2 ms·

Here’s my understanding. Let’s say you want to resample from 32kHz to 48kHz with a filter that has N taps. The sample rate ratio is 3:4, so you need 4N coeffic

Here’s my understanding. Let’s say you want to resample from 32kHz to 48kHz with a filter that has N taps. The sample rate ratio is 3:4, so you need 4N coefficients total. Basically, what you have is four N-tap filters, where filter K has a delay of K/4 samples. With higher values for N, you get lower ripple and lower aliasing. Resampling 44.1kHz to 48kHz, the ratio is 147:160 so you need 160N coefficients. This wasn’t a problem for very long, but it was inconvenient in the 1980s.

segfaultbuserr 6y ago
Why did you need to resample it? Because 48 kHz was already the most common industry standard before CD, is it correct?
- klodolph 6y ago
  That's the way the story goes. I can't really find supporting evidence, though. When I try looking at early digital media, it seems that it was fairly common for hardware to support both 44.1kHz and 48kHz, at the very least. Stuff like DASH, ProDigi, and DAT. I think it's more likely that 44100 was chosen for other reasons--either because it a multiple of the number of active NTSC or PAL lines per second, or because it has a convenient factorization (14x14x15x15), or both. 44.1kHz is a great sample rate for distribution. 48kHz is a little more convenient for creating sounds because it often uses less CPU power to process 48kHz signals (because the transition band is twice as large).