4 ms·

Ha, I knew where that link was going before I clicked it. That is the go-to video to dispel the "stair step" misconception. It worked wonders for me when I firs

Ha, I knew where that link was going before I clicked it. That is the go-to video to dispel the "stair step" misconception. It worked wonders for me when I first watched it. While many time-machine inventors would go back and kill Hitler, I wouldn't. I would find the first person to draw a "digital sine wave" and handle that problem.

TheOtherHobbes 4y ago
If you really understand digital sampling, there's actually quite a lot wrong in that video. The "16-bit is enough" line is absolutely wrong, because quantisation noise has a very uneven spectrum which is nothing like white noise. It has peaks at certain subdivisions of the sample rate which are quite a way above inaudibility. Which is why dither was invented - to smooth out those peaks. The "44.1kHz is enough" line is also wrong because there's no such thing as a perfect reconstruction filter. (It's mathematically impossible to build a perfect brick wall filter that works in real time.) Real reconstruction filters have trade-offs and the distortions always become more audible and obvious around Nyquist. They're less obvious at higher sample rates. There are also inter-sample peaks which can clip in the reconstruction filter and which are easier to avoid at higher sample rates. Digital also suffers from clock jitter. Early converters were just plain nasty, but these days jitter is barely a problem on cheap converters and not audible at all on pro-grade hardware. None of this is to suggest that digital sampling is stepped and analog sampling is smooth. That's just nonsense and always has been.
rightbyte 4y ago
> I would find the first person to draw a "digital sine wave" and handle that problem. If the rise time is way faster then sample period it looks like a "stair step" though. Most audio chips do some fancy interpolation on the other hand.

TheOtherHobbes 4y ago

If you really understand digital sampling, there's actually quite a lot wrong in that video. The "16-bit is enough" line is absolutely wrong, because quantisation noise has a very uneven spectrum which is nothing like white noise. It has peaks at certain subdivisions of the sample rate which are quite a way above inaudibility. Which is why dither was invented - to smooth out those peaks. The "44.1kHz is enough" line is also wrong because there's no such thing as a perfect reconstruction filter. (It's mathematically impossible to build a perfect brick wall filter that works in real time.) Real reconstruction filters have trade-offs and the distortions always become more audible and obvious around Nyquist. They're less obvious at higher sample rates. There are also inter-sample peaks which can clip in the reconstruction filter and which are easier to avoid at higher sample rates. Digital also suffers from clock jitter. Early converters were just plain nasty, but these days jitter is barely a problem on cheap converters and not audible at all on pro-grade hardware. None of this is to suggest that digital sampling is stepped and analog sampling is smooth. That's just nonsense and always has been.

rightbyte 4y ago

> I would find the first person to draw a "digital sine wave" and handle that problem. If the rise time is way faster then sample period it looks like a "stair step" though. Most audio chips do some fancy interpolation on the other hand.