Up Sampling

[Ron Texas]

Over at CA I see a lot of discussion about up sampling either to the highest PCM rate supported by the DAC or the highest DSD rate. The argument is a computer with it's resources can do a better job than the hardware inside the DAC chip. Furthermore, if fed a DSD stream, the DAC will do a better job of decoding it than a PCM stream.

Resampling to high rates of PCM with SOX is trivial from the standpoint of computer resources. Going to high rates of DSD requires at least a moderately powerful computer which is also completely quiet. I don't have one. Someone at Roon said most of the benefit is in going to high rates of PCM. I have seen statements that DSD 128 is better than DSD 64 because it gets the noise further from the audio band. Why DSD 256 or DSD 512 would be better is a mystery to me.

My personal experiments with high rates of PCM are inconclusive. Is there a real reason for any of this, or is it nonsense? If it's nonsense, then why?

 

[andreasmaaan]

EDIT: I've left this post here so the thread still makes sense, but I now understand that what I said here is essentially incorrect

I'm definitely not an expert on this but have just been asking similar questions in another thread and doing some reading and getting some answers from more knowledgeable forum members that may help with your question too.

With PCM digital, there is a low-pass filter placed just before the nyquist frequency (e.g. 22,050Hz for 44.1Khz PCM). This filter introduces ringing (pre- and/ or post- echo, depending on the filter). At 44.1Khz PCM, this ringing may be in the audible band (beginning around 20Khz or lower). By using a higher sample rate, the nyquist frequency, and therefore the transition range, and therefore the ringing, is pushed up to near the new (higher) nyquist frequency. For e.g. a 96Khz sample rate, this pushes the ringing up to the 40Khz range (exactly where depends on the filter type used). Get the sample rate high enough, and use an appropriate filter, and the ringing stays completely out of the audible spectrum.

Others will definitely be able to explain this in more depth, but that is what I understand to be the gist of the theoretical argument in favour of higher sample rate PCM.

I'm not sure if/why a computer would do it better than a dedicated DAC.

And I'm not sure about your questions regarding DSD, except to hazard that perhaps the higher rates push audible noise out of / higher above the audible frequency range.

 

[DonH56]

Running the DAC faster may or may not help in the end. You must somehow create new samples between the existing samples so, while the algorithms can be pretty sophisticated, it is still a guess. Greater sampling rate means greater noise bandwidth as well, more stringent requirements on DAC and support circuitry settling and stability, etc. When discussing bandwidth and slew/settling in most designs the engineering principle is to use just enough bandwidth, not too much, and not too little, to preserve signal fidelity.

Delta-sigma designs use noise shaping to push the noise above the audio (audible band). THey are sort of a special case in that they depend upon oversampling to provide the bandwidth needed to implement noise shaping. However, they use lower-resolution loops with oversampling and filtering to create a lower-speed, higher-resolution result. There's an article about them somewhere here that I put together some time ago. Since you are "boosting" HF noise, then increasing the oversampling can improve in-band SNR, and make the filter easier to implement. Some of that benefit is lost if you extend the bandwidth.

Put another way, originally the reason to move from say 64x to 128x oversampling was to improve the SNR in the desired signal band. If you double the signal bandwidth when you double the sampling rate, you haven't necessarily improved the net SNR, and in fact because there are noise sources besides quantization (conversion) noise, you may actually have lower net SNR.

It's complicated...

 

[andreasmaaan]

Do you happen to have a link to your article @DonH56 ?

 

[Ron Texas]

Here is a CA thread on the topic, FWIW:

https://www.computeraudiophile.com/forums/topic/44706-consensus-about-upsampling-to-512-dsd/

It's going beyond DSD 128 which mystifies me the most.

 

[DonH56]

Do you happen to have a link to your article @DonH56 ?

After a search here are a couple of introductory articles:
https://audiosciencereview.com/forum/index.php?threads/digital-audio-sampling-101.1919/#post-50385
https://audiosciencereview.com/foru...igma-delta-digital-audio-converters-dac.1928/

HTH - Don

 

[Ron Texas]

After a search here are a couple of introductory articles:
https://audiosciencereview.com/forum/index.php?threads/digital-audio-sampling-101.1919/#post-50385
https://audiosciencereview.com/foru...igma-delta-digital-audio-converters-dac.1928/

HTH - Don

I just took a look. Not easy reading.

 

[andreasmaaan]

Thanks @DonH56

 

[Fitzcaraldo215]

Here is a CA thread on the topic, FWIW:

https://www.computeraudiophile.com/forums/topic/44706-consensus-about-upsampling-to-512-dsd/

It's going beyond DSD 128 which mystifies me the most.

I have not tried much in the way of software upsampling of PCM or DSD. I don't see real evidence of it adding much, if anything. Yes, I am mystified by DSD128, 256 or 512, too. I prefer to stick with the sampling rate I have got in the file as distributed. Often, that is a downsampling of the higher Rez, as recorded sampling rate and/or a DSD/PCM conversion, but you cannot entirely accurately put back what you haven't got, which is @DonH56's point. And, the storage overhead mounts tremendously. The ideal might be to use the native, as recorded resolution and format, but that is not always available and much of that is really just for "headroom" in facilitating recording production. Some engineers agree with that, even my friend who provides me with DSD256 files.

