CHORD M-Scaler Review (Upsampler)

[balletboy]

Reading this thread, and your comments specifically, makes me wish that you would only have the option of listening in tightly controlled, level matched, multiple trial, blind listening tests where any rate of detection of difference below 90% would be as it should be, counted as invalid. Of course when you go and conduct sighted listening tests you are going to hear the difference you expect to hear. You already purchased Chord products so clearly you have confidence in the claims made meaning that you have a predisposition to hear a difference. If you really don't, great, prove it and participate in a true blind listening test. It should be very eye-opening and save you a buttload of money moving forward.

When I first heard the Vivaldi system it was because I walked into a random room at an audio show. The only product I saw was a pair of Wilson speakers (dCS use Wilson) and some large amplifiers (brand unknown). I'd never heard of Vivaldi as it was brand new, and the dCS people were sitting in front of it. So I did not know the source until I asked, and it could have been coming from some other sources they had there. So it was not controlled, but it was blind and not expectation biased. It was just one of those extremely rare 'Wow' moments I can count on one hand.

I did not own Chord at the time, I own a Mojo and have heard Dave several times. I may have an opportunity to listen to the M-scaler in about 2 weeks, but I have no intention of buying one, so there is no money to be saved. I'm just curious about it. I've not said the M-Scaler has any benefit as I've not heard it in use.

I just read what Rob Watt wrote - basically the high jitter is on purpose and irrelevant sonically for the low rate upsampling, and the M-Scaler is really intended to be used with x16 upsampling using the 2xBNC supplied into the Dave DAC. I assumed as much in an earlier post.

 

[Chester]

When I first heard the Vivaldi system it was because I walked into a random room at an audio show. The only product I saw was a pair of Wilson speakers (dCS use Wilson) and some large amplifiers (brand unknown). I'd never heard of Vivaldi as it was brand new, and the dCS people were sitting in front of it. So I did not know the source until I asked, and it could have been coming from some other sources they had there. So it was not controlled, but it was blind and not expectation biased. It was just one of those extremely rare 'Wow' moments I can count on one hand.

I did not own Chord at the time, I own a Mojo and have heard Dave several times. I may have an opportunity to listen to the M-scaler in about 2 weeks, but I have no intention of buying one, so there is no money to be saved. I'm just curious about it.

I just read what Rob Watt wrote - basically the high jitter is on purpose and irrelevant sonically for the low rate upsampling, and the M-Scaler is really intended to be used with x16 upsampling using the 2xBNC supplied into the Dave DAC. I assumed as much in an earlier post.

It’s not designed specifically for the DAVE.

 

[Rottmannash]

Please elaborate. First time I heard this term.

I believe he is referring to either/or.

 

[balletboy]

It’s not designed specifically for the DAVE.

Fine. Headfi owner suggested best with TT2 or Dave, but there is a happy user with the M-Scaler and Qutest - i.e. all the 2xBNC DACs.

Chord seem very popular with headphone users, which is why I bought one. I lost mine somewhere, was thinking of getting another (portable) one until buying a cheap pair of wireless headphones recently that I am happy with.

 

[Rottmannash]

When I first heard the Vivaldi system it was because I walked into a random room at an audio show. The only product I saw was a pair of Wilson speakers (dCS use Wilson) and some large amplifiers (brand unknown). I'd never heard of Vivaldi as it was brand new, and the dCS people were sitting in front of it. So I did not know the source until I asked, and it could have been coming from some other sources they had there. So it was not controlled, but it was blind and not expectation biased. It was just one of those extremely rare 'Wow' moments I can count on one hand.

I did not own Chord at the time, I own a Mojo and have heard Dave several times. I may have an opportunity to listen to the M-scaler in about 2 weeks, but I have no intention of buying one, so there is no money to be saved. I'm just curious about it. I've not said the M-Scaler has any benefit as I've not heard it in use.

I just read what Rob Watt wrote - basically the high jitter is on purpose and irrelevant sonically for the low rate upsampling, and the M-Scaler is really intended to be used with x16 upsampling using the 2xBNC supplied into the Dave DAC. I assumed as much in an earlier post.

The high jitter is "on purpose"???

 

[Somafunk]

You already purchased Chord products so clearly you have confidence in the claims made meaning that you have a predisposition to hear a difference.

I seriously doubt I’d be able to hear a difference between my mojo 2 and an apple dongle if we remove the mojo eq/crossfeed and db match the pair of them.

