Recent Changes - Search:




Resampler Measurements

All output files are generated from cPlay 2.0b26 (or later).

Both 16/44.1 and 32/44.1 1 kHz input test tones are used:

Upsampling 16/44.1 > 192 (SoX -vLb99.7 on left and SRC 145db right):

A closer look at the bandwidth option using 16 bit test tone (noise floor ~130db):

SoX at 99.7% offers all frequencies up to 22kHz (max is 22.05kHz). SRC offers ~96% and cannot be changed.

Upsampling 32/44.1 > 192:

Downsampling 32/192 > 44.1:

SoX SNR is better than 170db whilst SRC is better than 150db (notice small artefact ~150db above 20kHz).

Phase measurements done using a 2kHz (2 cycles per ms) 24/44.1 test tone. Transient is a sudden volume increase (50db) from -90dbfs to -40dbfs for 5ms:

SRC output at 192:

Both pre- & post-ringing is equally distributed.

SoX output at 192 (minimum phase):

No pre-ringing but greater post-ringing. Overshoot and ringing level is greater than SRC (due to higher rejection db, higher SoX bandwidth and no aliasing). Note that:

  1. SoX non-linear (non-50) phase setting affects HF phase response of original input signal (i.e. a trade-off is made wherein increasingly higher frequencies undergo greater phase shift). Intermediate setting (anything between 0..50) offers better balance regarding this trade-off.
  2. Observed ringing is beyond human hearing when "min(input, output)" is 44.1k or larger.
  3. Ringing effects can be further reduced when "alias" setting is enabled. This results in a trade-off wherein alias artefacts above passband (bandwidth) is added, e.g. for 44.1k input, 96% bandwidth and 192k output, distortion above 21.17kHz occurs. These levels are low (below 100dbRMS) and not in the audible range.
Edit - History - Print - Recent Changes - Search
Page last modified on March 26, 2010, at 06:15 AM