CHORD M-Scaler Review (Upsampler)

[Jomungur]

I remember people saying pass through is not as loud as without M and without M scaler the sound being better or preferable to passthrough, so don't think this pass through being worse is news.

OK, I didn't realize that. It's not easy to tell in a demo, because if you have the M Scaler in the audio chain, you can't just turn it off. You are forced to use its bypass mode or upsample. To compare bypass to w/o M Scaler, you have to disconnect and reconnect 2-3 cables every time you test.

 

[tmtomh]

There's a lot wrong with your post there. Human ears are less sensitive to lower frequencies. Amir has talked about this in one of his videos.

When you say "a sound has to be sampled only twice in order to be properly reconstructed" I don't think you understand what "properly reconstructed" means. It does not mean it's going to be perfect.

I'm not a sound engineer or electrical engineer and I know those two things. Don't be ignorant.

I'm sorry, but it appears that you don't know what it is that you don't know. A couple of points that might amplify or clarify my prior comment:

1. We are indeed less sensitive to low frequencies - but we are also less sensitive to high frequencies as well. So the variation in human hearing sensitivity has exactly zero impact on the point I was making.

2. Our hearing sensitivity is not linear, nor is it a simple up or down gradient from low to high frequencies. We are most sensitive between about 1-5kHz, and even in that narrow range of the greatest human hearing acuity, 1kHz sounds are always 5x higher "resolution" than 5kHz sounds because the 1kHz frequencies are sampled 5x as much, regardless of the sample rate. And once again, no one says 1kHz sounds are "higher res" or higher fidelity than 5kHz sounds in all our digital recordings.

3. Lest you think, "well, 5x as much sampling of 1kHz sounds vs 5kHz sounds isn't very much of a difference in sampling," remember that upsampling from 44.1kHz to 96kHz only represents an increase of about 2.2x in the sample rate, meaning most sounds will get sampled only 2x as much and some will get sampled 3x as much. So, as I noted previously, all of our digital recordings already have sounds that are "upsampled" compared to other sounds, even within the narrow range where human hearing is most sensitive.

4. Of course analogue reconstruction of a digital sample is not going to be perfect - that's why I wrote "properly reconstructed" rather than "perfectly reconstructed." But what you are missing is that the "imperfection" is only noise. This "imperfection" has absolutely nothing to do with sample rate. It has to do with the bit-depth. Now, it is true that if you upsample from, say, 16-bit/44.1kHz (CD quality) to 24-bit/96kHz (typical "high-res" quality), you will go from a bit depth that provides just over 96dB of signal-to-noise ratio to one that provides just over 144dB of signal-to-noise ratio. However, the problem with your claim there is fourfold:

a. First, 96dB is already plenty in real-world listening situations - if you listen on speakers in a very quiet home listening space, you will not be able to detect the difference between 16-bit and 24-bit reproduction. On headphones at very loud volumes you might, but still the chances are very slim.

b. Second, no stereo reproduction systems that I am aware of are capable of producing a S/N ratio equal to 24-bit. The most we can get is about 21-22 bit.

c. Third, the majority of 16-bit sources have been dithered down from 24 or 32-bit (it is useful to record, mix, process, and produce in higher bit depths because of all the processing steps, which can add many steps of recalculation/alteration of the original digital recording data along the way). And that dither is almost always noise-shaped, so that it provides an effective noise floor of -120dB rather than -96dB (by shifting some of the noise into those low and high frequency areas where human hearing is far less sensitive).

d. Finally, and most importantly for this particular discussion, increasing the bit depth via upsampling from 16-bit to 24-bit (or 32-bit, or 64-bit) simply adds a string of 8 zeros (or 16 zeros with 32-bit or 48 zeroes for 64-bit) to the original 16-bit PCM word for each sample. It literally does nothing to reduce the noise floor - only recording at higher bit depths in the first place will give you a lower noise floor. And again, if you record in 24-bit and then dither down to 16-bit using noise-shaping dither that's available even in consumer-grade and free/open-source digital audio software, you increase/degrade the noise floor from -144dB to -120dB, and both are beyond the range of human hearing so there is zero functional difference.

The bottom line is that higher sample rates do not increase fidelity, reduce distortion, or have any impact on noise. So I would say you are applying a misunderstanding (or I guess multiple misunderstandings) of digital sampling theory in making your argument.

I am not a professional engineer either, and as always I am happy to be corrected by any of our more knowledgeable friends here at ASR.

Thanks!

 

[sq225917]

Obvious troll is obvious.

 

[AdamG]

Thread Notice:

This is an Official Product Review Thread. Please keep your conversation firmly rooted in the Product Reviewed and/or the actual Bench Test Results/data. Several recent posts have been off topic and these posts then cause further thread drift. Please try as best you are able to get back on topic here. Further off topic posts may be deleted and may result in account sanctions. Please and thank you for your assistance and understanding.

