Oh Chris, I feel honoured to give the one and only real answer. Ahum.
First of all, the term "oversampling" as the result of an activity, is not the same as the result of the term "upsampling". Both activities are the same (or similar at least) : they increase the samplerate. However, with oversampling the objective is something which is in the internals of the DAC, and with upsampling it is about a functional aspect, for you.
When a DAC oversamples 8x, you won't notice it at the ouput rate (is still 44K1 or whatever it was fed with).
When a DAC upsamples, you will notice it at the output, and e.g. 44K1 has become 96K.
What makes all the answers at the (by now derailed) AA thread confusing, is that the various
causes for oversampling are mixed, not mentioned, or not mentioned all (in the one post). Example of this : when one says that oversampling is needed for the (brickwall) filter to operate, the other says that such a filter is needed because of oversampling took place. Both are right but it is not important to the one wanting to know whether he needs an nos-DAC or nor. What *is* important, is why either can be left out (if at all) and which actually causes what to be there/happen.
Yeah, I'm confusing deliberately, but I'll take that away as fast as I introduced it :
Everyting and all is about the type of DAC, which differentiate in two main characters :
1. Ladder oparating DAC;
2. Sigma/Delta operating DAC (or Delta/Sigma, never mind).
The Ladder DAC is the most genuin DAC because (I think) it's principles were used in the first CD players around, by itself because of the logic of the principle (which is dividing by 2 the same way bits for a byte with decimal value; the ladder DAC is also called R2R DAC (R becomes 2R).
An example would be the TDA1541 (chip).
Today many R2R DACs (as a chip) exist, but the 1541 has a specialty : it is so old, that it doesn't contain the logic of the necessity to have a filter built in. Might you be into digital photography, this is nearly the same as an Anti Alias filter in a digital camera; it needs to be there otherwise many photo's will fail (by moiré effects), but when not there, the photo's which didn't capture moiré prone subjects (like the roof of a house) are so much more bright and better.
So, the first CD players contained this lousy DAC without "AA filter". This was subject 1.
Subject 2 is about 16 bits not being enough in todays digital audio world, and 48bits being the kind of minimum needed when mixing/mastering audio. However, R2R DACs with more than (I think) 23 bits do not exist, so they invented another way of, say, emulating the R2R principle. This became the Sigma/Delta (Delta/Sigma) principle, which in fact is about fewer bits (could even be 1 I think) with which a kind of unlimited bits can be virtually created. In your Fireface such a S/D DAC resides, and it can do 32 bits I think.
So this was subject 2. Now we must combine things :
A S/D DAC needs oversampling (aha !) in order to "create" its virtual bits. It will depend on the number of real R2R bits in there - in combination with the number of virtual bits needed in the end - how much oversampling is needed. But it can be 8 or 16 or anything (up to 256 I think). Maybe I saw 4 somewhere.
So we can come to one intermediate conclusion at this time : A Sigma/Delta DAC needs oversampling to operate.
While "upsampling" - that being the same as oversampling for the internals of the DAC - is a nasty thing to happen, because squares become sinuses (the higher the converted (!!) samplerate the worse it becomes) ... the next thing happening is that the oversampling principle *needs* AA filters because of the process happening (never mind the process itself).
Next conclusion : A Sigma/Delta DAC needs AA filters to get the sound back to normal after oversampling.
What's left now is the good old R2R DAC which exist without and with AA filter;
Those with AA filter (which are about all of them), have that filter for one purpose only : the idea of it being necessary to perceive good sound. Of course, one can measure "aliases" when the AA filter is not applied, but as often in audio, things are a tradeoff, and this is - or looks to be one of them : what the AA filter repairs, destroys something else. Why ?
Well, here it comes : in order to let these well thought filters operate, first a good dose of oversampling has to be applied. Roughly it comes down to shifting noise and other anomalies to the inaudible spectrum, do some work there, and then go back to the original sample rate, applying some (necessary) filters again.
Kind of conclusion : Nearly all R2R DACs use oversampling in order to be supposedly better by means of filters for various reasons.
I say "supposedly" because I don't think there's science which can prove it true to be "better".
Last conclusion : R2R DACs allowing for no filtering and not needing oversampling to achieve their number of bits, leave the waves in tact.
The latter is a fact (leaving the waves in tact), and the difference with the R2R oversampling/filtering DACs is in the area of distortion of various kinds. Another fact is that no-oversamping cannot cause additional (Aliasing) distortions. So at least no filters are needed for *that*.
Which type of DAC is best is subjective. But note the big apples and oranges contest, where the filtered DAC measures better, if you only forget to measure the shape of the waves and forget that there might be more squarish waves than sines around.
Also note the huge dilemma : already 24 bit nos-DACs (chip wise) do not exist (not to my own research). I think 23 bits nos chips exist and by combining two you you'll get a 24 bit functional DAC. A functional 24 bit nos (box) DAC existed in the past, but I don't think currently its still in production. Now also note the
reason why you'd want a 24 bit DAC : digital volume control for 16 bit material (but that only allowing for 24dB attenuation without losses, which hardly is enough) and of course the playback of 24 bit material (that coming along with 88K2 or 96K samplerate).
Sigma/Delta DACs can do that easily enough, but they sound lousy compared to R2R DACs already by themselves, and additionally they oversample to get the (bit) job done.
Well Chris, was this something for an answer ?
Peter
