Upsampling 16/44.1 collection a good idea?

[antcollinet]

Just want make sure that I fully understand what you mean.

Did you mean I cannot hear anything with frequencey higher than 22kHz so extra samplerate over 44khz won't help? or
Did you mean I cannot hear the difference in the audible range between the audio signal re-constructed using 44kHz and 768kHz?

There is no difference in the audible range from the higher sample rate. The only thing that the higher sample rates give you is the ability to capture frequencies above 20khz. Which are inaudible.

Could they co-exist?

no they couldn't. Everything audible is measurable.

 

[antcollinet]

oversampling can give us more sampled points in a second. It helps the DAC to reconstruct the final analog signal with less artifacts for the audible range. So, it helps in the audible range.

Again nope.

You don’t need more points to perfectly reconstruct a band limited signal than 2x the maximum frequency in the band. More points are not needed for perfect reconstruction.

Look at the Monty video again. Look at the part where he shows a 20kHz signal. Perfectly reconstructed sine wave, even though there are only just over 2 samples per cycle.

 

[MaxwellsEq]

No it's not poor analogies, because they are both digital systems complying under the same Nyquist-Shannon Theorem. As I explained earlier audio is a lot simpler with only time and amplitude, while images have two spatial dimensions, at least three channels in amplitude and when it comes to video a time dimension as well. But they both is bound by the same rules and in the end it's our senses that are the limiting factor. And fyi even the very old analog film of the moving train that the Lumiere brothers showed in 1986 have aliasing. It's of course experienced differently, but it's the exact same principle as in digital audio.

1. Yes audio have come way further than image technologywise, but that's why I do compare it to 8K and 16K since that's where I think our eyes no longer can resolve any differences even though the technology itself haven't come that far yet.
2. The exact same as in audio, except that, again, audio have come further technologywise.

Static images are a poor analogy. The microphone and its preamplifier have a frequency range that comfortably extends beyond Red Book fs/2 as can the playback amplifier and speakers. So, in theory Red Book is not as good as the analogue chain it sits in. This is not true of lenses which are rolling off at or near the half sample frequency.

In audio the analogue bits can easily have a wider frequency range than Red Book. In photography they can't. I wish Red Book had been 48kHz (like professional gear at the time). But you can argue 88 or 96 make sense because they are probably comfortably wider than the analogue chain. With photography the lens is the limiting factor.

 

[KSTR]

As explained by others, offline oversampling makes sense if you need, for whatever reasons, a better (or a different) reconstruction filter than the ones the DAC chips typically offer. It has pitfalls like intersample overs and in general needs attention to processing details. That's basically all that has to be said, IMHO.

 

[Tell]

Static images are a poor analogy. The microphone and its preamplifier have a frequency range that comfortably extends beyond Red Book fs/2 as can the playback amplifier and speakers. So, in theory Red Book is not as good as the analogue chain it sits in. This is not true of lenses which are rolling off at or near the half sample frequency.

In audio the analogue bits can easily have a wider frequency range than Red Book. In photography they can't. I wish Red Book had been 48kHz (like professional gear at the time). But you can argue 88 or 96 make sense because they are probably comfortably wider than the analogue chain. With photography the lens is the limiting factor.

I guess you never used a particularly good lens then..
But yeah we're not there technologywise that we can capture and display images in all dimensions that exceeds our eyes capabilities, so I am talking about theory mostly, and also me working in 3D I see it from that "perfect" perspective as well. But the digital sampling theorem still very much applies just the same and it's all constrained by our human senses, which is what I have been trying to explain and compare.
Anyways, I don't really have neither the time or every to keep on with this conversation, I have explained my point of view that I still fully stand by and the person I was mostly talking to seem to have understood it so

To my understanding,

1. oversampling is not to make us hear anything higher than 22kHz. It is a big mis-understanding in the general public (probably from the HiRes marketing). I got tricked before too. Yes, I agree with you it won't help as this is not what we want.

2. oversampling can give us more sampled points in a second. It helps the DAC to reconstruct the final analog signal with less artifacts for the audible range. So, it helps in the audible range.

BTW, with the final reconstructed analog audio signal, I think there is no concept of "resolution' (please correct me if I am wrong). We are indeed hearing a combination (aka superimposed) of different analog sine wave (at different frequencies and amplitudes). I think we cannot use the term resolution like what we use in the digital domain.

1. Yup

2. Well it might help in theory but in practice I don't think it does.

Guess it depends on how you define "resolution". I'd say that a 20khz analog sinewave have higher resolution than a 5khz one, just like we measure camera lenses with how many lines in the real world they can resolve.

 

[melomane13]

10k sine wave without oversampling:

View attachment 364862

i think you have wrong copied the significations of images.
the first is
"output from the Topping E30 with the F1 filter – Super-slow Roll-off” – engaged"
the second is:
"filter the output, things become “perfect” again. Here’s 10kHz from Topping E30 with F1, Linear Phase Fast Roll-off"

so isn't oversampling but filtering

 

[KSTR]

the first is
"output from the Topping E30 with the F1 filter – Super-slow Roll-off” – engaged"

AKM's "super-slow roll-off" is an emulation of "filterless NOS".

 

[antcollinet]

My understanding from @pkane that there is difference but not a lot. Please correct me if I am wrong as I don't want to mislead people here.

The difference @pkane is talking about comes from the choice of filter - not from the sample rate.

If you choose the ideal sharp filter in both cases, then there will be no audible change. Just as you saw yesterday, you can create differences by an innapropriate choice of filter. You might prefer the effects of those filters - but the result on the audio is a less accurate reconstruction, compared to the encoded waveform.

Note - resampling might create a tiny measurable difference. But if the same (or even just similar) filter characteristic is used that difference is not going to be audible.

