[StratocasterMan]

I was reading about one of the Cambridge Audio products, and I found this information about their 

ATF2™ Upsampling:​

 
http://www.cambridgeaudio.com/content.php?PID=889&COID=329&Title=ATF2%99+Upsampling
 
It seems to me they might be making some pretty dubious claims. They are saying that they can take the digital data contained on a CD, modify that data, and to quote:  "allow the digital signal to closer represent the true analogue sound of the studio mastered audio data."
 
Let me see... take some data, modify it, and then make it closer to the original data??? Aren't they saying that they can take a CD and guess how it should have been mastered in the first place?
 
Wouldn't that be like taking a rock, adding some material to the rock, and then saying you have now made the rock the way it was supposed to be?

 

[eucariote]

2X oversampling is required for the reconstruction filter (that creates smooth analogue waves from digitally encoded signals) or the top 1/2 of the recorded frequency spectrum is filtered out according to Sr. Nyquist.

 

[xnor]

I guess what they mean is that if you connect the samples with straight lines it looks closer to the analogue waveform after they've upsampled and low pass filtered it. I guess I don't need to explain what's wrong with that.

 

[bellsprout]

upsampling is standard, but that article is bile...

 

[Emperortoro]

To be fair, that link is not an article.  It is an overview that talks about bullet points in a general fashion.  This link provides a bit more information on ATF2.  If you are unsatisfied with the degree of detail provided, I'd suggest contacting either Cambridge Audio or the distributor for your country.  I suspect that, like many companies, they don't publish every bit of detail on the web.  They probably have more to share for those that are truly interested.

 

[ultrabike]

Sounds fancy. It seems they are describing an approach to remove "standard" CD  signal jitter, because it somehow matters a great deal.
 
Polynomial based interpolation (usually a Farrow structure) is used in some digital communication systems to close timing in the digital domain. They require higher than fs sampling, a "slip buffer", and use an interpolant (timing error) to close the loop through the Farrow structure. I guess the interpolant here may be the difference between the original clock and the "jitter-free" clock. Interpolation may be cubic, spline, linear etc. Linear usually sucks, cubic and spline approaches usually yield better results. Anyway, in short it seems they are describing an approach to resample the data with a "jitter-free" clock, i.e. removing jitter.
 
I've used Farrow structures in the past to close timing loops and simulate jitter. They may have something in there, dunno.