 

[A Surfer]

When I first heard the Vivaldi system it was because I walked into a random room at an audio show. The only product I saw was a pair of Wilson speakers (dCS use Wilson) and some large amplifiers (brand unknown). I'd never heard of Vivaldi as it was brand new, and the dCS people were sitting in front of it. So I did not know the source until I asked, and it could have been coming from some other sources they had there. So it was not controlled, but it was blind and not expectation biased. It was just one of those extremely rare 'Wow' moments I can count on one hand.

I did not own Chord at the time, I own a Mojo and have heard Dave several times. I may have an opportunity to listen to the M-scaler in about 2 weeks, but I have no intention of buying one, so there is no money to be saved. I'm just curious about it. I've not said the M-Scaler has any benefit as I've not heard it in use.

I just read what Rob Watt wrote - basically the high jitter is on purpose and irrelevant sonically for the low rate upsampling, and the M-Scaler is really intended to be used with x16 upsampling using the 2xBNC supplied into the Dave DAC. I assumed as much in an earlier post.

What Robb is saying could very well be true, but unless it creates an audible difference that can be reliably detected in properly controlled blind listening conditions, regardless of what is measured, it is absolutely moot. That is all I care about, separating claims and implied benefit from those that can be demonstrated with blind listening testing. The differences between the best DACs tested here and those that perform below SOTA doesn't matter at all in terms of audibility. Why we do like to see great engineering even if it becomes academic is that it brings with it a trickle down effect and improves the class of products altogether over time.

So even though the last few Topping DACs, for example, all measure transparently and have for several years unless I'm missing something, those are actually relatively affordable and can be used for a very important purpose in the audio chain. Something like the M-Scaler is by way of comparison obscene in terms of cost, and additionally, there does not appear to be any valid reason for it to exist at all. Other than to generate profit.

It is attempting to solve a problem that doesn't exist in any meaningful sense other than as a mathematical, theoretical construct that perhaps, just might, under the very most auspicious moments, align with a great many other factors, to possibly, maybe, create this tiny effect when the proper configuration of the stars and planets align. But of course you need to spend thousands of dollars to eek out that 0.215% improvement in sound, a theoretical improvement mind you, and just maybe and only if your cables are not holding your system back. But wait, what about the power coming into my house? How dirty is it? And my god, what about the clocks in my DAC? How many sources of inaccuracy are there? How can anybody enjoy music under such conditions? Don't you people know how much better your music sounds when you spend that additional $20 000? Why even bother otherwise? All of that digital noise, it is ruining the music and we all just put up with it? We let companies like Topping, and SMSL, Gustard, Schiit produce affordable gear that doesn't give us the benefits of the M-Scaler? I question why we even go on under such conditions.

I'm being deliberately ridiculous, but I would feel comfortable saying that creating a product like the M-Scaler designed to solve the "problem" that it purports to address, and for the price it sells for seems rather ridiculous as well. For the record I have owned a Mojo and was on the first review tour of the Hugo 2. This tirade wasn't directed at you balletboy. Your posts have been reasonable and you have not been making wild claims so this was really not aimed at you.

 

[Guermantes]

Why disappointed? He's saying the jitter is a red herring, which is true with Chord DACs at least.

I couldn't quite follow what he is saying in the italicized section below, it's too technical. Quote from Watts below. Anyone know how to explain it in laymen's terms (serious responses only please, not pointless insults).

"The Hugo M scaler does not have technical issues that IMO need fixing. The added quantisation noise is there simply because with 2x and 4x upsampling I use Gaussian dither set to a high level - and this is done simply because it sounds much better, closer to the noise shaped outputs at 16x. Of course using it in the intended mode, 16x, the noise and distortion level is identical to the input noise due to the aggressive noise shaping. So noise is higher, but so what? You still won't hear it, but you will hear the benefits of Gaussian against the usual triangular dither. That is, you will hear better depth and detail resolution, unless you are deaf, or not blessed with good hearing, or not interested in the SQ and musical performance, in which case why on earth would you be interested in an expensive up-sampler?"

He is talking about the probability distribution function of the dither noise values. There are three "shapes" that are generally used:

- rectangular: equal probability for all values;
- triangular: centre value has the highest probability
- gaussian: a classic bell-curve distribution that mimics analogue noise

Triangular is the one that is most used. Perhaps he believes that gaussian is more natural and analogue-like, I'm only guessing.

