CHORD M-Scaler Review (Upsampler)

[Ken Tajalli]

I see, it's a resampler, not just a bit-depth-increaser.

In real life it's hard to imagine how this could really help fidelity or sound quality.

The only reason to use a sampling rate higher than 44.1khz is to represent frequencies higher than 22,500hz. Even if you could hear those frequencies, you'd have to be adding content (what kind of content, anyway?) that doesn't exist in the original recording, for the ultimate analog output of the resampler to be any different than the original. Why would you want to do that?

Actually there is merit in the idea.
By upsampling, one can implement more precise and sharper filtering. This can have benefits in-band. By shifting the filtering to higher frequencies and sharper slopes, one can have less audio band artefacts.
Think of it like watching a movie, not in standard 24 fps, but in 100 fps.
Not the same thing, but a loose comparison to show that faster moving objects would look smoother, eventhough, the eye can only manage 50fps or so.

 

[Ken Tajalli]

Yes, in a sense the steps are smaller.

It may use a bit more energy because it's more CPU intensive, but realistically nothing in digital audio playback is that difficult for today's CPUs to handle, until you reach 384 Khz / 32-bit with multiple FIR filters, or something.

My Huawei cheap phone can manage 768kHz while implementing DSP, using Neutron media player!
Not difficult at all.
Audio data rate is low, even for a phone CPU. And USB 2 can manage the rate happily.

 

[staticV3]

Less noise, not more precision.

Noise is identical in both graphs as the files are undithered, so only the self-noise of the DAC, output stage, and ADC remain, all of which can be assumed constant between both tests.

The difference visible in the test is entirely due to the higher resolution of 24 bit audio.
I think it's fair to call that more precise.

Edit: I guess you could say that the lower resolution of 16 bit causes higher quantization noise, so less resolution allows for less precise sampling, resulting in quantization noise.

 

[kemmler3D]

Actually there is merit in the idea.
By upsampling, one can implement more precise and sharper filtering. This can have benefits in-band. By shifting the filtering to higher frequencies and sharper slopes, one can have less audio band artefacts.
Think of it like watching a movie, not in standard 24 fps, but in 100 fps.
Not the same thing, but a loose comparison to show that faster moving objects would look smoother, eventhough, the eye can only manage 50fps or so.

That's fair, you can create some breathing room for the filters, although I'm not sure they typically cause audible artifacts either way.

 

[SIY]

Noise is identical in both graphs as the files are undithered,

And you can end the sentence right there. Even if dithering is not deliberately applied, it's there anyway in actual signals because of noise limitations. Of course, Stereophile is unlikely to explain this...

 

[Ken Tajalli]

Noise is identical in both graphs as the files are undithered, so only the self-noise of the DAC, output stage, and ADC remain, all of which can be assumed constant between both tests.

The difference visible in the test is entirely due to the higher resolution of 24 bit audio.
I think it's fair to call that more precise.

Edit: I guess you could say that the lower resolution of 16 bit causes higher quantization noise, so less resolution allows for less precise sampling, resulting in quantization noise.

I think what he means is that, more bits only serve to encode smaller signals, the ones below the ability of 16 bits, and that's where the analogue noise lies. so not much is achieved.
However, for initial recording, 24 bit is better, but not for reproduction.

 

[kemmler3D]

I think what he means is that, more bits only serve to encode smaller signals, the ones below the ability of 16 bits,

Strictly speaking I think you get more accuracy at any level with 24 vs. 16. However, you can only really hear the digital noise relative to extremely low-level signals with 16 bits.

With 24 bits you can't actually hear the digital noise at all, because that's below the physical capability of existing electronic technology to reproduce. I think it may be below the thermal noise at room temperature, or something like that.

 

[staticV3]

Not the same thing, but a loose comparison to show that faster moving objects would look smoother, eventhough, the eye can only manage 50fps or so.

I take it you've never used a high refresh rate monitor before.
60Hz->120Hz is incredibly obvious.
120Hz->240Hz is still pretty noticeable.
240Hz->360Hz gets tricky to discern, but is doable if you're trained and the monitors' pixel response times are fast enough.

 

[Ken Tajalli]

Strictly speaking I think you get more accuracy at any level with 24 vs. 16. However, you can only really hear the digital noise relative to extremely low-level signals with 16 bits.

With 24 bits you can't actually hear the digital noise at all, because that's below the physical capability of existing electronic technology to reproduce. I think it may be below the thermal noise at room temperature, or something like that.

