What is the point of upsampling?

[Julf]

Do it through measurements. No need for snarky remarks and the paranoid attitude.

Here's looking at you, babe.

That digital transmission and decoding/conversion is perfect. Or even audibly prefect.

Who here has claimed that? What we are saying is that properly designed digital transmission and lossless encoding/decoding does reproduce the exact bits that went in - or fail badly, resulting in dropouts or no sound at all. The bits don't change in subtle ways, just like your spreadsheets or downloaded programs don't get small errors from being transmitted, so from that point of view bits are bits.

 

[tuga]

Who here has claimed that? What we are saying is that properly designed digital transmission and lossless encoding/decoding does reproduce the exact bits that went in - or fail badly, resulting in dropouts or no sound at all. The bits don't change in subtle ways, just like your spreadsheets or downloaded programs don't get small errors from being transmitted, so from that point of view bits are bits.

The data doesn't change. But reconstruction is not perfect.

 

[Julf]

The data doesn't change. But reconstruction is not perfect.

That's a different story. The bits are still bits. What goes in comes out - as ling as it is bits.

 

[Holmz]

The first DAC I designed was back in the 70's (R2R ladder), but who is counting...

Did you play stairway to heaven on ^it^.

Or was that on vinyl?

 

[Julf]

Did you play stairway to heaven on ^it^.

Or was that on vinyl?

The source was digital - software-synthesized sounds and music. It often took the whole night for the computer to produce 60 s of music. I am glad we have slightly more computing power these days.

 

[earlevel]

This question is best asked by professional DAC developers, because they are the ones who use upsampling for some purpose, to implement filters for example.
Since there are no DAC developers here, the answers will be varied (to say the least).

Isn't this a little like saying that only people who design car engines understand why car engines have certain design traits? Not even a seasoned mechanic truly understands car engines, because they don't actually design them, but just use existing components?

Most of us don't design car engines—or DACs—because there is no need to, there are competent ones already out there, and it's pretty costly and time consuming to design and build a new one.

I'm sure that a lot of people on this board understand DACs very well, and can answer questions from people who don't understand them quite so well.

I think your main point was likely more to the fact you'll get varying levels of knowledge, and in some cases perhaps people won't understand the details as well as they think they do. But that's true of any topic in a public forum, including how to cook a rib roast. I haven't read the thread, but I suspect that someone who started out with no idea of what sense it makes to raise the sample rate, post-sampling, probably has a little better idea. They are not people who are likely to design their own DACs, so they are probably comfortable with a little idea of the reasoning. That is...I doubt they need an explanation from someone who designs DACs.

 

[bhobba]

To answer the original question, the point of upsampling is when you put, say, 48kz material into a DAC at 48kz, you get a cascade of reflected images at higher frequencies of the recorded material theoretically forever. Some don't worry about it, and you have what is called a NOS DAC that some like. But since the frequencies go on forever, it can cause problems with amps, etc. So what you do is upsample it. A straightforward process is duplicating the samples to get 96k, 196k, etc. It is called 2, 4,.... etc oversampling. Eight times, oversampling is common. Then, those cascaded copies of the signal occur at a much higher frequency that is easy to filter without affecting the audible band. DACs often do that for you. However, some believe a better quality filter (technically with many taps) affects the sound quality. Rob Watts is a proponent of that with his M-Scaler; there are plenty of reviews on the device, and Rob has videos on it and how he designs DACS. First, how he designs DACs with massive upsampling:

Then, on the M-Scaler:

I like the M-Scaler, but others, some even more experienced Audiophiles than me, don't.

I know a DAC designer who solved the issue elegantly. He uses a standard DAC chip and the best quality Sowter Audio Transformer he can buy as output. That is virtually flat to about 70KHz and rolls off rather steeply after that, preventing any high-frequency stuff from being output. I have compared his DAC to much more expensive DACs like the Direct Stream, and IMHO, it is better. It is not better than a Chord TT2, but that is MUCH more expensive and not better than the even more costly Grandinote DAC - but they are the only two better DACs I have found.

Of course, better here means to my ears sounds better - if it is better, it depends on what you mean by better.

Thanks
Bill

 

[AnalogSteph]

I know a DAC designer who solved the issue elegantly. He uses a standard DAC chip and the best quality Sowter Audio Transformer he can buy as output. That is virtually flat to about 70KHz and rolls off rather steeply after that, preventing any high-frequency stuff from being output.

