Can you hear upscaling?

[eq1849]

OK, I think you are right, it is valid to test the M-Scaler with any DAC you would want. But from the GoldenSound review, it may not be reasonable to expect peak performance by combining it with another company’s DAC.

“If the MScaler had a USB output option it may have even been worth routing the output of the MScaler through a high quality DDC.
Unfortunately, given the use of Chord’s proprietary dual-BNC output, this is not possible unless you’re happy to stick to 192khz upsampling.
Luckily, this does not really affect Chord DACs specifically as their design makes them highly resistant to jitter.

Chord DAVE jitter fed by MScaler.
Practically perfect”.

Furthermore, the GoldenSound review says earlier on:

“The WTA is a convolution of a rectangular window with cosine tapers. Giving excellent time domain accuracy, though at the expense of increased spectral (frequency domain) leakage.
Filter windows are always a trade off and there is no perfect design. The WTA filter just is heavily geared toward time domain accuracy.
Though with such a high tap count the window itself is of less importance anyway and you will get better time AND frequency domain accuracy than a lower tap count filter built into a DAC.

Additionally we can see from the flat noise floor that the M-Scaler seems to be using standard TDPF dithering. This is a bit surprising as I’d expected it would be using a more advanced noise shaping option given the compute power in this device. It could simply be that Chord feels their DAC’s internal noise shapers for the PWM stage make this irrelevant, but for use with other DACs it would be nice if there was a higher quality dithering/noise shaping method built in.”

This is interesting because in the original interview posted in this thread Rob Watts discusses how in his opinion small timing issues can degrade how listeners experience music reproduction. I do not have enough knowledge or context to understand if the quoted finding in the GoldenSound review is what Watts is talking about when he mentions timing, but according to the GoldenSound review, the M-Scaler filter has “excellent time domain accuracy”.

kn

Perhaps they are referring to something along the lines of how fast the sampling interval is?

44.1khz - 22.7uS(micro seconds).

192khz - 5.4 uS.

384khz - 2.7 uS.

768khz - 1.35 uS.

and so on...

 

[bigshot]

I think it should be self evident that upscaling doesn’t add any more information. Take a 64k mp3 with artifacting and upscale it to 24/96… it still sounds like a 64k mp3.

 

[gregorio]

This is interesting because in the original interview posted in this thread Rob Watts discusses how in his opinion small timing issues can degrade how listeners experience music reproduction. I do not have enough knowledge or context to understand if the quoted finding in the GoldenSound review is what Watts is talking about when he mentions timing, but according to the GoldenSound review, the M-Scaler filter has “excellent time domain accuracy”.

While I’ve no reason to doubt GoldenSound’s published measurements, some of his conclusions/findings are straight out of the audiophile BS manual. I personally find this particularly egregious, he obviously has some serious knowledge in some areas but instead of using it to help audiophiles reject some of the marketing BS, he often does the opposite and reinforces it. I expect that from amateur or ignorant audiophile reviewers but it’s not good coming from someone who isn’t ignorant! A classic example from your quote: “If the MScaler had a USB output option it may have even been worth routing the output of the MScaler through a high quality DDC.” - What?! So, you pay thousands for a unit with horrific jitter and then try to fix it by spending another bunch of money on a DDC, that also has pretty poor jitter but nowhere near as bad as the MScaler. What an engineer or someone who hasn’t drunk the audiophile koolaid would do is obvious, simply buy a modestly priced DAC to start with and plug your USB straight into it, missing out both the snake oil MScaler and the nonsense DDC. The modestly priced DAC will re-clock the USB input and reduce the jitter to way lower levels than the MScaler and even significantly lower than a DDC!

To answer your question though, it’s really not clear what they’re going on about, the “timing issues” appear to refer to the supposed pre/post filter ringing but Watts’ filter seems to have the same ringing as any DAC with a linear phase anti-image filter. Maybe he means “timing” in the sense of where the transition band starts and how long it lasts (it’s transition band width), which is inaudible or maybe he’s referring to the timing resolution of 16/44.1 but there’s no issue with the timing resolution of 16/44.1 and I don’t see how his filter design would affect that anyway. The only timing issue I can see is the latency of his MScaler!

In your opinion what is the difference between Upsampling and Upscaling?
Both seem, IMHO, to be inserting INTERPOLATED samples between the ACTUAL samples to increase the sampling rate on playback.

I’m not qualified to answer that question, as I don’t know very much about image processing. From my very limited understanding, upscaling is the process of taking an image of a certain original size and blowing it up and/or increasing the number of pixels and that appears to involve some quite complex resampling and processing, commonly achieved with AI (machine learning) technology these days. There’s no equivalent in audio, there are no pixels in audio and no equivalent to blowing up the image size. Oversampling in audio is typically achieved by simply adding a bunch of zero value samples between the existing samples, which does not affect the mathematical outcome but raises the Nyquist point, allowing for a relaxed, simple/cheap, analogue reconstruction filter. A higher sampling rate during the DAC process does not introduce information that is not in the original 16/44.1 data.

G

 

[gregorio]

Perhaps they are referring to something along the lines of how fast the sampling interval is?
44.1khz - 22.7uS(micro seconds).
192khz - 5.4 uS. ….