 

[amirm]

I couldn't quite follow what he is saying in the italicized section below, it's too technical. Quote from Watts below. Anyone know how to explain it in laymen's terms (serious responses only please, not pointless insults).

"The Hugo M scaler does not have technical issues that IMO need fixing. The added quantisation noise is there simply because with 2x and 4x upsampling I use Gaussian dither set to a high level - and this is done simply because it sounds much better, closer to the noise shaped outputs at 16x. Of course using it in the intended mode, 16x, the noise and distortion level is identical to the input noise due to the aggressive noise shaping. So noise is higher, but so what? You still won't hear it, but you will hear the benefits of Gaussian against the usual triangular dither. That is, you will hear better depth and detail resolution, unless you are deaf, or not blessed with good hearing, or not interested in the SQ and musical performance, in which case why on earth would you be interested in an expensive up-sampler?"

There is a major flaw in his argument. The content before it hits M-scaler is already dithered. M-scaler processes that data and to avoid quantization error, adds another layer of noise. In that regard, it degrades the signal to noise ratio of the system regardless of which form of dither it uses.

In other words, there is no free lunch here. The incurred loss of signal to noise ratio better be accompanied with higher fidelity which sadly, we don't have here.

That aside, his assertion is wrong about superiority of Gaussian noise relative to Triangular Dither. Here is the paper that is considered the bible of dither:
AES paper, DIGITAL DITHER 2412 (C-8)
Stanley P. Lipshitz
John Vanderkooy
Audio Research Group
University of Waterloo
Waterloo, Ontario,

As you see, triangular dither is superior in both performance and level of noise it adds to the system. This is why it has become the de facto standard in digital audio processing.

As you see at the bottom, the reason to use Gaussian is because it is simpler/cheaper to implement. Not that it is better. In a $6,000 device, I expect the better dither to be used, than the cheaper one that raises the noise floor yet again.

 

[Blumlein 88]

"The Hugo M scaler does not have technical issues that IMO need fixing. The added quantisation noise is there simply because with 2x and 4x upsampling I use Gaussian dither set to a high level - and this is done simply because it sounds much better, closer to the noise shaped outputs at 16x. Of course using it in the intended mode, 16x, the noise and distortion level is identical to the input noise due to the aggressive noise shaping. So noise is higher, but so what? You still won't hear it, but you will hear the benefits of Gaussian against the usual triangular dither. That is, you will hear better depth and detail resolution, unless you are deaf, or not blessed with good hearing, or not interested in the SQ and musical performance, in which case why on earth would you be interested in an expensive up-sampler?"

There are different types of dither, and they have different noise floors. Rectangular is pretty much an even level of noise across the frequency spectrum. Triangular dither is very slightly sloped upward with more noise at higher frequencies and just past the top frequency along with less noise in the lower frequencies. Gaussian dither is similar to this with a noise floor that goes up with frequency. So he is saying above 20 khz his choice of gaussian dither causes a higher noise floor at lower upsampling. With higher upsampling to 16x the 20 khz range is low enough in the total bandwidth it is in the lower noise region of gaussian dither so you don't see the higher noise floor just above 20 khz. I see Amir and Guermantes have replied, but I'll leave my less technical and slightly incorrect explanation.

BTW, one can frequency shape the dither to greatly lower noise in one part of the spectrum and raising it (or placing it) elsewhere in the spectrum. For instance lowering noise in the 3-5 khz range and raising noise in the above 10 khz range we don't hear well. Doing this it is possible to use 16 bit files which can really record signals down to -120 db or a bit more.

 

[amirm]

BTW, one can frequency shape the dither to greatly lower noise in one part of the spectrum and raising it (or placing it) elsewhere in the spectrum. For instance lowering noise in the 3-5 khz range and raising noise in the above 10 khz range we don't hear well. Doing this it is possible to use 16 bit files which can really record signals down to -120 db or a bit more.

Strangely he says he is only noise shaping at 16X multiplier. There is plenty of room to noise shape 88 and 176 kHz.

 

[Tachyon88]

Its getting to a point where the word "measurements" is considered hate speech over at the other form! lol, the seething is intense.

 

[thin bLue]

CHORD....what have you done...