Mind you, if you just need a lowpass of a decent order and nothing else, I can think of substantially less costly ways of implementing one... two Sowter transformers by themselves are about as much as an entire rather competent DAC (e.g. Topping E50). I'd rather be throwing a few passives and an opamp or two at the problem.

 

[Philbo King]

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?

Upsampling is a good way to determine 'true peak' values of intersample overs.

 

[Killingbeans]

I like the M-Scaler, but others, some even more experienced Audiophiles than me, don't.

Others just acknowledge that it does nothing audible:

CHORD M-Scaler Review (Upsampler)

This is a review, listening tests and measurements of the CHORD Hugo M-scaler upsampling digital transport. It is on kind loan from a member and costs US $5,295 (some show it as $5,650). We have the usual lightshow of buttons as other CHORD digital products but here, due to less overloading...

www.audiosciencereview.com

 

[bhobba]

Mind you, if you need a lowpass of a decent order and nothing else, I can think of substantially less costly ways of implementing one... two Sowter transformers by themselves are about as much as an entire relatively competent DAC (e.g. Topping E50). I'd rather throw a few passives and an opamp or two at the problem.

Yes. It's how he differentiates his DACs. He once used a diamond transistor (an IC that implements a 'ideal' transistor look alike, but the current can flow both directions between base and emitter) output stage that sounded good to my ears. But he now prefers the transformers - maybe just for marketing. His DACs are cheap enough, anyway. We all know a lot of this stuff is just marketing. The Direct Stream does it that way but upsamples to DSD. I guess his marketing ploy is we do essentially the same as the direct stream but at a much cheaper price. I don't know what is happening, and I have a direct stream, but his DACs sound better. I don't use it because I directly connect to my amps, and his DAC requires a pre-amp. Plus, the difference isn't much.

Thanks
Bill

 

[bhobba]

Others just acknowledge that it does nothing audible:

Oh, the claims and counterclaims.

All I can say is having organised blind listening tests, especially where expensive equipment and resulting egos are on the line, is a royal pain. I remember one where a DAC had a different earthing path - straight away, the 'losing' DAC makers said that invalidated it with some extra agro thrown in for good measure. Two good friends were at each other's throats for a while - they calmed down and are back to being good friends.

Now I don't bother organising them, I suggest people shut their eyes while a reliable friend switches the component in and out - after matching volumes, of course.

Thanks
Bill

 

[Killingbeans]

I don't know what is happening, and I have a direct stream, but his DACs sound better.

Probably just some impedance issue f¤¤king up the frequency response in a way you happen to enjoy.

Now I don't bother organising them, I suggest people shut their eyes while a reliable friend switches the component in and out - after matching volumes, of course.

Was going to say that's better than nothing, but honestly it probably isn't. High risk of getting a false sense of security.

I'd say there's no need to bother with even that. For most people this hobby is all about wonder and waffling. Trying to introduce reality into the mix is just deeply insulting.

 

[AnalogSteph]

I really have to drag my Latitude E6330 over to be measured one of these days. I swear it benefits audibly from upsampling (or rather starts to sound normal... it's mostly there in double speed, quad speed is a tad better). Either there's something screwy going on with the drivers (which is quite possible, they're old Win8.1 ones on Win10), or the ultrasonic aliases are messing with the output stage or the digital filter is terrible or something. I have no reference for how an IDT 92HD93 should perform. My E6510 does not have that particular issue but also uses another chip (92HD81B) and is still on Win7.

 

[AnalogSteph]

My E6510 does not have that particular issue but also uses another chip (92HD81B) and is still on Win7.

Silly me. E6520 and 92HD90, respectively.

 

[Bebelalu55]

Hello, I just joined the forum today.
Sorry if my question has already been answered, but there is so much to catch up with!

When upsampling, does the filter adjust itself to half of the incoming frequency?
I assume this is the case, otherwise there would be no benefit.
I am puzzled because the pretty filter curves always show a cut off point around 20K for a 44.1 or 48KHz sampling rate (e.g. SMSL SU6).

 

[DonH56]

Hello, I just joined the forum today.
Sorry if my question has already been answered, but there is so much to catch up with!

When upsampling, does the filter adjust itself to half of the incoming frequency?
I assume this is the case, otherwise there would be no benefit.
I am puzzled because the pretty filter curves always show a cut off point around 20K for a 44.1 or 48KHz sampling rate (e.g. SMSL SU6).

