Upsampling 16/44.1 collection a good idea?

[terryforsythe]

To summarize, the discussion between antcollinet and myself began with the following post by me:

My understanding is that the primary benefit of oversampling is to improve the effective SNR with respect to quantization noise. Specifically, dithering is used to redistribute the quantization noise over a wider frequency spectrum extending well above the audible frequency range, and then the higher inaudible frequencies are filtered out. The audible difference primarily is detectable in quiet passages where the music signal is low, for example a decaying piano tone. See, e.g., https://science-of-sound.net/2016/01/quantization-noise-and-bit-depth/

That understanding was reached by researching the subject and piecing together various pieces of information - this thread peaked my curiosity in this subject. Admittedly, this subject can be a little confusing, and it took careful reading and analysis of the various references to reach my understanding.

antcollinet replied with the following:

But that only works at the adc stage, or when downsampling. It is not (if I understand correctly) possible to shape the quantisation noise in an existing quantised signal by upsampling. It is already baked in.

This appears to me to be the underlying issue in the back and forth discussion between antcollinet and myself, as well as the discussion between danadam and myself.

As noted, this subject can be confusing, and I understand why it can be hard to grasp. But, with further research I found the paper "Designing and Evaluating a Delta-Sigma DAC for Hi-Fi Audio" (https://liu.diva-portal.org/smash/get/diva2:1745070/FULLTEXT01.pdf). That paper ties everything together in a single reference, fully disclosing using dithering in an audio DAC to perform noise shaping in order to redistribute the quantization noise over a wider frequency spectrum, and then filtering out the higher inaudible frequencies. The result is a reduction in the quantization noise in the audible frequency band, performed by the DAC.

I understand that danadam performed an oversampling test with results that do not correlate to the above paper nor the other references I cited. danadam did not post which DAC and/or oversampling algorithm was used for the test. Based on the results, though, it appears that whatever algorithm was used for the test probably did not implement noise shaping as discussed in various references that were cited.

At this point in time I have nothing further to contribute to this subject.

 

[antcollinet]

That dithering can only be applied at the time of quantisation - IE during the ADC

You need to add yor emphasis also to the words *for the original file”

what i was pointing out is once the signal has been quantised for that file (using dither at the time of quantisation) that noise is now added to the signal in that file. you cannot subsequently use dither to alter that noise which is already there.

please don’t quote me further if you are going to misrepresent what i said. Im not particularly interested in having to correct you each time.

 

[terryforsythe]

what i was pointing out is once the signal has been quantised for that file (using dither at the time of quantisation) that noise is now added to the signal in that file. you cannot subsequently use dither to alter that noise which is already there.

Read the last paper I cited. It discloses doing exactly what you say cannot be done.

 

[KSTR]

Read the last paper I cited. It discloses doing exactly what you say cannot be done.

You are using it out of context. The context is quantization noise etc in the Delta-Sigma Modulator itself.
But at the stage of upsampling the input stream prior to the DAC proper (the thread topic) it is fully out of context and does not apply. That does not conflict with the notion that after upsampling you will need to re-dither the signal when transforming it back into the integer domain to avoid additional quantization noise and distortion here, just like with any other manipulation of the sample stream, audio processing 101. And for that dithering, you now can use the added bandwidth and even apply noise shaping.
But you cannot improve what's already baked in into the input stream.

 

[antcollinet]

Read the last paper I cited. It discloses doing exactly what you say cannot be done.

No, it doesnt.

 

[Brian Hall]

LOL.... there are always two groups of extreme people...

1. MQA Hater vs MQA Supporter
2. Upsampling Supporter vs non-Upsampling Supporter
3. Tube Amp vs Class-D Amp Supporters

Too many in the audiophiles world. I think it is interesting to have such (provided that all the discussions are fact based, healthy, meaningful)

1. MQA is a scam and should be shunned.
2. DACs are great at doing any upsampling they need for great filtering. No point in doing it ahead of time. Programs like HQPlayer are an unneeded waste of money.
3. Tube amps are old overpriced outdated tech for people who don't like accurate reproduction and want to have their audio distorted.

 

[Purité Audio]

1. MQA is a scam and should be shunned.
2. DACs are great at doing any upsampling they need for great filtering. No point in doing it ahead of time. Programs like HQPlayer are an unneeded waste of money.
3. Tube amps are old overpriced outdated tech for people who don't like accurate reproduction and want to have their audio distorted.

Yup absolutely , although some valve amps may not add audible distortion but their owners ‘believe’ they do.
Keith

 

[Brian Hall]

Exactly, you made a very good point.

It does make a diference to the musical experience (at least to me).

Placebo effect. You believe it does something so your brain thinks it does.

If it makes no difference, who would waste time and money for doing it? agreed?

People who incorrectly believe it makes a difference because of the placebo effect.

If you tried it (and really try to understand and feel it open heartly) and you still don't like it, it is ok, YMMV.
However, if you didn't try, I bet you may miss one of the easier and cheapest way to enhance your musical experience.

