Brian's post shows harmonics that are down 96 dB. You should NEVER expect a 16 bit representation to to better than that because that's the total dynamic range available to you for representing a signal.
I think it is important to note that we are not just concerned with one single frequency when recording music. We will be recording a wide range of frequencies, of which many will be harmonically related to the sample rate. So I would think that dithering, which prevents these harmonics from occurring, seems to be beneficial.
As far as the level of the harmonics, they are buried below -96 db, as are the highest levels of noise for even the most severely noise shaped spectra. So wouldn't this imply that truncation vs dither should be inaudible? I would guess the differences ARE audible, so there must be something going on with dither comparisons besides simply the noise floor. 
Just thinking out loud here.
Wait a minute... We need to be careful to understand that these harmonics are the result of the repeated appearance of exactly the same quantized samples, cycle after cycle. That is to say it only happens when exactly the same waveform appears repetitively and the period of that waveform is exactly an integer multiple of the sampling frequency. Show me where that ever happens in live recorded music. Even when a band uses a synthesizer to produce a single, continuous tone, there is undoubtedly enough ambient noise that exactly the same samples do not appear on each cycle of the recorded waveform. So in real life, these harmonics caused by quantization will never be seen. In real life, other real sounds prevent the repetitive occurence of a unique sequence of samples. (This is also the reason that you don't get much compression when you zip a .wav file. The algorithm used by pkzip, or WinZip was developed by a guy named Phil Katz. Phil's algorithm takes advantage of the repetitive patterns that exist in most files in order to store them more efficiently. There just aren't that many repetitive sequences in a .wav file with any reasonable bit depth.) In real life, other sounds perform the function that dithering does on pure synthesized tones, but without adding noise. My premise is that if the problem doesn't exist in real life, why add noise in order to cure a non-existent problem?
Again, I'm willing to keep an open mind here, but it's going to take some compelling evidence to sway me. I'd like to see real world examples and evidence that you are thinking about what's really going on. I'm not easily swayed by marketing hype written by someone whose intent is to get you convinced that you need the latest equipment unless I can also verify for myself that what they say is true. Unfortunately, I think that the taper community needs a good, healthy dose of scepticism when it comes to new technology. We seem to have lost the ability or the desire to understand the subtle nuances of the technology we use. Without careful scrutiny, we are unable to tell the difference between snake oil and miracle cures. Worse, it appears to me that dithering is an area where we're willing to buy a cure for a disease that does not in fact exist in the wild.
Topic: 24Bit dither to 16bit vs. recording just 16bit (Read 9694 times)