What is the point of upsampling?

[Tks]

Perhaps I need to learn more about digital sampling itself, but I just can't fathom what upsampling is supposed to do? Like from 44.1 to 192 gets me what exactly in the end?

 

[amirm]

There is no value in upsampling. Best case is that you get what you put in there. Worst case is some loss.

 

[Blumlein 88]

Wouldn't you think with some gear and filters upsampling might reduce imaging artifacts upon playback? I will of course note with proper design and proper filter choice imaging upon playback isn't a problem.

So there should be no value in upsampling, but maybe sometimes with some gear there is the chance of a little value.

 

[Tks]

Figured as much..

So this guy for instance is wasting his time utterly?

http://archimago.blogspot.com/2019/04/measurements-roon-16-upsampling-digital.html


Also, finally, why does this exist in the first place if doesn't do anything. Like audio cables' sonic quality and such are essentially nonsense claims (aside from the realm of IEM's for instance or gear with impedences that will cause a FR to shift heavily with odd cables that have like 9 ohms as some folks have spoke about)..

But with upsampling, it doesn't make sense to me even from a marketing perspective. Like what could you even lie about to sell the practice?

EDIT: Heck, infact, it seems totally detrimental when you go to something like 192..

 

[Blumlein 88]

I would also make this comment. If maybe there are situations where upsampling might make a small difference, it is a very small difference. As in almost anything else could make more difference. People who talk about the night and day difference are full of it.

 

[Tks]

I would also make this comment. If maybe there are situations where upsampling might make a small difference, it is a very small difference. As in almost anything else could make more difference. People who talk about the night and day difference are full of it.

If I may ask, is there any website or source I can visit to hear what imaging artifacts even are? Perhaps even an artifically induced test just to demonstrate what such a thing even sounds like?

 

[sergeauckland]

There may be historic benefits to up or oversampling. Philips's original DACs were 14 bits, 4 time oversampled, which if you do the maths, gets close to 16 bit resolution. Also, when filtering was analogue, it's much easier to avoid cutting into the audio bandwidth with an oversampled signal as the sampling frequency is then four times higher.

With modern DACs and digital filtering, there doesn't seem to be any benefit except marketing.

S

 

[mansr]

Converting a digital (sampled) signal to a continuous analogue waveform requires interpolation to produce the values between sample points. Doing part of this interpolation digitally (upsampling) simplifies the analogue circuitry and gives better results. That's all there is to it. Whether software upsampling is audibly superior to that built into DAC chips is debatable.

 

[solderdude]

The potential upside of upsampling is that the used filter in the upsampling algorithm of the upsampler is 'better' than the brickwall filtering in the used DAC.

When a DAC has a poorly implemented filtering at 44.1 rates this may be audible. When this is pushed to 2x or 4x sample frequency the upside may be improved SQ for 44.1 files and lets face it most recordings are 44.1.

When the used DAC has a good filter inside and proper analog post filtering there should be no benefit.

A potential benefit would be for NOS DACs. In this case the typical 'roll-off' is not there assuming the upsampling algorithm is well implemented.
So for NOS DACs there certainly are benefits (smaller and higher freq. 'steps') but at the expense of the typical NOS roll-off.

Signalist HQ player seems to have upsampling logorithms that may well be better than some of the filters inside DACs acc. to some.

So... for some DACs there may well be audible improvements, or measurable (not necesarilly audible) but certainly not for all DACs .

 

[Tks]

@mansr

Gives better results in terms of what you would say? Accuracy?

Also, what are the limits of upsampling? If it was better, why stop at any value as long as you have something like core i9 CPU’s for example (not sure if upsampling is like video rendering, where it has incredible CPU core scaling, or whether it’s more frequency clocks sensitive).

Also, is this upsampling done in real-time, or can you “pre-render” and simply have a finished upsampled song for instance, ready to be played by hardware that can at least read the file format?

Would that then invalidate the need for expensive DACs that have hardware decoders for these sample rates? Basically concluding that as long as I have enough processing horsepower, and a simple DAC module that is at least EMI shielded in some way, I could do without having to have such an expensive DAC for instance?

 

[Tks]

The potential upside of upsampling is that the used upsampling algorithm used in the upsampler is 'better' than the brickwall filtering in the used DAC.

When a DAC has a poorly implemented filtering at 44.1 rates this may be audible.

So... for some DACs there may well be audible improvements, or measurable (not necesarilly audible) but certainly not for all DACs .

A lot here I need to do more reading about, like filters and NOS.

