Looks like Rob Watts is claiming that his WTA Filter is better than filters that preserve the original data(aka Schiit Closed form filter)?
And this:
How does a DAC know what else to reproduce other than the data it is given(e.g. garbage in, garbage out)??? Unless Rob Watts have some kind of method/maths which compensates for analogue to digital converter's signal loss? This sounds like MQA type of solution.
http://www.audiostream.com/content/mqa-ltd
There would actually be two possible ways:
1) As you suggested, it's
theoretically possible that the DAC could "know about" and "compensate for" specific errors that were caused by other components when the audio signal was encoded. While this sounds like a great idea, it usually falls down in practice - first, because some types of errors simply cannot be corrected perfectly, and second, because being able to do so relies on knowing a lot more about the original signal chain than we usually do.
2) It's perfectly reasonable to claim that simply avoiding causing any additional errors contributes to creating a more perfect reproduction of the original.
There's a sort of "option 1a" that entails making good guesses about problems, and then making alterations based on the assumption that they are present, and hoping that the end result is closer to the original than what you started with. A perfect example of this is the software used to "recover missing detail" from pictures. If you have a picture, taken with a telescope, which shows a bunch of blurry little white blobs, and some short parallel white lines, since you know that what you expected to see were a bunch of tiny white points, you can
assume that the blurry dots were supposed to be stars but they are a bit out of focus, and that the short white lines were created when the telescope failed to remain still and so smeared similar dots in a single direction. You can then calculate a mathematical correction that will get you remarkably close to what was there to begin with. However, this all relies on the assumption that you're looking at a picture of stars.
You can use that same or similar software to "sharpen" a picture of something else, such as a human face, or a license plate number. You can even base some of your assumptions on the way in which pictures tend to get blurred when a camera isn't perfectly focused. This will give you a "pretty good guess" that sometimes produces remarkably good results. However, it also sometimes produces bad results, because your assumptions aren't always true. (Modern software can even be written such that, assuming you are hoping to make a license plate number readable, the software can "detect how well it worked", and even adjust its operating parameters accordingly. This would allow it to try different settings, and finally use the one that produced a result that was closer to what it
expected or "
hoped for". However, in reality, it's still a guess.) To take the extreme example, if I was the photographer, and I
DELIBERATELY shifted the picture out of focus, then your assumption that it should be sharp is wrong, and, even if you could do so perfectly, making it sharp will "destroy" it.
Personally, I would leave anything that deliberately alters the signal in the "mastering process". (If I was remastering a CD, and I happened to
know that it was converted with a specific brand of A/D converter, and also had a way to correct the specific errors introduced by that encoder, then I would so so.... although, even then, if my correction process generates other new errors, I have to decide whether my new version is really "better" or not.)
(This question comes up frequently in legal cases. If I start with a fuzzy blob that's supposed to be the bad guy's face on a security video, with enough signal processing I can probably "sharpen" it to the point where it looks like a human face. However, can I trust it to look like the right face? Or did my software do such a great job that it essentially created a face from insufficient information, in which case who it happens to look like is almost purely random? Or, even worse, does it offer so many options that, if I keep trying different settings, I can produce a result that looks like whomever I want it to - at which point I cheerfully declare "that's the guy" and stop trying new options?)