My music playback scenario now is primarily from SACD as DSD64 rips in DSF files. I use JRiver to convert that to PCM352k (the default) and to SoX downsample it to 176k for compatibility to with Dirac's 192k upper limit. It's all done on the fly, so no added storage overhead, and no problems at all. Likely, I will go to PCM352k playback when Dirac 2.0 with its 384k upper limit is out, eliminating the downrezzing. I doubt it will make any noticeable difference, though. It's just mainly an easy hygiene thing to keep the signal path as simple as possible.

But, it's like the multiplying megapixels in cameras - not much use to the average guy. Except, more is always better, right? Well, after a point reached long ago, no, except maybe for marketing, but not much in performance in typical use. OTOH, I do find hirez audio to be of some audible advantage over RBCD resolution. I have got it already in my SACD/BD library files as distributed, so I am happy to use it.

 

[Ron Texas]

I have not tried much in the way of software ... And, the storage overhead mounts tremendously...



My music playback scenario now is primarily from SACD as DSD64 rips in DSF files. I use JRiver to convert that to PCM352k (the default) and to SoX downsample it to 176k for compatibility to with Dirac's 192k upper limit. It's all done on the fly, so no added storage overhead...

But, it's like the multiplying megapixels in cameras...

Up sampling on the fly eliminates the storage problem, but needs a fast and silent computer.

The obvious truth is there is only so much information in a PCM file, so changing it into something else adds no information. It probably changes how the DAC works with it.

I playback dsf by converting to PCM 352k and down sample if needed. I could feed 352k to my Grace M9xx. That DAC is DSD capable up to 256, but I don't like getting up to fool with the volume control. JRiver can control the volume of some DAC's, but not mine.

I shoot a 44mp resolution DSLR. The results are stunning, and easily better than the 12mp I used 5 years ago. High resolution files allow for more cropping. It's better than the 36mp camera I was using, but probably due improvements in the camera other than resolution.

 

[Thomas savage]

After a search here are a couple of introductory articles:
https://audiosciencereview.com/forum/index.php?threads/digital-audio-sampling-101.1919/#post-50385
https://audiosciencereview.com/foru...igma-delta-digital-audio-converters-dac.1928/

HTH - Don

Thank you mate

 

[Fitzcaraldo215]

Up sampling on the fly eliminates the storage problem, but needs a fast and silent computer.

The obvious truth is there is only so much information in a PCM file, so changing it into something else adds no information. It probably changes how the DAC works with it.

I playback dsf by converting to PCM 352k and down sample if needed. I could feed 352k to my Grace M9xx. That DAC is DSD capable up to 256, but I don't like getting up to fool with the volume control. JRiver can control the volume of some DAC's, but not mine.

I shoot a 44mp resolution DSLR. The results are stunning, and easily better than the 12mp I used 5 years ago. High resolution files allow for more cropping. It's better than the 36mp camera I was using, but probably due improvements in the camera other than resolution.

Yup, agreed, Ron.

On cameras, I was thinking of my wife and her friends, typical users. No argument that your use of high megapixels is appropriate for your heavier duty, more exacting needs, especially enlargements,cropping, etc. It's as I said about ultra hirez. Engineers find high sampling rates useful in the many steps of recording production, for example, keeping edits, splices and such a bit cleaner, more precise and less potentially intrusive or keeping a-d filter artifacts further from the audible band. Similarly, I think a 24-bit recording chain's added headroom at the low level end can benefit releases on 16-bit CD. But, not very much useful, audible information, if any, comes through to the listener directly on playback from the ultra sampling rates.

 

[svart-hvitt]

Upsampling doesn’t give you more information; so it’s not hifi-er.

But I am open for perceptual aspects that may be pleasing.

I tried HQ Player to upsample Tidal 1644 to 2496. My speakers, Genelec 8351, upsample to 2448 and 2496 anyway, so doing this in software made some sense, I thought.

Did I perceive a gain in pleasure? No. I don’t know if I really preferred the original. All «tests» were sighted.

Please note that my speakers my complicate the overall chain a bit due to their internal DSP.

Today, I don’t use upsampling. I prefer the original, which may be 1644 and 24xxz in Qobuz.

Just my anecdotal 2 cents.

 

[Pio2001]

With PCM digital, there is a low-pass filter placed just before the nyquist frequency (e.g. 22,050Hz for 44.1Khz PCM). This filter introduces ringing (pre- and/ or post- echo, depending on the filter). At 44.1Khz PCM, this ringing may be in the audible band (beginning around 20Khz or lower).

Hi,
This ringing is completely inaudible. You can try the following : lowpass a musical sample at 10 kHz instead of 22050 Hz.
With a brickwall filter, the ringing is obvious. It is a 10 kHz sine playing over the music.
But if I push the filter frequency above my hearing threshold (13500 Hz), then I can't hear it anymore.

But in a DAC, the filter is not a brickwall. It is a progressive filter that starts to attenuate near 20 kHz, until 22 kHz. I've tried this kind of filter at 10 kHz. And there is no ringing at all !