Upsampling cannot add more to the music, but can reduce the noise, if you keep the signal bandwidth the same when you upsample. That increases the SNR (signal to noise ratio).

Upsampling does not add any new information to the signal; it cannot make something out of nothing, and the information band is set by the input bandwidth. There is no signal information above the incoming band to be added by upsampling. Increasing the output bandwidth by upsampling just adds noise if you do not filter it.

What upsampling can do is to spread the quantization (conversion) noise over a wider bandwidth. The overall noise from quantization is the same for given resolution no matter the bandwidth, it will be about 6N where N is the number of bits for any sampling rate. Think of all the noise in a 22 kHz band for a 44 kS/s DAC, then double that rate to 88 kS/s, and the noise now falls in a (doubled) 44 kHz band. By keeping (filtering) the bandwidth at 22 kHz, you can eliminate (filter away) one-half the noise, and thus improve SNR by about 3 dB.

Here is the original picture:

The quantization noise floor (dashed red line) is all within the 22 kHz signal band, so all the noise (grey box) is in the same 22 kHz band, and the filter affects signal and noise.

Next look at the picture when upsampled:

The signal bandwidth is not changed, so keeping the filter at 22 kHz does not affect the signal. However, the quantization noise (dashed red line) now goes to 44 kHz, and is at half the level it was before. When you filter at 22 kHz, the noise above 22 kHz is removed, so the remaining noise (grey box) is halved.

No change to the signal, but reduced noise, if you keep the same signal bandwidth.

With most DACs providing SNR well beyond what we can hear (use), you cannot hear the SNR improvement, so in the real world upsampling offers little to no benefit. It may actually harm the signal, because faster sampling usually increases distortion.

HTH - Don

 

[antcollinet]

Upsampling cannot add more to the music, but can reduce the noise, if you keep the signal bandwidth the same when you upsample. That increases the SNR (signal to noise ratio).

Upsampling does not add any new information to the signal; it cannot make something out of nothing, and the information band is set by the input bandwidth. There is no signal information above the incoming band to be added by upsampling. Increasing the output bandwidth by upsampling just adds noise if you do not filter it.

What upsampling can do is to spread the quantization (conversion) noise over a wider bandwidth. The overall noise from quantization is the same for given resolution no matter the bandwidth, it will be about 6N where N is the number of bits for any sampling rate. Think of all the noise in a 22 kHz band for a 44 kS/s DAC, then double that rate to 88 kS/s, and the noise now falls in a (doubled) 44 kHz band. By keeping (filtering) the bandwidth at 22 kHz, you can eliminate (filter away) one-half the noise, and thus improve SNR by about 3 dB.

Here is the original picture:

View attachment 368178
The quantization noise floor (dashed red line) is all within the 22 kHz signal band, so all the noise (grey box) is in the same 22 kHz band, and the filter affects signal and noise.

Next look at the picture when upsampled:
View attachment 368179
The signal bandwidth is not changed, so keeping the filter at 22 kHz does not affect the signal. However, the quantization noise (dashed red line) now goes to 44 kHz, and is at half the level it was before. When you filter at 22 kHz, the noise above 22 kHz is removed, so the remaining noise (grey box) is halved.

No change to the signal, but reduced noise, if you keep the same signal bandwidth.

With most DACs providing SNR well beyond what we can hear (use), you cannot hear the SNR improvement, so in the real world upsampling offers little to no benefit. It may actually harm the signal, because faster sampling usually increases distortion.

HTH - Don

But you can't reduce the quantisation noise that is already in the file which has already been quantised. That is there embedded in the upsampled data. All you can do is reduce the new quantisation noise that is added by the DAC DSP processes.

 

[DonH56]

But you can't reduce the quantisation noise that is already in the file which has already been quantised. That is there embedded in the upsampled data. All you can do is reduce the new quantisation noise that is added by the DAC DSP processes.

That depends upon the upsampling algorithm and DAC resolution.

 

[antcollinet]

That depends upon the upsampling algorithm and DAC resolution.

It doesn't. Once the noise is embedded in the data, there is no way of separating it from the signal - any more than you could reduce tape hiss from a digital recording from a tape.

There was a demo by @danadam in another thread. Hang on...

...here it is:
https://audiosciencereview.com/foru...4-1-collection-a-good-idea.53641/post-1954662