I certainly hope not, as that would indicate they don’t have much understanding of digital audio or do have a good understanding but are lying! The sampling interval just defines the amount of time between each sample, it does not define the timing resolution, which in the case of 16/44.1 is several orders of magnitude smaller!

G

 

[Davesrose]

I’m not qualified to answer that question, as I don’t know very much about image processing. From my very limited understanding, upscaling is the process of taking an image of a certain original size and blowing it up and/or increasing the number of pixels and that appears to involve some quite complex resampling and processing, commonly achieved with AI (machine learning) technology these days. There’s no equivalent in audio, there are no pixels in audio and no equivalent to blowing up the image size. Oversampling in audio is typically achieved by simply adding a bunch of zero value samples between the existing samples, which does not affect the mathematical outcome but raises the Nyquist point, allowing for a relaxed, simple/cheap, analogue reconstruction filter. A higher sampling rate during the DAC process does not introduce information that is not in the original 16/44.1 data.

G

The most basic way of resizing an image would be multiplying pixels with new pixels of the same values (so similar with audio as far as no difference compared to original). However, there's been quite a few algorithms used throughout the years-well before current AI (resampling an image while resizing it). One traditional method has been to sample the pixel and neighboring ones. The new pixels generated around it being the average of the original pixels. One potential issue is the image may look diffused with borders. A common way of sharpening an image is use an "unsharp mask": in which the image "mask" is blurred and combined with the original image (to make borders look sharper). When it comes to image processing and AI, I've found the most impressive features are things like automatically masking a background and replacing it with a new one. Such as taking a photo that has a cloudy sky, and insert a blue sky with no manual adjustments used. I've seen some people try to upsample movies to 4K 60fps, and I've found they look awful. Normally the detail is lost and nothing like native 4K (looks like bad "digital noise reduction" and the interpolated 24 to 60fps looks unnatural).

 

[71 dB]

In your opinion what is the difference between Upsampling and Upscaling?

Upsampling: Resampling a signal upwards, say from 44.1 kHz sampling rate to 58.8 kHz.
Upscaling: Resizing a digital image upwards, say from 600x400 pixels to 750x500 pixels.

Both seem, IMHO, to be inserting INTERPOLATED samples between the ACTUAL samples to increase the sampling rate on playback.

Upsampling can be done by inserting interpolated* samples between the actual samples if the upsampling coefficient is an integer, say 3. If it is a simple ratio such as 4/3, we can do upsampling by the nominator (here 4) first and then downsampling by the denominator (here 3). If the upsampling coefficient is not simple, say 3989/2999 (e.g. from 44.1 kHz to 58.65785262 kHz), this method becomes computationally VERY taxing and impractical. Instead the upsampled signal is constructed by summing windowed sinc-functions (centered at each old sample point and multiplied by the sample value) at the new sample points, most of which are going to hit somewhere between old sample points.

* The method of interpolation dictates the accuracy of upsampling. Linear interpolation is fast, but the result is low quality. This might be adequate in certain applications such as children's toys. Using higher order polynomials for interpolation increases accuracy and gives mid quality results, but for high fidelity applications zero padding is a common method of upsampling, because it doesn't distort the old bandwidth at all. Instead it generates a lot of noise above the old Nyquist which must be filtered out.

 

[XTF1]

Upsampling: Resampling a signal upwards, say from 44.1 kHz sampling rate to 58.8 kHz.
Upscaling: Resizing a digital image upwards, say from 600x400 pixels to 750x500 pixels.

I believe it is trivial to re-encode a 16-bit audio file into 24-bit by, for example, padding zeros—no gain in quality, no way to “create” something that was not here to start with.

Is there a name for this process, or family of processes?

 

[Joe Bloggs]

I believe it is trivial to re-encode a 16-bit audio file into 24-bit by, for example, padding zeros—no gain in quality, no way to “create” something that was not here to start with.

Is there a name for this process, or family of processes?

Zero padding?

 

[XTF1]

Zero padding?

Yes…!!!
Thanks for correcting me, Bloggs Joe !!!

 

[Joe Bloggs]

Zero padding would be the noun and padding zeros would be the verb phrase. Looks like you already had everything correct under your nose

 

[Vamp898]

I just tested it. I upscaled an 48kHz file to 384kHz and let DeltaWave analyze the difference.

Zero, nothing.

So no, upscaling changes nothings.

That doesn't mean this upscaler doesn't change the sound. Maybe its not transparent and changes the frequency response or whatever. But if, than that would be unrelated to the upscaling

 

[VNandor]

I just tested it. I upscaled an 48kHz file to 384kHz and let DeltaWave analyze the difference.

Zero, nothing.

So no, upscaling changes nothings.

Is there really nothing left in your null file or just a tiny amount that doesn't matter?

 

[Vamp898]

Is there really nothing left in your null file or just a tiny amount that doesn't matter?

There is noise in the >=20kHz region that was created by the upscaling process. It was minus some houndret db, so noise floor. The original recording has zero information there (because it stopped at around 22kHz), and the upscaled version has niose floor there.

But everything below ~22kHz is 99.99999% identical. Not an tiny amount that doesn't matter, no difference.