One other thing I was wondering about. Is there a point of deminishing returns? Like let’s say you want to upsample not 4x, but instead 40x (if such hardware configuration existed that can do such a thing). What sort of measurable differences in terms of scaling occur at those levels. Regardless of audibility?

 

[solderdude]

It is all about reconstruction filtering and the essential analog post filtering being easier to construct (less steep, less ringing and higher above the audible range).
4x above the audible range is already more than enough and easy to build in hardware.
Going higher is not beneficial at all except for those building and selling those devices as well as for the dealers and reviewers/magazines.
It won't be beneficial for us mortals.

You will always encounter people claiming 16x is audibly better than say 4 times. Doesn't mean it is true.

In the majority of cases 2x, 4x let alone 40x upsampling will bring no audible benefits.
Only in some specific cases it might. 4x or 40x fill not change or increase anything but the price.

You would also need to understand that upsampling requires steep digital filtering which also ring around 20-22kHz even when upsampling say 40x.
The filter action may be different than that built inside the DAC and in most cases will not be audible at all.

 

[March Audio]

@mansr

Gives better results in terms of what you would say? Accuracy?

Also, what are the limits of upsampling? If it was better, why stop at any value as long as you have something like core i9 CPU’s for example (not sure if upsampling is like video rendering, where it has incredible CPU core scaling, or whether it’s more frequency clocks sensitive).

Also, is this upsampling done in real-time, or can you “pre-render” and simply have a finished upsampled song for instance, ready to be played by hardware that can at least read the file format?

Would that then invalidate the need for expensive DACs that have hardware decoders for these sample rates? Basically concluding that as long as I have enough processing horsepower, and a simple DAC module that is at least EMI shielded in some way, I could do without having to have such an expensive DAC for instance?

Accuracy? Not really. If the ADC/DAC obeys the Nyquist rules (bandlimited) then your waveform will be perfectly reconstructed without additional interpolation. As mentioned its about making filtering easier, but that has become a bit of a moot point with current DACs.

This video is always worth watching to get the basics of ADC/DAC and busts a few myths along the way.

 

[RayDunzl]

My DAC resamples everything to 211kHz.

I get no choice about it, and I don't ever think about it until some thread like this comes up.

 

[daftcombo]

What about a dCS upsampler?
https://www.hifinews.com/content/dcs-scarlatti-upsamplerdac-£6500£12000

 

[Theo]

Maybe there would be a difference if you possess a quite poorly designed DAC?

 

[Tks]

Accuracy? Not really. If the ADC/DAC obeys the Nyquist rules (bandlimited) then your waveform will be perfectly reconstructed without additional interpolation. As mentioned its about making filtering easier, but that has become a bit of a moot point with current DACs.

Thank you very much March, I'm going to watch right now!

Btw, how're you and your projects going, last I seen you were conjuring an amp for us :}

 

[Krunok]

I would also make this comment. If maybe there are situations where upsampling might make a small difference, it is a very small difference. As in almost anything else could make more difference. People who talk about the night and day difference are full of it.

The difference can come only from the information that was lost/distorted in the process of upsampling. No new data can be created during upsampling, you can only loose some in the process.

 

[pkane]

The difference can come only from the information that was lost/distorted in the process of upsampling. No new data can be created during upsampling, you can only loose some in the process.

The point of upsampling is to move the reconstruction/anti-aliasing filter well above audible frequencies, where it can be more gentle and not affect the audible spectrum. A high-quality filter at 22khz is hard to make in hardware without distorting phase and without cutting into frequencies below 20khz. A gentle filter at 88khz is much easier, and it can distort there all it wants without affecting the audible range. It's a simple engineering solution to a problem that could otherwise become audible... at least to some of us (not to me, not for a while )

 

[AnalogSteph]

Software resamplers in general can be useful when hardware sample rate support is limited and the quality of whatever resampler is provided by hardware, drivers or the sound stack is less than ideal. Those who have fought with last decade sound hardware will probably know what I mean. (Random tidbit for Asus Xonar users: Looks like those or their Windows drivers may only be able to upsample but not downsample.)

In software you can afford throwing substantially more computing power at the problem, and delay is much less of a concern in playback of canned material. Hence you can achieve lower (periodic) passband ripple and better filter ultimate rejection. For example, your average consumer grade DAC will have a reconstruction filter with an ultimate rejection of somewhere around 60 dB, and you'd need a top-grade part for 100+ dB. I don't remember what something like SoX VHQ will do, but 120+ dB for sure - with negligible CPU loading on processors close to 15 years old if optimized well.