Sample Rate Conversion and Anti-Aliasing
This link, as long as it works, will take you to a page where you can A/B graphs of various audio products' Sample Rate Conversion effectivity.
Sample Rate Conversion Graphs of various audio products
In each case we see a sine wave sweep, and the blue background shows amplitude by color, where silence would be black, low amplitude dark blue, and normal listening volumes ranging from yellow to white. From blue to yellow goes through magenta, red, and orange, so one can tell where the anti-aliasing artifacts are audible.
Some Examples
DP 6 vs. Logic (Leopard)
DP 6 vs. Ableton Live 7 (not high quality)
Thus we see that DP's aliasing occurs at extremely low (inaudible) levels of less than -140 dB, while Logic dithers a constant low-level noise floor of around -135 dB to mask any aliasing effects. Ableton Live 7 in this example appears not to have any protection against aliasing. The result would be unusable. Ableton Live 7 has a High Quality mode which fares much better, but I chose this one simply as an example of the other extreme in aliasing.
Aliasing and Anti-Aliasing
For those who do not understand aliasing, it is the result of any two different sample rates, frame rates, resolution scales, or other patterns, when superimposed over each other. Thus, two screens of different sizes, superimposed, will produce moire patterns. One often sees movies where the frame rate creates an illusion of bicycle spokes, wagon wheel spokes, hubcap patterns, airplane propellors, or helicopter rotors, which appear to move slowly, or even backward, because the frame rate of the film or video captures each spoke in almost the same position as the previous spoke, or a little bit ahead or behind, thus "freezing" the motion or "moving it" very slowly. In digital audio, two different sample rates superimposed will produce all kinds of phantom sounds, or aliases of the original. The effect grows unbearable as the pitch nears the Nyquist limit -- the sample rate limit that can still define a pitch, usually considered about double the frequency of the sound. Beyond that limit the pitch becomes ambiguous, and aliasing can produce unpredictable results. Thus the need for dithering, which "blurs" the effect by introducing low-level noise. (kind of like blurring the jaggies in a digital picture, where lines that slant relative to the monitor's pixel patterns will produce "stair-stepping." The blurring fills in the stair-stepping with noise in the form of many copies of the jagged steps superimposed between the steps.) Thus
Anti-Aliasing becomes essential for any situation which suffers in quality loss from the superimposition of 2 or more repeated patterns that don't match each other.
Interestingly, our eyes have a built-in anti-aliasing feature called "flicker-fusion." Flicker-fusion operates mainly toward the center of the eye, where most our our attention is focused. The perimeter of our vision does not do this, which causes instant distraction when something moves at the edge of our vision field. Evolutionarily speaking, we are able to react to predators instantly, at the periphery of our vision, whereas that kind of attention to motion would drive us crazy at the center of our vision, and we would not be able to focus on anything. To wit, CRT monitors with low refresh rates can even cause nausea from flicker-fusion reactions, as can fluorescent lighting.
Poor anti-aliasing can introduce several kinds of artifacts into digitally sampled music: aliases (electronic whistling, twittering, etc.), harmonics, comb-filtering, sub-harmonics, and probably others (this is just off the top of my head). Having established that aliasing is the achilles heel of digital recording, and anti-aliasing is absolutely necessary, it is important to know that the tools you are using are doing something about it. One shouldn't just assume that someone else has taken care of it. As can be seen above, one can get into trouble quickly that way. But how can you know? This website (linked at the top of this post) provides us with real tests and what appear to be reliable results. Check it out!
With thanks to user XYZ (that's his screen name here, not a generic reference) for pointing us to this site in
This Thread.
Shooshie