I don't agree.
0 dBfs is same for both scales, all the way down to -96 dBfs both would be identical, at lower than that, 24 bit still has bits to work with, when 16 bits have run out!
But since noise is at around those levels, and the ear can not hear it either, 24 bits potential at reproduction is minimal.

I take it you've never used a high refresh rate monitor before.
60Hz->120Hz is incredibly obvious.
120Hz->240Hz is still pretty noticeable.
240Hz->360Hz gets tricky to discern, but is doable if you're trained.

I have, but a high refresh rate monitor is not what I was talking about. I was talking about high frame rate material.
A US DVD at 60Hz interlaced looks smoother than a PAL DVD at 50Hz, and in modern TV's, such as my Samsung Q95 QN95A, higher settings of "Judder reduction" would make any movie look like a Camcorder footage!
I use it for Football. On movies, I bring it down, to get a little judder, to make it look like film.
Silly, but I prefer it.

 

[kemmler3D]

I believe that upsampling can facilitate other forms of DSP,

Well-implemented DSP will use a higher sampling rate internally if it's necessary, and then downsample for output. That's not to say all DSP is well-implemented. Some VSTs will fail if you feed them the wrong sample rate.

 

[kemmler3D]

I don't agree.
0 dBfs is same for both scales, all the way down to -96 dBfs both would be identical, at lower than that, 24 bit still has bits to work with, when 16 bits have run out!
But since noise is at around those levels, and the ear can not hear it either, 24 bits potential at reproduction is minimal.

I have, but a high refresh rate monitor is not what I was talking about. I was talking about high frame rate material.
A US DVD at 60Hz interlaced looks smoother than a PAL DVD at 50Hz, and in modern TV's, such as my Samsung Q95, higher settings of "Judder reduction" would make any movie look like a Camcorder footage!
I use it for Football. On movies, I bring it down, to get a little judder, to make it look like film.
Silly, but I prefer it.

Yes, but consider the fact that almost all signals are zero-crossing, so even the large-amplitude signals involve the <-100dB bits constantly.

Also, AFAIK 24-bit PCM files are usually linear, they don't encode according to dB, but rather in equally sized steps from -inf to 0dBFS. The additional steps yield a smaller minimum value that can be represented, (therefore lower noise floor) but the additional resolution is present between any two values compared to 16-bit.

I might be confused but I think this entails more accuracy at any given amplitude. (I think) It's just like bit-depth in colors. When you go from 8- to 16-bit color, you don't just gain more resolution in the dark colors, you can represent many colors in between each of the colors you could previously represent.

What we care about are the additional decibels of noise floor, but the "steps" are smoother throughout. It's just that that is not a limiting factor on sound quality until you get close to the noise floor.

 

[staticV3]

I have, but a high refresh rate monitor is not what I was talking about. I was talking about high frame rate material.

I, too, was talking about high frame rate material. Obviously, the same 50fps video will look identical no matter if displayed on a 50Hz display, or a 500Hz one.

But the human eye can process frame rates much higher than 50fps, otherwise the differences between monitor refresh rates wouldn't be as noticeable as it is.

 

[Ken Tajalli]

Yes, but consider the fact that almost all signals are zero-crossing, so even the large-amplitude signals involve the <-100dB bits constantly.

No No. the notion of a sinewave going positive and negative does not apply here. Bits sample loudness levels. The level never goes below zero.
By DC blocking after decoding, you get your familiar sinewaves or whatever. No crossover distortion here.
Somebody correct me.

Also, AFAIK 24-bit PCM files are usually linear, they don't encode according to dB, but rather in equally sized steps from -inf to 0dBFS. The additional steps yield a smaller minimum value that can be represented, (therefore lower noise floor) but the additional resolution is present between any two values compared to 16-bit.

No. the steps are same, on 24 you get more steps beyond.
Picture it as two ladders with same step spacing, one having 100 steps, the other 200 steps. One would be twice as long, but steps are same size.

I might be confused but I think this entails more accuracy at any given amplitude. (I think) It's just like bit-depth in colors. When you go from 8- to 16-bit color, you don't just gain more resolution in the dark colors, you can represent many colors in between each of the colors you could previously represent.

What we care about are the additional decibels of noise floor, but the "steps" are smoother throughout. It's just that that is not a limiting factor on sound quality until you get close to the noise floor.

I let someone else to chip in.

