My understanding of oversampling has to be wrong - where have I gone wrong?

[mike7877]

OK, so take the new 4499. It supports 128x or 256x oversampling, and it's configured to run at 128 all the time because THD+n is -8dB worse at 256x.

So we have the 4499EX 7 bit DAC (in the vertical), and we have 44.1 sampled 128x over (by the 4191EQ), in the horizontal, which is another 7 bits.

With oversampling, correct me if I'm wrong here, but we oversample the horizontal to make up for the lack of vertical resolution. The total of the two can't be any better than the total sum of each, because in the end they become the same thing, no?

7 + 7 = 14 bits

With the 4191EX driving the 4499EQ, say you wanted something vertically exactly half way between bit 5 and 6... 64 of the samples would be bit 5, and 64 of the samples would be bit 6. I don't know if the first 64 would be 5 and the second 64 would be 6, if it would alternate between 5 and 6, or something in between, but whatever the case, being able to switch between 5 and 6 only gives you 128 different steps between 5 and 6, making 14 bits.

Right? Or where have I gone wrong?

Also, 44.1 *128 = 5.6448MHz, *256 = double that - DSD rates. So when you send in files over 44.1, you get diminished returns

 

[DonH56]

Sampling is not a linear process. If by "horizontal" you mean frequency and "vertical" you mean amplitude then oversampling extends the Nyquist frequency, and then filtering the undesired upper frequencies out (with or without noise shaping such as delta-sigma modulation) reduces the in-band (audible) quantization noise floor, which in turn can provide a greater SNR and thus more bits. The process is not a linear sum but usually root-sum-square (RSS) and noise shaping alters the noise spectral density so again a simple sum is unlikely to be accurate.

Note the theory applies to quantization noise; distortion and other noise sources can change the results. A simple hand-waving example is that wider bandwidth means more noise, and higher sampling rate often means higher distortion since the circuits do not perform as well at higher speeds (for a variety of reasons, including limited slew, reduced loop gain for feedback, and so forth).

I did not follow the paragraph on the 4191EX so maybe somebody else can help with that.

You should really read some articles, texts, webinars etc. to gain a better understanding of the process. Trying to learn, or explain, such a big topic is almost impossible to do on a forum.