16 bit vs 24 bit sampling , much better?

[analogika]

I am not an expert on this, but I believe the 16 bit vs 24 bit would make a bigger difference when you are recording a live performance, where the volume levels will be uncertain. Clipping during the capture is fatal, so you have to avoid that by leaving some headroom. As the recording engineer for a live performance, you really don't know just how hard will the drummer whack the snare or the cymbal, or how loud the guitarist will get when the high energy moment arrives. So you end up having much of the capture take place well below the level at which clipping begins. In this scenario, I believe the 24 bit vs 16 bit advantage is large. But if you are just sampling something in a controlled environment, where you can precisely control the volume of what is captured, and do it over if you get it wrong, I believe the advantage for 24 bit will be much less.

I'm almost sure it's not like you are saying. When an analog signal is converted into digital, each voltage value is converted to a digital binary number at a frequency that now is not important. So with 16 bits you have 2^16 (65536) different numbers representing the various voltages of your signal. With 24 bit you can be more detailed because you have 16.777.216 numbers. I think this is like cutting a pizza in 8 or 16 pieces, the pizza is the same area but the slices are smaller. Audio-speaking, you can reconstruct your signal with more detail thanks to the fact you sampled the little differences.

That's wrong.

The signal can be constructed with exactly the same amount of detail up to the Nyquist frequency (sampling frequency /2, so 22.05 khz for 44.1 kHz sample rate), because there is only ever one possible waveform that represents the sample points recorded. (That bit of information right there broke my head for a long time.)

What the higher bit depth does is effectively lower the noise floor. It thereby increases the available dynamic range; it does not provide a more detailed dynamic range.

In effect, when you're recording a very dynamic signal in 16 bits, you have to drive it as hot as you can so that the softest parts of the signal don't get lost in bit nirvana below the noise floor. This means that you're at risk of clipping the input unless you limit it beforehand.
In 24 bits, you don't have to turn up the signal nearly as much when recording, because you have all that dynamic range at the bottom. So you virtually never have to risk clipping the signal.

[mircolord]

I am not an expert on this, but I believe the 16 bit vs 24 bit would make a bigger difference when you are recording a live performance, where the volume levels will be uncertain. Clipping during the capture is fatal, so you have to avoid that by leaving some headroom. As the recording engineer for a live performance, you really don't know just how hard will the drummer whack the snare or the cymbal, or how loud the guitarist will get when the high energy moment arrives. So you end up having much of the capture take place well below the level at which clipping begins. In this scenario, I believe the 24 bit vs 16 bit advantage is large. But if you are just sampling something in a controlled environment, where you can precisely control the volume of what is captured, and do it over if you get it wrong, I believe the advantage for 24 bit will be much less.

I'm almost sure it's not like you are saying. When an analog signal is converted into digital, each voltage value is converted to a digital binary number at a frequency that now is not important. So with 16 bits you have 2^16 (65536) different numbers representing the various voltages of your signal. With 24 bit you can be more detailed because you have 16.777.216 numbers. I think this is like cutting a pizza in 8 or 16 pieces, the pizza is the same area but the slices are smaller. Audio-speaking, you can reconstruct your signal with more detail thanks to the fact you sampled the little differences.

That's wrong.

The signal can be constructed with exactly the same amount of detail up to the Nyquist frequency (sampling frequency /2, so 22.05 khz for 44.1 kHz sample rate), because there is only ever one possible waveform that represents the sample points recorded. (That bit of information right there broke my head for a long time.)

What the higher bit depth does is effectively lower the noise floor. It thereby increases the available dynamic range; it does not provide a more detailed dynamic range.

In effect, when you're recording a very dynamic signal in 16 bits, you have to drive it as hot as you can so that the softest parts of the signal don't get lost in bit nirvana below the noise floor. This means that you're at risk of clipping the input unless you limit it beforehand.
In 24 bits, you don't have to turn up the signal nearly as much when recording, because you have all that dynamic range at the bottom. So you virtually never have to risk clipping the signal.

