What is the point of upsampling?

[dc655321]

And, all this started with me wanting to learn a little about "upsampling." Wow. Be careful what you wish for. ; -)

Thar be dragons in them thar rabbit holes!

Thanks for your patience and taking the time to educate (a little) a non-engineer. I've enjoyed it.

I have to say that your willingness to ask questions, attempt to absorb the responses, ask more questions is a nice change of pace around here.
Kudos to you!

 

[bwinlr]

@SIY in another thread pointed to an AES paper and article in Linear Audio by Michael Urwins on his investigation on the reasons behind the 'vinyl attraction'.
https://www.audiosciencereview.com/...nderstand-the-appeal.6430/page-23#post-170710

Here is a link to Urwins' Linear Audio article which included more details than the e-brief. It is a pretty long read at 26 pages, but if you are interested in this topic, it is well worth your time.
https://linearaudio.nl/sites/linearaudio.net/files/v10 mu.pdf

Thanks for the link.

Bob

 

[LTig]

At the very real risk of beating a dead horse, what accounts -- in your estimation -- for the audible difference that I hear between LPs and CDs? It's not my imagination; it's real and apparent. And, I also realize this is very, very subjective, but I love music and listen carefully to it -- it's not background stuff. I have a nice sound system that is highly-resolving musically, so equipment isn't really playing a big role in coloration.

I would appreciate your thoughts on this.

What has not been mentioned yet is frequency response(FR). It's not easy to get a flat FR:

• pickups may have non flat FR
• the capacitive load of a MM pickup and its inductance can lead to wild variations in FR above 8 kHz if not matched properly
• phono preamps may have non flat FR

It's quite a challenge to find a combination of pickup, cable and phono preamp which truly delivers flat FR. If this is done just by subjective audition chances are high you end at a FR similar to a loudness correction - it will sound fuller. If you compare such an analogue chain with the flat FR of a CD the CD must lose.

It's somewhat easier when you use a MC pickup. Here the load capacitance plays no role because the inductance is very low and changes in FR are above the audible range. You just have to match the FR of the pickup and the FR of the phono preamp, and both are usually quite flat to begin with. The disadvantage is he price - both pickup and phono preamp are more expensive.

 

[Dro]

Not quite upsampling, but a very related question regarding this: All serious DACs oversample to allow for an easier reconstruction filter. Yet when we look at the DAC reviews, we see all the filtering work still happening around half of the source sample rate. Why? Does the oversampling require a preceding filter around 1/2 fs?

 

[LTig]

Not quite upsampling, but a very related question regarding this: All serious DACs oversample to allow for an easier reconstruction filter. Yet when we look at the DAC reviews, we see all the filtering work still happening around half of the source sample rate. Why? Does the oversampling require a preceding filter around 1/2 fs?

Yes. The upsampling process creates multiple images above fs/2, starting at fs/2. These have to be removed by a low pass filter at fs/2. However it is easy to implement this filter in the digital domain - much easier than building one in the analog domain without upsampling.

 

[RayDunzl]

we see all the filtering work still happening around half of the source sample rate.

Increasing the sample rate does not increase the frequency range of the existing source data.

If the sample rate is doubled, you will see the original "cutoff" at 1/4 the new sample rate, for example.

 

[watchnerd]

I upsample practically everything I listen because the default setting of the ADC/DAC in my electronics is 24bit/96khz.

I can't lower it, but I can raise it.

 

[bwinlr]

Thar be dragons in them thar rabbit holes!



I have to say that your willingness to ask questions, attempt to absorb the responses, ask more questions is a nice change of pace around here.
Kudos to you!

Well, thanks. And as a big fan of "A Clockwork Orange," I enjoyed seeing your avatar of "Alex," I.e., Malcolm McDowell. Met Malcolm one time at an airport.

 

[DonH56]

Here is one video illustrating the loudness war issue:

 

[Tks]

I almost brace myself for music with DR 6 and such... So annoying and fatiguing

It doesn't even sound half of what real music does. Fucking car radios and now earbuds, pandering to that horseshit.

 

[DonH56]

Not quite upsampling, but a very related question regarding this: All serious DACs oversample to allow for an easier reconstruction filter. Yet when we look at the DAC reviews, we see all the filtering work still happening around half of the source sample rate. Why? Does the oversampling require a preceding filter around 1/2 fs?

A (baseband) sampling system can capture signals from DC to half the sampling rate, fs/2, the Nyquist frequency. Oversampling means increasing the sampling rate without increasing the signal bandwidth; that is the "over" part. If you double the sample rate and the signal bandwidth, then you are not oversampling.

And of course as @RayDunzl said you do not increase the source bandwidth by oversampling.

Oversampling allows greater resolution by moving the quantization ("sampling") noise above the signal bandwidth through some sort of modulation, e.g. a delta-sigma modulator. The modulator does the noise shaping that increases SNR in the frequency band by pushing the quantization noise to higher frequencies. See e.g. https://www.audiosciencereview.com/...igma-delta-digital-audio-converters-dac.1928/

Oversampling does move the Nyquist frequency higher, so the DAC's output image filter can be moved higher in frequency. See e.g. https://www.audiosciencereview.com/...ital-audio-converters-dacs-fundamentals.1927/ for a picture of what the images look like. But that does not get rid of the noise above the signal band; you still need to filter that away.

HTH - Don

 

[kaka89]

I am trying Roon recently and really enjoy the upsampled sound. It sounds clearer and more exciting.

I am using RME ADI-2 DAC so the quality of my DAC should be perfect, but stills I hear a *subjective improvement* with Roon's upsampling feature.

 

[pozz]

@bwinlr Nice to have you here. I've enjoyed the way this thread progressed.

 

[Julf]

I am using RME ADI-2 DAC so the quality of my DAC should be perfect, but stills I hear a *subjective improvement* with Roon's upsampling feature.

Sure. I also hear a subjctive improvement from changing the color or my loudspeaker stands. It all makes for good anecdotes, but that's all.

 

[kaka89]

Well I don't have an ADC so there is no way I could provide an objective measurement.

Technically, Roon's digital filter could be implemented differently, so I won't be surprised the final product is different than RME.
It is basically a different music file after all.

 

[kaka89]

One thing I don't understand is Roon consumes so much CPU power to do upsampling.

If other DACs does upsample exactly the same as Roon, how could other DACs do it with a small chip? It is either DAC chips are highly optimized or Roon did it very poorly.

 

[Julf]

Well I don't have an ADC so there is no way I could provide an objective measurement.

You could arrange for a double-blind ABX test.

Technically, Roon's digital filter could be implemented differently, so I won't be surprised the final product is different than RME.

It could, of course, but I would be surprised if either was so badly implemented that the difference would be audible.

It is basically a different music file after all.

It is a different file, it is the same music (audio waveform).

 

[Julf]

If other DACs does upsample exactly the same as Roon, how could other DACs do it with a small chip? It is either DAC chips are highly optimized or Roon did it very poorly.

DAC chips are indeed highly optimized. They use specialized signal processors, while Roon uses a general-purpose CPU.

 

[LTig]

One thing I don't understand is Roon consumes so much CPU power to do upsampling.

If other DACs does upsample exactly the same as Roon, how could other DACs do it with a small chip? It is either DAC chips are highly optimized or Roon did it very poorly.

Doing it inside a highly specialized hardware chip can be dramatically faster than doing it in software on a standard CPU.

 

[scott wurcer]

Doing it inside a highly specialized hardware chip can be dramatically faster than doing it in software on a standard CPU.

Actually BrutFIR easily outperforms the cheaper SHARK DSP's if you give it most of the CPU.