2
$\begingroup$

I've read that upsampling performed in digital music playback can color the sound, produce artifacts, etc. For example, an audio file ripped from a CD might be 44.1Khz/16-bit, and then upconverted to 48Khz/16-bit and played via an optical digital audio output. Audiophiles say this is bad because the signal needs to be "bit perfect" to be reproduced correctly.

Is this correct? My vague knowledge of DSP from grad school leads me to think that all upsampling should do is increase the bandwidth of the signal. But, since the source signal is bandlimited, there shouldn't be anything new in the added bandwidth. And I don't see why the process would color the sound.

What am I missing?

| improve this question | |
$\endgroup$
  • $\begingroup$ What conversion tool did you have in mind? A good result requires a good conversion tool, but if it's done right you'll get all the fidelity you need. If an audiophile can tell the difference, then there's a problem with the audio system, not the data. $\endgroup$ – MackTuesday Apr 30 '14 at 15:57
  • $\begingroup$ I guess the imperfections they refer to come from the filtering involved in the upsampling procedure. However, these artifacts can be made very small if complexity is no issue. $\endgroup$ – Matt L. Apr 30 '14 at 16:30
  • $\begingroup$ So, the specific application is streaming audio via an Apple TV. [network source]-->Apple TV-->Optical audio out-->DAC. Apparently, everything going in to or out of the Apple TV converts to 48Khz/16-bit. I could understand the DAC coloring the sound, but assuming nothing on the digital path is lossy, I don't understand why an ideally-implemented upsampling stage would degrade the signal. $\endgroup$ – Anna Dickinson Apr 30 '14 at 17:01
  • 1
    $\begingroup$ There are no ideal digital filters (that are stable and causal and that can be implemented with finite complexity). They will introduce amplitude and phase distortions. These distortions can be made small by increasing the complexity of the filters. Check this link: en.wikipedia.org/wiki/Sample_rate_conversion $\endgroup$ – Matt L. Apr 30 '14 at 20:45
  • 2
    $\begingroup$ If you specify the degree of "degradation" to as small of an epsilon as you want, the interpolation that computes the new samples in between can accomplish that as long as you're willing to pay for it with computational effort. if your interpolation looks at 64 samples (32 before and 32 after your interpolated sample), no one, including dogs, can hear any degradation. you can implement a pretty damn good brick-wall polyphase LPF with 64 samples. $\endgroup$ – robert bristow-johnson Apr 30 '14 at 21:57
4
$\begingroup$

The up-sampling process will always change the signal in some measurable way. However, if it's done properly the changes are negligible and don't result it any audible difference. Most commercially sample rate converters (hardware or software implementations), do a really good job at this.

Off course, if done badly, upsampling can result in clearly audible signal degradation. I'm not familiar with Apple's implementation but I would assume that they got this correctly.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.