 

[Doodski]

Off topic.. deleted.

 

[testp]

oops, replay below was for @Jomungur, about one of his dealers.. sry

sure, but there is something that most persons in most places will experience (regular forest even), that is lower ambient noise at evening/night due to most creatures taking it easy that time etc.. i hear my audio more clear for sure.

::
i'll also keep it on the topic from now, less traffic..

 

[AdamG]

Obvious troll is obvious.

Since you didn’t quote anyone, not sure whom you are referring? Best option is to either send us a pm or report the activity.

 

[sq225917]

Maybe not that obvious, i meant mccririck

 

[garbz]

What is your definition of fraud? When is it okay to sell things that don't work? When is it not?

If you come to me and ask me to create an upscaler for you, and I do, and it upscales (as in this case) it's not fraud. The device does work. It upscales.
If you come to me and say "I need you to make me something that improves sound" and I provide an upscaler it *is* fraud.

Fraud is only ever defined in a claim, not in a capability. The statement said "Is Rob Watts a fraud?" The answer is clearly no, he designs several products that do what they say on the box (and some of which perform very well too). Whether or not they are overpriced garbage isn't a deciding factor as to something or someone being fraud.

 

[raif71]

I'm sorry, but it appears that you don't know what it is that you don't know. A couple of points that might amplify or clarify my prior comment:

1. We are indeed less sensitive to low frequencies - but we are also less sensitive to high frequencies as well. So the variation in human hearing sensitivity has exactly zero impact on the point I was making.

2. Our hearing sensitivity is not linear, nor is it a simple up or down gradient from low to high frequencies. We are most sensitive between about 1-5kHz, and even in that narrow range of the greatest human hearing acuity, 1kHz sounds are always 5x higher "resolution" than 5kHz sounds because the 1kHz frequencies are sampled 5x as much, regardless of the sample rate. And once again, no one says 1kHz sounds are "higher res" or higher fidelity than 5kHz sounds in all our digital recordings.

3. Lest you think, "well, 5x as much sampling of 1kHz sounds vs 5kHz sounds isn't very much of a difference in sampling," remember that upsampling from 44.1kHz to 96kHz only represents an increase of about 2.2x in the sample rate, meaning most sounds will get sampled only 2x as much and some will get sampled 3x as much. So, as I noted previously, all of our digital recordings already have sounds that are "upsampled" compared to other sounds, even within the narrow range where human hearing is most sensitive.

4. Of course analogue reconstruction of a digital sample is not going to be perfect - that's why I wrote "properly reconstructed" rather than "perfectly reconstructed." But what you are missing is that the "imperfection" is only noise. This "imperfection" has absolutely nothing to do with sample rate. It has to do with the bit-depth. Now, it is true that if you upsample from, say, 16-bit/44.1kHz (CD quality) to 24-bit/96kHz (typical "high-res" quality), you will go from a bit depth that provides just over 96dB of signal-to-noise ratio to one that provides just over 144dB of signal-to-noise ratio. However, the problem with your claim there is fourfold:

a. First, 96dB is already plenty in real-world listening situations - if you listen on speakers in a very quiet home listening space, you will not be able to detect the difference between 16-bit and 24-bit reproduction. On headphones at very loud volumes you might, but still the chances are very slim.

b. Second, no stereo reproduction systems that I am aware of are capable of producing a S/N ratio equal to 24-bit. The most we can get is about 21-22 bit.

c. Third, the majority of 16-bit sources have been dithered down from 24 or 32-bit (it is useful to record, mix, process, and produce in higher bit depths because of all the processing steps, which can add many steps of recalculation/alteration of the original digital recording data along the way). And that dither is almost always noise-shaped, so that it provides an effective noise floor of -120dB rather than -96dB (by shifting some of the noise into those low and high frequency areas where human hearing is far less sensitive).

d. Finally, and most importantly for this particular discussion, increasing the bit depth via upsampling from 16-bit to 24-bit (or 32-bit, or 64-bit) simply adds a string of 8 zeros (or 16 zeros with 32-bit or 48 zeroes for 64-bit) to the original 16-bit PCM word for each sample. It literally does nothing to reduce the noise floor - only recording at higher bit depths in the first place will give you a lower noise floor. And again, if you record in 24-bit and then dither down to 16-bit using noise-shaping dither that's available even in consumer-grade and free/open-source digital audio software, you increase/degrade the noise floor from -144dB to -120dB, and both are beyond the range of human hearing so there is zero functional difference.

The bottom line is that higher sample rates do not increase fidelity, reduce distortion, or have any impact on noise. So I would say you are applying a misunderstanding (or I guess multiple misunderstandings) of digital sampling theory in making your argument.

I am not a professional engineer either, and as always I am happy to be corrected by any of our more knowledgeable friends here at ASR.