Upsampling before the DAC does nothing that can improve the final audio unless you have a bad DAC that is incompetent at filtering.

 

[KSTR]

In that case, IMO, dithering does involved in the DAC process. Am I correct?

Yes. In a "classic" 1-bit Delta-Sigma DAC, dithering is taken to the extreme, it's the underlying working principle.
What was linear pulse code modulation (LPCM) at a low sample rate at the input is converted into some sort of semi-random pulse density modulation (PDM) at a very high sample rate.
EDIT: In modern multibit D/S-DACs we have a combination of both, a low-resolution LPCM DAC running at high rate and putting out heavily dithered data to arrive at the desired precision.

 

[Brian Hall]

Just wondering do you have any objective proof or evidence to support your subjective claim?

If you have any objective evidence to show that there is no way to further improve the exisiting solution for the DAC process, I think that is something all of us would like to see and then no more discussion for this topic is needed, IMO.

Simple logic. There is no point in doing something ahead of time that good DACs are already great at doing as part of their normal operation.

From Schiit Audio (they make some great DACS):

"Upsampling involves increasing the sample rate of audio before it reaches the DAC. Now, some folks believe that upsampling can enhance the audio quality by reducing artifacts and improving the overall sound reproduction. But here's the Schiit truth: it's a bit of a controversial topic.

You see, modern DACs are quite capable of handling different sample rates without the need for upsampling. They have their own internal filters and processes to ensure accurate and faithful audio reproduction. Upsampling may introduce additional processing and potential artifacts, which can be a bit counterproductive.

Now, don't get me wrong, there are cases where upsampling can be beneficial, especially if you're dealing with poorly mastered or low-quality audio files. Upsampling can help in smoothing out some rough edges and improving the overall listening experience. But for high-quality audio sources, the benefits may be minimal."

So if you believe you are getting any benefit by doing upsampling before sending the data to your DAC, it must mean you listening to low quality files or ones that were badly mastered.

 

[Purité Audio]

Just wondering do you have any objective proof or evidence to support your subjective claim?

If you have any objective evidence to show that there is no way to further improve the exisiting solution for the DAC process, I think that is something all of us would like to see and then no more discussion for this topic is needed, IMO.

No one to my knowledge has ever published any properly conducted comparisons where HQ player has been reliably picked including the developer of the software.
Keith

 

[Brian Hall]

I think that is something all of us would like to see and then no more discussion for this topic is needed, IMO.

People have been telling you there was no need for this for 16 pages already. Upsampling ahead of time is a waste of time and does nothing to improve the audio. You want to believe it helps so you think it does.

It is the same thing with Roon's upsampling that can be applied before sending audio to an endpoint. It does nothing to improve the end result, but some people are fooled into thinking they are getting better sound by seeing bigger numbers on their DACs display.

It is the same thing with "HiRes" audio files. Just another scam.

 

[KSTR]

DACs are great at doing any upsampling they need for great filtering.

Upsampling (to an integer multiple, say N) is trivial, just insert N-1 zero sample after each sample. The filtering is hard part and most on-chip DAC filters are compromises wrt to attenuation at fs/2 and beyond, and passband ripple (creating pre-/post-echos, not to be confused with the pre-/post-ringing of the impulse response).

 

[melomane13]

I believe most people in ASR would consider this as your own personal opinion.

To me, I am more than happy to see people to share their own personal experience / opinion but looks to me some of the fellow members may have a different viewpoint about it as they believe these personal opinion is no use here. Good luck.

it is not an opinion but a fact that there is no audible difference between 16/44 and 24/96

 

[Brian Hall]

I believe most people in ASR would consider this as your own personal opinion.

To me, I am more than happy to see people to share their own personal experience / opinion but looks to me some of the fellow members may have a different viewpoint about it as they believe these personal opinion is no use here. Good luck.

A clarification. By "HiRes" being a scam, I am referring to bit rate and sampling rates above lossless CD quality 16/44.1.

24/96 and 24/192 "HiRes" are pointless for playback.

I understand the reasons for recording at higher resolution, but mixing down for playback to CD quality is all that is needed as it captures everything humans can possibly hear.

 

[Brian Hall]

Thanks for your personal opinion (please correct me if I am wrong with the objective proof / evidence as I may miss them).

Let's assume two cases in your proof:

1. Yes, you do have the objective proof / evidence. That's good, then your claim is not an opinion as you said, it is a fact then.
2. No, you don't have any proof / evidence. Then, your claim would be classified as personal opinion.

Agree?

Let's assume it was indeed case 1 above.

Then let's consider what it means: it would mean (let's called it statement 1) "it is a fact that there is no audible difference between 16/44 and 24/96"

Correct? Excellent. I hope you agree.

Ok, given statement 1 is true, let's consider statement 2: "it is a fact that there is no audible difference between 16/44 and 24/768"

Based on my simple logical reasoning, statement 1 does not imply statement 2. Is my understanding correct?