 

[amirm]

But mScaler tries to be not-that.
although 44.1k digital audio can absolutely manage 20kHz, but to get a clean high frequency, requires clever filtering. mScaler uses Rob Watts proprietary algorithm to cleverly guess extra inner-samples to intelligently upscale 44.1k many times.

There is absolutely nothing clever there, nor any guessing. He is simply using a longer filter. The filter is just multiply+add operations (and weighting). It has no intelligence or knowledge of what is being upsampled. DACs have the same filters but with shorter length.

The only thing "proprietary" is the window function he uses. At least that is his claim. Window functions are dime a dozen so I don't much any weight behind that either.

 

[SIY]

Strictly speaking I think you get more accuracy at any level with 24 vs. 16.

No, you don't That's counterintuitive for sure, but it's true. Noise decorrelates quantization.

The difference is noise floor, no more, no less.

 

[Soria Moria]

I take it you've never used a high refresh rate monitor before.
60Hz->120Hz is incredibly obvious.
120Hz->240Hz is still pretty noticeable.
240Hz->360Hz gets tricky to discern, but is doable if you're trained and the monitors' pixel response times are fast enough.

It’s hard to say how many frames we can discern on a monitor. I personally wouldn’t assume we can distinguish 120 Hz from 360 Hz if the frames are the only variable, but they aren’t. If you watch people who review displays (let’s use LinusTechTips for this example), I’m fairly certain I remember him during a review of an OLED display saying it has ‘perfect motion clarity’ or something. That is obviously not true. If you trail an object moving across the screen even on a 120 Hz OLED display it will look blurry. It’s because modern displays use a sample-and-hold technique. Blurbusters are the experts on this and it’s definitely worth reading about: https://blurbusters.com/faq/oled-motion-blur/

CRTs aren’t sample-and-hold which is why they have the fabled motion clarity (although they suffer from slight phosphor decay times).

Higher framerates means less persistence which means increased clarity. I’m certain that this is where the difference people see lies. However if this wasn’t a variable I would assume anything beyond 120 Hz isn’t possible to see for most people.

 

[Ken Tajalli]

There is absolutely nothing clever there, nor any guessing. He is simply using a longer filter. The filter is just multiply+add operations (and weighting). It has no intelligence or knowledge of what is being upsampled. DACs have the same filters but with shorter length.

The only thing "proprietary" is the window function he uses. At least that is his claim. Window functions are dime a dozen so I don't much any weight behind that either.

As I said, he claims that, or from what I have read from his posts, it appears he claims that.
He does not give too much info into the algorithm's inner workings. He has even implemented an anti reverse-engineering built into the FPGA code. Laugh all you like .
Whatever it does, I couldn't hear it! at times, I thought I could, but not enough to part with money.

 

[Nathan Raymond]

ok, so there are actually smaller steps between levels in 24 bit compared to 16 bit?

One thing - does 24 bit use more energy to convert?

 

[kemmler3D]

No No. the notion of a sinewave going positive and negative does not apply here. Bits sample loudness levels. The level never goes below zero.

Uh, PCM formats (well, all audio formats) definitely encode polarity, i.e. values below zero, just open any audio file in audacity. Either the middle value is chosen as zero or one of the bits encodes sign.

No, you don't That's counterintuitive for sure, but it's true. Noise decorrelates quantization.

The difference is noise floor, no more, no less.

Right, but isn't there an equal amount of quantization noise at any amplitude, and you just can't hear it except at low levels? How would that not be the case? (leaving aside dithering).

Picture it as two ladders with same step spacing, one having 100 steps, the other 200 steps. One would be twice as long, but steps are same size.

In integer PCM formats, you have 0dBFS and then either 65535 steps below that for 16 bits, or 16,777,215 steps below that for 24 bit. Amplitude goes from zero to 100%, it is not specified what that percentage is of. 24-bit has more steps in between zero and 100%. Except in logarithmic formats, (very rarely used for music) more bits are not allocated to a given amplitude range.

Either way, in the digital domain, the "length of the ladder" is imaginary or arbitrary, just the same as a pixelated drawing of a ladder. It has no real-world defined amplitude or length until you give it one on playback.

 

[SIY]

Right, but isn't there an equal amount of quantization noise at any amplitude, and you just can't hear it except at low levels? How would that not be the case? (leaving aside dithering).

"Leaving aside dithering" makes as much sense as "leaving aside anti-aliasing" or "leaving aside anti-imaging." It's part and parcel of digital audio.