I agree with you that 96KHz is a good choice.
24 bits? That is not the case.
First, 24 bits means 144db dynamic range, that is not the case.
Second, almost all manufactures specify the dynamic range with A weighting, which adds a couple or more dB to the specifications.
Third, the dynamic range specification is only one thing. The various distortions specs are very importent (not just a 1KHz sine wave).
Also, many applications require low latancy. It takes a lot of parameters to specify a converter well.
But I agree with you that 96KHz is the best choice for AD. I avoided making AD with 192KHz. The market forced it, so I did it on my last gold AD.
The dynamic range is 126dB (20-20zKH no A Weighting), thus true 21 bits performance. The distortions are around %0.0003, very low latancy.
It was not easy...
If a relevant portion of a recorded track was at half of full scale, and the mixer wanted to digitally amplify it at full scale, a 16-bit peak sample would have only used 8 bits, which would now be scaled up to full scale. That channel would be limited to 48dB S/N, it seems to me, and that bit of gain-riding could lead to noise pumping or noticeable gaps in smooth crescendos and dimenuendos, depending on how it was done. Yes, that’s an extreme example, but I’ve seen worse moves during mixing in post, sometimes from big changes in EQ. If the sample had 24 bits of depth, that scaled-up sample would start with 12 bits—a big difference.
(Yes, all the surrounding hardware may do no better than 20 bits, but the mixing move is done in the digital domain these days. My Yamaha pro-audio digital PEQ does an AD on the way in and a DA on the way out, and at 16/48. But it’s used in a playback chain—different requirements.)
Are any of these effects audible? Depends on the mix, and whether other content masks it. But my idea for recording in 24 bits is to allow for those subsequent large adjustments. Once those are made, I downsample it for playback. Recording digitally from live analog recordings or directly from live performance seems to me a different use case that leads to different requirements.
For recording straight from analog sources that won’t be remixed just to make a digital version, I would agree 16 bits is fine and probably 20-30 dB better than the source anyway.
And maybe I’m wrong, but sampling the signal at a higher bit depth than its analog performance adds about 50% to the file size compared to 16 bits, which may be less than the performance of analog sources especially when heavily manipulated. Low cost for mitigating a low risk, perhaps.
Rick “has had to gain-ride low tracks from field recordings more than once” Denney
) but in digital it will be capped at quantization level.