Now I’m going to suggest you stop wasting everyones time JAQ ing (Just askign questions) and do a bit more reading and learning. It is all here.

 

[antcollinet]

Oversampling is not something just working in theoretical space. According to wikipedia, there are three main reasons for performing oversampling: to improve anti-aliasing performance, to increase resolution and to reduce noise.

Here you can see how it works in reality here:

10k sine wave without oversampling:

View attachment 364862

10k sine wave with oversampling:

View attachment 364863

Source: How to pick the best filter setting for your DAC – Addicted To Audio

AGAIN NO.

The difference there doesn’t come from oversampling.

It comes from applying a filter vs not applying a filter (slow rolloff)

I explained it to you yesterday - it appears you are not listening - in-fact now I think you are sealioning (look it up - just another troll). I’m going to suggest you stop wasting everyones time and do some quiet learning. Read around the forums and actually try to take some of it on board.

I’m out.

 

[pkane]

On his page, he stated "the F5 filter is described by Topping as “Super-slow Roll-off”. But it’s known also as “Optimised Transient” (in a Pro-Ject Audio DAC) and zero oversampling, or NOS, for non-oversampling."

@antcollinet is correct. A NOS filter doesn't mean non-oversampling -- that's just the marketing term used to get audiophiles to pay more for an unnecessary feature. That filter is just a very poor reconstruction filter when used without oversampling. The example you posted shows that it doesn't work well at all when applied without oversampling. So, a filter that's designed not to work without over sampling requires some other device (like software or hardware oversampler) to do the work that's normally done by the DAC. What a surprise!?

As I said before, use of proper reconstruction filter makes all this off-line or real-time software oversampling unnecessary. And Topping includes proper reconstruction filters.

 

[pkane]

From you understanding, if we feed a 44.1/16bit PCM source for a pure 10k sine wave to a real NOS DAC (i.e. not using any digital filtering at all) and perform real measurement with a oscilloscope, what would we get?

1. a similar stair-step output as I show above
2. or a perfect smooth sine wave?

#1. No filtering is an improper way to reconstruct PCM signal. This was covered nearly 100 years ago by Shannon, Nyquist, and others.

 

[Tell]

Some measurements and subjective test from Archimago: https://archimago.blogspot.com/2018/11/nos-vs-digital-filtering-dacs-exploring.html

 

[KSTR]

A NOS filter doesn't mean non-oversampling

Of course it does. When viewed as a black box, one cannot tell a difference between a NOS emulation DAC and a true NOS DAC, they both show the same stair-steps and the same frequency response from the ZOH (until some top-end filtering becomes effective, rounding off the sharp edges of the stair steps). The actual main reconstruction filter function is "repeat samples N times", like 16 times for an AK4490/93 running at 44.1 or 48kHz.

 

[KSTR]

In order to regenerate a smooth sine wave output, oversampling is necessary (please corret me if I am wrong)

No. Filtering is necessary. It could be done in analog, following the true NOS DAC output but that impractical. That's why oversampling was invented, to reduce the burden on the analog filter and move most of the reconstruction into a digital filter which is much simpler to realize.

 

[pkane]

Of course it does. When viewed as a black box, one cannot tell a difference between a NOS emulation DAC and a true NOS DAC, they both show the same stair-steps and the same frequency response from the ZOH (until some top-end filtering becomes effective, rounding off the sharp edges of the stair steps). The actual main reconstruction filter function is "repeat samples N times", like 16 times for an AK4490/93 running at 44.1 or 48kHz.

What? Oversampling is a practice of increasing sampling rate 16x or whatever. A filter, NOS or otherwise, does not by itself increase sampling rate, so it’s a misnomer to call a filter NOS. A DAC can be NOS, but not a filter.

 

[KSTR]

As I need a NOS DAC for my unsampled music, do you think these modern DAC with NOS filters would be good enough? Or a real NOS would be better?

Technically, there is no difference, therefore there is no audible difference as well.

 

[KSTR]

What? Oversampling is a practice of increasing sampling rate, usually by ZOH. A filter, NOS or otherwise, does not by itself increase sampling rate, so it’s a misnomer to call a filter NOS. A DAC can be NOS, but not a filter

I think we know and mean the same thing, Paul.
The digital part of a reconstruction filter in a DAC device can either be a true/proper filter of some sort or the mentioned "sample repeater" so that the digital filter is actual equivalent to NOS behavior. On top of that, in any modern DAC and many NOS DAC we have some sort of analog filter to keep the dirty RF out, so to say.
EDIT: FWIW, the AKMs have a 2-tap FIR averager in the final switched-capacitor output stage, and likely also some IIR filter after the main filter after the upsampler. That main filter always reverts to the sample repeater at 8x and 16x rates, that is the proper filters are only effective at <352kHz sample rates.

 

[terryforsythe]

I've read them, thank you. You are missing the point. 1) there's a reference; 2 there's a preference. They are not the same things.

In audio, the reference usually comes in the form of frequency sweeps, pink noise, impulse signals, etc. These things are used almost universally by component engineers. There is not a golden music recording that is used universally as a reference.

 

[terryforsythe]

Note - resampling might create a tiny measurable difference. But if the same (or even just similar) filter characteristic is used that difference is not going to be audible.

Is my understanding correct?

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/

 

[terryforsythe]

Getting back to the original question asked by the OP, the answer depends on whether the algorithms the OP will use to upsample the recordings will do any better of a job at improving the effective SNR than the oversampling implemented by the OP's DAC, assuming the DAC is performing oversampling. Also, it depends on whether upsampling the recordings, with the DAC again performing oversampling, will create any other problems or serve to audibly improve the effective SNR even further. I don't know the answers to those questions. What I can say is that my DACs seem to be doing a good job - I never have noticed quantization noise when playing 44.1/16 recordings.