Since there is already nothing audible for me at 10 kHz, I can't imagine how the same thing could be audible at 22 kHz !

By using a higher sample rate, the nyquist frequency, and therefore the transition range, and therefore the ringing, is pushed up to near the new (higher) nyquist frequency. For e.g. a 96Khz sample rate, this pushes the ringing up to the 40Khz range (exactly where depends on the filter type used).

This is wrong for two reasons.
First, in order to upsample from 44100 Hz to a higher sample rate, you need to apply exactly the same anti-alias filter as your DAC's at 22050 Hz in your resampler. So your "problems" will occur at that frequency anyway, not higher.
And second, since you have no information above 22050 Hz, you will have no ringing above that frequency. For ringing to occur, there must be something that rings.

By the way, this is valid at 22050 Hz too. If your musical content is filtered below (for example at 20 kHz), then you can filter with anything you want at 22 kHz, even a Sinc brickwall, nothing will happen. No pre-echo, no post-echo, as long as there is no energy in the transition band of your filter.

All this is valid for standard DACs. In the audiophile world, there are some incredibly flawed designs, such as NOS DACs (DACs that don't oversample at all), or DACs with a set of filters preserving "transient response" (which was never altered to begin with). These ones have audible problems well below 20 kHz, and the above is not true for them.
Upsampling is very good with these DACs, because it rejects their flaws above the audible range. In fact, with them, upsampling is nothing else than reverting to a standard oversampling with a standard filter, performed in software, rejecting their horrible filters above the audible range (in the case of a pure NOS DAC, what acts as a "filter" will be the bare zero-order-hold circuit that generates the square steps in the analog domain).

Delta-sigma designs use noise shaping to push the noise above the audio (audible band). THey are sort of a special case in that they depend upon oversampling to provide the bandwidth needed to implement noise shaping.

You mean for DSD or PCM ?

 

[andreasmaaan]

This ringing is completely inaudible.

That might be true. I didn't claim the ringing was audible, but rather that it might fall within the audible range.

First, in order to upsample from 44100 Hz to a higher sample rate, you need to apply exactly the same anti-alias filter as your DAC's at 22050 Hz in your resampler. So your "problems" will occur at that frequency anyway, not higher.

Thanks, it seems I'd misunderstood this

 

[DonH56]

You mean for DSD or PCM ?

The modulator itself does not really care as far as noise shaping is concerned.

 

[astr0b0y]

My reason (poorly studied, very likely incorrect) for upsampling to DSD is the belief that the digital waveform structure of DSD is more like an analog waveform in that it doesn’t have the stepped appearance of PCM, or has a less stepped appearance. The higher the DSD rate the "smoother" the waveform. Therefore when the DAC is converting to an analog output it will more easily be able to reconstruct the smooth original analog signal.

It feels good to get that off my chest. It occurs to me after writing and rereading it how ludicrous it sounds.
However, many many people are doing this upsampling. Maybe it’s one of the new audiophile snake oils?

 

[n2it]

I have not tried much in the way of software upsampling of PCM or DSD.

My music playback scenario now is primarily from SACD as DSD64 rips in DSF files. I use JRiver to convert that to PCM352k (the default) and to SoX downsample it to 176k for compatibility to with Dirac's 192k upper limit. It's all done on the fly, so no added storage overhead, and no problems at all. Likely, I will go to PCM352k playback when Dirac 2.0 with its 384k upper limit is out, eliminating the downrezzing..

I playback dsf by converting to PCM 352k and down sample if needed. I could feed 352k to my Grace M9xx.

I'd call converting DSD64 to PCM352 a form of upsampling - while there is certainly no perfect DSD to PCM conversion, 96/24 seems to cover the full spectrum of DSD64.

Also, why would you down sample if not needed? This downsampling would inherently lose data. Why not just convert DSD64 directly to 96k (or 176k if you want to be really really conservative) and not have to ever deal with downsampling and save a lot of disk space too?

 

[Blumlein 88]

My reason (poorly studied, very likely incorrect) for upsampling to DSD is the belief that the digital waveform structure of DSD is more like an analog waveform in that it doesn’t have the stepped appearance of PCM, or has a less stepped appearance. The higher the DSD rate the "smoother" the waveform. Therefore when the DAC is converting to an analog output it will more easily be able to reconstruct the smooth original analog signal.

It feels good to get that off my chest. It occurs to me after writing and rereading it how ludicrous it sounds.
However, many many people are doing this upsampling. Maybe it’s one of the new audiophile snake oils?

Very much worth time to watch and ponder this video.

 

[DonH56]

My reason (poorly studied, very likely incorrect) for upsampling to DSD is the belief that the digital waveform structure of DSD is more like an analog waveform in that it doesn’t have the stepped appearance of PCM, or has a less stepped appearance. The higher the DSD rate the "smoother" the waveform. Therefore when the DAC is converting to an analog output it will more easily be able to reconstruct the smooth original analog signal.

Not really... Here is another link that compares the two: https://www.mojo-audio.com/blog/dsd-vs-pcm-myth-vs-truth/

I have not read it thoroughly but it has pictures.