 

[raif71]

CHORD....what have you done...

seems to have some sort of dischord effect between people

 

[Blumlein 88]

Strangely he says he is only noise shaping at 16X multiplier. There is plenty of room to noise shape 88 and 176 kHz.

I noticed that too which didn't make any sense. Even at 1x you can used shaped dither, and at 2x and 4x there is lots of room for placing noise out of band. Plus he said at 2x and 4x he used Gaussian dither set to a high level. I thought one of the benefits of Gaussian dither if you choose it was that you could use a lower level of dithering. But I'm no expert on that.

 

[Jomungur]

Thanks, Guermantes, Blumlein, Amir and others for answering my question on Watt's quote on dithering. It's starting to make more sense, although not completely. I need to study this further.

I'm understanding all of this correctly, then:

1- Take an album on Qobuz that has a 48khz version and a 192khz version.

2- Presumably the 192khz version was recorded in that format. Call the original192khz version Output A.

3- If you run the Qobuz 48khz version into the M Scaler at a 4x sample rate, the M Scaler will output a 192khz version. Call this upsampled 192khz version Output B.

4- Because of the way the M Scaler processes the original 48khz recording, Output B will *not* be the same as Output A, even though both are 192khz versions of the same album. In fact, it is likely to sound different because of the changes of dithering, possible noise, etc.

Is that right? If so, that goes against how many people are thinking about the M Scaler (I'm not talking about people on this forum). I believe many people think the M Scaler works so that Output B is actually the same as Output A, giving you the promise of getting 192khz versions of songs as though they were recorded in that format from 48hz original streams. That's what I thought anyway when it was first described to me. A sort of sonic alchemy; even better than the real thing (sorry listening to U2 right now). I do not think people would purchase the M Scaler if they realized Output B is actually different than Output A.

 

[ExUnoPlura]

Slightly off topic but of interest to me is the newest issue of Stereophile (August 2022) that takes multiple passing swipes at ASR (maybe Archimago, who knows). Page 3 we have a discussion of thoughts following meeting the Purifi team at High End Munich:

Purifi’s people are serious engineers, but don’t make the mistake of thinking they mindlessly ally themselves with the simple-minded objectivists who populate certain online discussion forums. They understand that science and engineering must be carefully deployed. Members of the Purifi team seemed to me as impatient with measurements-happy reductionists as with the radical antimeasurements crowd. Sure, it’s depressing when a forum post asks, “Is it possible for a component that measures well to sound good?” But owning an Audio Precision analyzer doesn’t make you an expert, and the unguided application of analytical tools can lead you far astray.

This gets a follow-up emphasis in the review of Genelec G Threes by Herb Reichert:

I was impressed that a globally respected manufacturer of professional-grade active loudspeakers chose not to loudly declare their obsession with accuracy, science, and measurements, choosing instead to present a poem by a 19th century Romantic poet.

And this is shortly after a stirring collection of thoughts about cables and an on-site interview with measurement professionals who work out of a very cool nuclear reactor facility. Perhaps @John Atkinson can enlighten us about the editorial goals of Stereophile that traditionally has been a balance of measurement and listening perspectives. Is it just protecting the fiefdom or is there a goal to this mishmash of perspectives?

 

[aj625]

Main purpose of m scaler is to filter and upsample to "16x" so that filtering of dac is mostly bypassed. If you do 2x or 4x then filtering of dac comes in picture. That's why i asked attenuation at 16x with "compatible dacs". Non chord dacs are not compatible with dual bnc so 8x could have been used for those dacs.

 

[amirm]

4- Because of the way the M Scaler processes the original 48khz recording, Output B will *not* be the same as Output A, even though both are 192khz versions of the same album. In fact, it is likely to sound different because of the changes of dithering, possible noise, etc.

It will never be the same because a part of the 192 KHz spectrum was chopped off when they created 48 kHz. Nothing in this world would get you back that lost spectrum. An idealized M-Scaler would simply do no harm other than forcing the DAC to operate at higher sample rate. An non-ideal version which is what we have, is adding more noise to the 48 kHz version, making it a worse copy yet again.

 

[Rottmannash]

My first post in the thread was basically asking why he only conducted the listening test with 2x upsampling when he could also have tried 4x and 16x.

Did you ask him? If not please do so we can move on from this circle of hell. I'll do it for you- @amirm can you please address why you didn't conduct a listening test with 4x and 16x? @spooky really wants to know. edit-never mind-I noticed @amirm answered his questions on a previous page.