Jayarr: When one looks at a signal at a given spot, it is called "sampling" the data. With digital, these "samples" are taken at very specific but limited points along the data path. One is not taking in ALL of the data, but simply "sampling" it and basing their deductions on averages and trends.
Over-sampling looks at that same data many times i.e. it doesn't look at more data. It does this to verify that that the interpretation of the data actually matches the data that was recovered. As such, over-sampling is more about error correction than it is about increased data recovery.
As far as upsampling goes, this differs from over-sampling. My understanding is that a lower ( standard ) sampling rate provides fewer "check points" of each bit of data. With fewer places along the signal path to extract and compare data, you will suffer from reduced linearity. Reduced linearity results in a loss of information. After all, the less we check up on something and keep track of it, the more that we have to assume about what is going on.
Depending on how and where the data / signal is checked, you might lose resolution on the peaks, between peaks and nulls, etc... By increasing the number of samples taken i.e. up-sampling, there is the potential for less divergence from signal integrity AND an increase in resolution / transparency.
Think of a sine wave i.e. the "hill and valley" looking thing. If we checked the signal at the positive peak ("hill"), in the middle as the signal descends from the peak, at the lowest point of the negative peak ("valley") and back in the middle as the signal ascends again, we could discern quite a bit of information. We would know the absolute amplitude of each peak and the duration of the peaks. From there, we could plot or try to replicate what the signal looked like. If we did this using the data that we recovered from our limited "check points" or sample rates, it would look like a rise accumulating into sharp point falling rapidly to another point that was of opposite or inverted polarity with another rise coming up. Obviously, this looks NOTHING like a sine wave, so we have to guess or "fill in the blanks" in order to approximate what is really taking place. Compared to an analogue signal, which tracks the entire waveform, we've lost TONS of resolution.
By increasing the number of "check points" or sampling rate where data is recorded, we now may be able to discern more of the natural shape of the sine wave. By doing so, we increase linearity by losing less data / having fewer "blank spots" to fill in. We can now tell that the peaks are not necessarily blunt points but rounded mounds. We can also tell that the rise and decay are gradual yet linear radiuses and there are no jagged or abrupt changes in amplitude getting to the top or bottom.
While this explains the increase in "resolution" and / or "transparency" that many experience with a good upsampler, it doesn't explain the difference in amplitudes or dynamic ranges. Or does it?
By producing a higher average signal due to recovering more data, the differences between the positive ( loud ) and negative ( quiet ) passages offers more of an aural contrast. On top of this, we are no longer guessing at how fast the signal rises and falls by filling in the blank spots, we have enough data to know just how fast and where the amplitude of the signal should change at. Hence, the increase in apparent dynamics.
As you mentioned, filtration, bandwidth and transient response all come into play here. Most of this has to do with the analogue section of the Digital to Analogue conversion ( DAC ) taking place. It is also where most of the differences in what we hear with "digital" front ends come into play. By increasing bandwidth, which requires greater speed, and reducing the side-effects of filtering, which also improves transient response ( speed ), there is less blurring or smearing of the signal. Less blurring or smearing with increased bandwidth means better resolution and increased transparency.
Having said that, units and / or formats ( like DVD-A and SACD ) that incorporate higher sampling rates are more likely to recover more of the data. In plain English, this just means that such an approach "lost" less than other, lower samping rates. As such, this is a step forward from "standard" redbook playback.
Units and / or formats that take the side effects of filtering into account and try to minimize / remove them from the signal path are also a step forward. Once again, these are evident in both the DVD-A and SACD formats and redbook players that attempt to do such things.
Each of these on their own are good things. Combining a higher sampling rate with improved linearity due to less filtering / better filtration design can really make a very noticeable improvement. Having said that, redbook players that impliment both of these design advantages are still at a disadvantage compared to either DVD-A and SACD. That is because redbook is limited to working within a given format that is of lower quality to begin with. The other formats offer the potential for much better performance due to building upon a more technologically advanced platform. As it is right now, we are only now "FINALLY" getting to the point where we've learned how to optimize the performance of redbook via manipulating the limitations within this format.
Having said that, the main limitation to any of these formats is in the recording techniques and equipment used to record, mix, master and duplicate the music that we listen to as consumers. As limited as redbook is in bandwidth and "potentially noisy" and "mechanically limited" as vinyl is, i've heard some rather astounding presentations of both. Given that we can do so much with these "limited" formats, i don't think that going to higher resolution playback equipment is the answer or the main problem. That main problem is in the recording / mass production area and you and i won't be able to solve any part of that. That is, unless we start voicing our thoughts / opinions to the record companies / engineers in some manner that they are forced to listen.
Bin: I have an EVS Millennium II in one of my systems. This is an up-sampling design with very limited filtering. My Brother is using an EVS Millennium 1B in his system and my Father is using an EVS Millennium 1A in his system. Both the 1A & 1B are 24/96 units with limited filtering. I have performed some modifications to my unit and that of my Father's, but not my Brother's ( as of yet )... : ) Sean
>