There will be no human audible difference between 16/44 and 24/anything greater than 44.

The only possible audible difference would be if a better master / mix was used to produce the "HiRes" versions.

The CD quality version would sound identical if the same master was used for both.

Or are you saying your amplifier and speakers reproduce those crazy high frequencies and you are a superhuman who can hear frequencies above 20 khz?

 

[melomane13]

Thanks for your personal opinion (please correct me if I am wrong with the objective proof / evidence as I may miss them).

Let's assume two cases in your proof:

1. Yes, you do have the objective proof / evidence. That's good, then your claim is not an opinion as you said, it is a fact then.
2. No, you don't have any proof / evidence. Then, your claim would be classified as personal opinion.

Agree?

Let's assume it was indeed case 1 above.

Then let's consider what it means: it would mean (let's called it statement 1) "it is a fact that there is no audible difference between 16/44 and 24/96"

Correct? Excellent. I hope you agree.

Ok, given statement 1 is true, let's consider statement 2: "it is a fact that there is no audible difference between 16/44 and 24/768"

Based on my simple logical reasoning, statement 1 does not imply statement 2. Is my understanding correct?

you are wrong.

1) of course, if I wrote a "fact" it is because I have the possibility of providing scientific experiments, for example from AES, which show that in a controlled test, it is not possible to hear differences between a 24/96 signal and the same with an ADC/DAC 44.1/16 in the middle.

Moreover, I have personally experienced it.
I invite you to do the same

2) since it is not possible to hear differences between 16/44 and 24/96 it is obvious that you can even have an infinite bandwidth without this adding anything to the sound felt.

I hope you won't debate this point with sophisms

 

[Tell]

To say it like this. Bit depth is all about noise. 16bit gives you a signal to noise ratio of 96dB, or with shaped dither (moving that dither noise to parts of the frequency response where we are less sensitive) it can have a SNR up to 120dB, and if you are going to play that back at 120dB the noise floor of your music will still be around 20-30dB below the noise floor in a typical quite room. Hence you will have to play _extremely_ loud to even hear that noise floor and damaging your ears while doing so. So upping that to 24bit is quite useless since you'll be deaf before hearing that difference in noise.
And 44.1khz gives you a bandwidth of 0-20khz which covers what's consider that human hearing range. Apparently some children can hear up to 22khz, but there really ain't that much musical information up there anyways, but one could still argue that CD should have gone with 48khz to really cover all the human needs. But in reality I doubt anyone will ever really hear that difference in normal music, especially since adults that actually care about that barely even hears over 16khz or even that. So going up to 768khz samplerate which could capture audio up to maybe 350khz won't give anything audible for us humans whatsoever.

Tldr; unless you're a bat that wants to destroy your ears there really is no use for anything higher than 16bit 44.1khz

 

[melomane13]

Please have a look of my post above. I hope it will make sense to you now. Cheers.

you are wrong again since you don't respond to what I wrote:
either
you know you're wrong but you don't want to admit it, because you just want to be right instead of looking for the truth together.
Or
you sincerely believe in what you say, but you do not have the desire to seriously confront the arguments with a true scientific approach.

whatever it is I'm tired of this useless discussion.

 

[danadam]

Your interpretation is inconsistent with the evidence. See section 4.1 of "Designing and Evaluating a Delta-Sigma DAC for Hi-Fi Audio" (https://liu.diva-portal.org/smash/get/diva2:1745070/FULLTEXT01.pdf). Specifically, that section describes noise shaping in the DAC, and Figure 4.1 shows the resulting reduction in quantization noise in the audio spectrum (fs/2).

That's about the noise introduced by the quantizer in the DAC's modulator, the "Q" block in figure 1.2. It's not about the noise that's already present in the input file.

danadam did not post which DAC and/or oversampling algorithm was used for the test.

The signals generation and upsampling was done with SoX.

Based on the results, though, it appears that whatever algorithm was used for the test probably did not implement noise shaping as discussed in various references that were cited.

That was the illustration that when you do the quantization (to 8-bit in that example) and you have more bandwidth, then the noise in the 0-22k band is indeed lower (file 01 in 8/44k vs file 03 in 8/352k). It was also the illustration that once you did the quantization having less bandwidth, you cannot "undo" the noise in the 0-22k band by upsampling (files 01 and 02 in 8/44k and 8/352k respectively vs file 03 in 8/352k).

Introducing noise-shaping will not change that conclusion. The only difference will be that the blue (above 20k) and yellow graphs won't be flat but... er... shaped. SoX can do noise-shaping only in 44k and 48k, so the only way I have to illustrate that, is by starting with lower sampling rate:

• File 01 is -20 dBFS 1 kHz tone generated in 8/11k.
• File 02 is the file 01 upsampled to 8/44k with noise shaping.
• File 03 is -20 dBFS 1 kHz tone generated in 8/44k with noise shaping