Thanks for the explanation! I imagined it was more like ADC conversion on embedded systems where more bits meant only more detailed representation of the voltage. I'l re-read your post a few times just to be sure I understood it

[mircolord]

That's actually correct, just that the "more detailed representation" you mention translates into "higher dynamic range"( between the lowest and hoghest possible values) and it is calculated as 6dB/but so you have 96dBs for 16bit and 104dB for 24bit sampling, but this does not influence the spectrum (frequencies) accuracy (so the timbre) which instead depends on sampling frequency.

[mircolord]

It really hurts my head now (as said before) although it seems 100% reasonable, one day I'll have my true Eureka-Moment about that

[Mr_-G-]

You can see (and hear!) this by yourself, for example taking it to the extreme in Audacity and using the "Decimator" plugin: if you keep the sampling frequency constant and lower the bit depth to 3 or 4 bits, you can see how the noise and distortion build up drastically. So it changes the spectrum (and drastically) but you can still can recognise the content to some extent (if it did not do anything, there would not be a "bit cruncher" effect!).

[cgrafx]

In simple terms:

Sample frequency = maximum frequency you can capture and playback
Bit depth = Dynamic Range

[analogika]

In simple terms:

Sample frequency = maximum frequency you can capture and playback
Bit depth = Dynamic Range

Correcterer:

Sample frequency = twice the maximum frequency you can capture and playback
Bit depth = possible dynamic range, figured downward from the maximum of 0 dB (down to -96 dB at 16 bits, and -144 dB at 24 bits, undithered).

[sweelinck]

On this topic and dynamic range, I was testing my home studio monitors today using the test audio files on https://www.audiocheck.net and ... see in particular their https://www.audiocheck.net/audiotests_dynamiccheck.php, try it and then read and the "note" on that page about 16bit vs 24 bit

Yes. Per Bob Katz (among the best audio engineers in the business) in his book “Mastering Audio” our hearing generally has 20 bits of dynamic range. If we compress to 16 bits using dithering, it fools us into perceiving there is more dynamic range than there actually is with 16 bits, so we find that adequate. 20 bits would be ideal for a distribution and playback medium, while 24 bits is useful as a working format to have better floating point precision until we dither down to 16 bits for the distribution medium.

[maxpiano]

On this topic and dynamic range, I was testing my home studio monitors today using the test audio files on https://www.audiocheck.net and ... see in particular their https://www.audiocheck.net/audiotests_dynamiccheck.php, try it and then read and the "note" on that page about 16bit vs 24 bit

Yes. Per Bob Katz (among the best audio engineers in the business) in his book “Mastering Audio” our hearing generally has 20 bits of dynamic range. If we compress to 16 bits using dithering, it fools us into perceiving there is more dynamic range than there actually is with 16 bits, so we find that adequate. 20 bits would be ideal for a distribution and playback medium, while 24 bits is useful as a working format to have better floating point precision until we dither down to 16 bits for the distribution medium.

Ok, but wharever Mr. Katz says, have you tried that test yourself? You may be surprised

[sweelinck]

Pink noise probably cannot be dithered and still be perceived as pink noise. Moreover, the test is sensitive to how the digital gain is set.

The authors of the content in that link agree with Bob Katz, per the textual content. To get an apples-to-apples comparison you need to take 24-bit content, dither it down to 16 bits, and compare the two for a decent variety of music content. Yes, 20 bits would be better to capture the full dynamic range. If you acquire some SACDs and an SACD player you can listen to music with full dynamic range. On the plus side, these are mastered with disregard of whether they sound good in car stereos or boomboxes— which may lead to CDs being mastered to exploit less than the full 16 bits.. On the minus side, while they are fine if you just want to sit down and listen to the music with full attention, other uses may lead to your wanting to adjust the volume up during quiet passages and down during louder passages.