Quote:

... I think I just recognized myself, there!

Actually, I completely agree with the 16/24 argument, but I think that, sometimes, we assume that better equipment makes better recordings, without acknowledging the difference made by the skill of the operator.

In my view, the differences between 16/24 are hard to detect, whilst the skill of the operator makes a huge difference. Some of us who remember working with tape, when you had a noise floor you could measure on the VUs and nothing was tracked without a compressor to preserve the s/n ratio without switching in the Dolbys, are quite happy to stick with 16 bit.




Rog,

Hoping that the analog chops we worked so hard to acquire could be translated to digital audio processing is common to most of us oldsters.
______________________________________________________________

In Linux, audio information is transmitted from one application to another for special-purpose processing, and the bit depth of the transmitted files can be 32 bit, or 64 bit. Files can even be encoded as 32 or 64 bit floating point.
I had not grasped the significance of this until reading Scott's description of "quantization noise" :

"What 24 bit recording does do is give you a better signal to quanitization noise ratio than lower bit depths.

For every bit depth there is an additional 6 dB of SQNR. You can go off and google and wikipedia this to get the complex math but I'll try to keep it simple here.

Here's the way to think about this. Let's pretend you have a pure sine-wave at a fixed peak-to-peak voltage. Let's pretend that we take this signal and we amplify it so that the the peak-to-peak voltage is the same as the peak-to-peak voltage that the A/D converter can handle. You can quantize this into 16 bits, or 2^16 values, or one of 65,536 possible values. The sine wave will have very tiny little stairstepped voltage values into which the wave is encoded.

Now, let's take the same signal, and quantize it into a 24 bit A/D converter, which has one of 16,777,216 possible values. As a result, the stair-steps are WAY smaller with 24 bit A/D conversion than with 16 bit conversion.

The 'stair-steppy-ness' of the signal is the quantization noise. Yet another way to think about this is to think of the smallest possible signal that can be encoded, a signal so quiet that the A/D converter switches between the lowest possible value at zero and the next highest value. It switches back and forth between 0000000000000000 and 0000000000000001. When those values eventually get sent back to D/A conversion, that twitching between the two values tweaks the output D/A and generates an unintended analog output noise: Quantization noise. (For those of you that know binary formats, and the difference between signed and unsigned stuff - bear with me here, just trying to make a point)

For a 24 bit recording, if we adjust the signal down so that the switching back and forth occurs between 000000000000000000000000 and 000000000000000000000001, the lowest two values, and as it goes back to D/A the tweak is much smaller and hence the quantization noise is much smaller. In fact the quantization noise will be roughly 48 dB lower than the 16 bit quantization noise.

Note - no magic with the dynamic range occurs.

The big benefit to recording with 24 bit recording over 16 bit is that one doesn't really have to worry nearly as much about using the full dynamic range of the A/D converter in order to get a nice signal to quantization noise ratio, and the little stair steps that occur in the digitized data are peanuts in comparison to 16 bit.

In other words, you can be quite a bit less careful about it, and just get on with recording."

Thank you, Scott!

just looking for clues...
Oren.
http://www.masteringmatters.com