Originally Posted by StanleyB1
...the side effects of upsampling. The padding of the waveform with data that was not present in the original digitally converted audio signal...
This statement reflects a common misconception on Head-Fi about upsampling.
:confused_face_2 : Maybe it's just distorted uses of that word though.Upsampling
- by definition - does NOT
. It merely recreates the samples that the original sampling process would have observed had it been running at a different sampling rate. This is a mathematically exact process (implementation correctness, quantization/clocking issues aside). It can be achieved because an accurate sample of an appropriately band-limited signal uniquely and precisely defines the analog waveform that was sampled.
Or to come from another angle, upsampling can't recreate any information that was lost during the sampling process. It's important to realize that any claims to the contrary, or any deviation from the mathematically exact upsampling algorithm referred to above means the "upsampling" process is (a) no longer purely upsampling, and (b) is making stuff up.
For example, "upsampling" can't determine the samples that would have been made with a different band-limiting filter (e.g. one rolling of at a higher frequency, as you might prefer to use if you sampled at a higher frequency in the first place). The information (if any) that resides in the frequencies that now get past the band limiting filter was removed by the original sample's band-limiting filter and cannot be "reconstructed", "retrieved" or "inferred".
What you MIGHT see in practice is that some so-called "upsampling" processes take shortcuts to simplify the problem, and marketers are more than happy to claim this is "upsampling" when it is not. This situation is implied in the DACmagic marketing material which (IIRC) talks about "upsampling" via "straight line interpolation" and shows how their upsampling process is better. There's only one problem with this - straight line interpolation is NOT upsampling, because the points created by this process do NOT lie on the analog waveform that is defined by the samples that make up the original signal. By definition, using "samples" that don't lie on the waveform introduces errors - i.e. distortion. Whether or not you like the end result (and the interaction of that process with the D/A hardware which may respond better to higher sample rates in some respects) is hard to predict - but this may be the reason some people think upsampling introduces artifacts - because shortcut "upsampling" methods do.