate - I'd rather have 16/96 if trying to responsibly over bound human hearing ability than 24/44

 

http://www.meridian-audio.com/w_paper/Coding2.PDF

 

if you want to play the game with only a 50% increase still go for sample rate - would end up pretty much per Stuart's suggestion in the paper - and today's dither algorithms can improve over what he had available

Quote:

...actually listen to audio at differing bit depths:
 

Homepage of Alexey Lukin

8,12 and 16 bit audio samples as well as more dither theory on that site

also if you want to see how various recoding format map over human perceptual limits:

http://www.meridian-audio.com/w_paper/Coding2.PDF

the paper is very technical (by head-fi standards at least, it is a “popularization” of a JAES convention paper)
in support of his argument for a particular coding scheme Stuart summarizes some of “conventional” audio engineering/psychoacoustic understanding of human perception limitations and their relation to reproduced audio – you might want to jump to the figures/graphs at the end of the paper and then search back into the text for the explanatory context

yes, I really like the Bob Stuart paper


Edited by jcx - 9/28/13 at 1:27pm
post #4 of 7
Thread Starter 
Quote:
Originally Posted by esldude View Post
 

http://xiph.org/video/vid2.shtml

 

This excellent 24 minute video will answer your question, and give you a good picture in your mind as to what is really happening. 

 

If done correctly higher sample rates would only give additional info at higher frequencies, and identical waveforms as lower sample rates at those lower frequencies.  So if you cannot hear more than 20 khz, there no additional info perceptible to you.  Hi-rez in regards to sample rate is actually a misnomer.  It should be called broader bandwidth recording.  As you don't really need the extra bandwidth it isn't helpful. And yes, 20 khz waves would only be sampled twice per wave, and that is enough to reconstruct the sine wave at that frequency.

 

Now 24 bit vs 16 bit does theoretically give additional resolution at levels you might hear though practically at the playback end it may not matter.  Or matters very little and rarely.  But spend just the few minutes watching that video above it is very instructive.

 

 

Thanks for that link, a lot of it was over my head, but it certainly shows that 192khz isn't all that necessary to getting high quality audio. Looks to me like 24/96 is really all you would ever really need and even that is a bit of overkill.

post #5 of 7
Quote:

Originally Posted by esldude View Post

 

And yes, 20 khz waves would only be sampled twice per wave, and that is enough to reconstruct the sine wave at that frequency.

 

Please allow me this pedantic correction:

The sampling theorem requires greater than 2 samples per cycle. (Example: 2.0001 would theoretically be enough..)

The filters in current DACs start rolling off at about 20 to 21 kHz with a 44.1 kHz sampling rate. That's roughly 2.2 to 2.1 samples per cycle.

 

Of course reconstruction doesn't abruptly stop there, but the filters are usually very steep so the level of those >20 to 21 kHz sine waves will drop extremely fast.

 

 

Using software resamplers, you can easily push the cutoff frequency to 21.5 kHz or higher.

post #6 of 7
Quote:
Originally Posted by HPiper View Post

Looks to me like 24/96 is really all you would ever really need and even that is a bit of overkill.

Yes, that's overkill. Using 16 bits at 44.1 KHz is plenty, and it's a heck of a lot better than any analog recording setup. Even the best analog tape recorders struggle to achieve the equivalent noise floor of 12 or 13 bits. And analog distortion? Distortion amounts of 1 to 3 percent are common and typical.

--Ethan
post #7 of 7

24/96 is really useful - don't think many studios are recording at 16/44 or even 48k

 

 

read the Stuart paper - I think there is a case for higher sample rate than 44.1k

 

and elsewhere it looks like there is likelihood that human sound frequency perception extends to 22-24 kHz, even if only for prepubescent girls - not proof

 

the combined advantages in reconstruction filter relaxed requirements and noise shaped perceptual noise floor improvement make me vote for more samples, certainly over more bits at the lower word rate - even if someone could prove "no one can hear..."


Edited by jcx - 9/30/13 at 11:35am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › Sampling Rate