Thanks!

I'm intrigued by the point no 2 where you said 1khz sound is 5X sampled than 5khz or less res in 5khz. I've never saw this to be the case in my understanding of sampling theory but I'm beginning to see it now. I wonder if out there, there is a system that manage to standardize the sampling of various frequencies so that for all frequencies there would be same resolution ie in your example, 1khz and 5khz would have the equal number of samples and hence same resolution. Hmmm

 

[jhenderson0107]

I'm intrigued by the point no 2 where you said 1khz sound is 5X sampled than 5khz or less res in 5khz. I've never saw this to be the case in my understanding of sampling theory but I'm beginning to see it now. I wonder if out there, there is a system that manage to standardize the sampling of various frequencies so that for all frequencies there would be same resolution ie in your example, 1khz and 5khz would have the equal number of samples and hence same resolution. Hmmm

This characteristic does not result in any deliterious effect. If an input signal is band-limited to 20 kHz and sampled at 40 kHz or more, all frequencies within the captured 20 kHz bandwidth will be rendered perfectly in the data file.

 

[Blumlein 88]

I'm intrigued by the point no 2 where you said 1khz sound is 5X sampled than 5khz or less res in 5khz. I've never saw this to be the case in my understanding of sampling theory but I'm beginning to see it now. I wonder if out there, there is a system that manage to standardize the sampling of various frequencies so that for all frequencies there would be same resolution ie in your example, 1khz and 5khz would have the equal number of samples and hence same resolution. Hmmm

Think harder. I guess you could have multiple DACs sampling at every possible sample rate. And somehow combine that. Yet you don't know recording music what frequencies would be in play. Plus in short, there is nothing to be gained doing such a thing versus what is done now. Nothing.

What confuses people is errant vocabulary. Resolution has been confused with sample rate. Sample rate determines frequency response limits. Not resolution. Number of samples doesn't create additional resolution. Hi_res being high sample rate is a marketing thing.

 

[chris719]

This is certainly interesting. However, not a real proof that the S/PDIF signal level was a root of a problem with the D70. As a proof, you would need to attenuate the S/PDIF swing to 0.6V by means of a voltage divider and then, only then if the issue has disappeared, you might have said it was the issue without any doubts. As of now, it is only a hypothesis.

I have my doubts that this is the issue. I struggle to see how this would cause what is observed if the receiver is able to lock. Most SPDIF inputs tolerate logic levels even if it’s out of spec. Older receivers even performed better with logic levels. Not that it matters because this product sucks anyway.

 

[amirm]

But you do seem to keep investigating this product though, despite not having the time, so not an unreasonable comment from @theREALdotnet

Just saying

I hate you guys.

What he is asking requires a bit of soldering and BNC connectors for that use which I don't have around at the moment. So it is not a quick thing.

What was a quick thing was to adjust the output of my AP analyzer to same level of 1.84 volt. It made no difference to D70s. I even went up to 2.5 volt max output of AP and it still functioned.

Maybe it is the ultra fast rise time that is causing problems. Not sure now.

I edited my other post to reflect this.

 

[raif71]

I have my doubts that this is the issue. I struggle to see how this would cause what is observed if the receiver is able to lock. Most SPDIF inputs tolerate logic levels even if it’s out of spec. Older receivers even performed better with logic levels. Not that it matters because this product sucks anyway.

On the toleration of levels, perhaps you're right as @amirm stated that the Gustard seems to be ok. Which Gustard DAC was that, @amirm ?

 

[theREALdotnet]

Maybe it is the ultra fast rise time that is causing problems. Not sure now.

I still think this is a possibility, in conjunction with a suspected impedance mismatch.

The coax cable doesn’t have to be long to exhibit transmission line properties at the frequencies involved here. I seem to remember that a digital cable shows transmission line properties (such as signal reflections on impedance mismatch) if the propagation time isn’t at least 8 times shorter than the signal rise time (ballpark figure). I assume the rise time in S/PDIF signals is somewhere in the 25-30ns region (guessing, happy to be corrected), which means a coax cable like RG6 becomes a transmission line at just 75cm length. Not very long at all. Signal reflections will likely shift the threshold points of the signal edges sufficiently to introduce jitter.

This all only matters if there truly is an impedance mismatch, of course. We don’t yet know whether or not there is. Using a 50 ohm connector instead of the proper 75 ohm part alone isn’t going to have a huge impact. If, however, the M Scaler output is actually 50 ohm then this could play a role in how the D70 performs.

I suppose an eye diagram at D70 end of the cable, while loaded with the D70‘s input, might show reflections if there are any. Without soldering up a tap at the BNC plug, the easiest way would be opening the D70 and probing at the back of the S/PDIF connector.

Also, measuring the actual output impedance of the M Scaler would give a clue – loading with either a 50 or a 75 ohm resistor right at the connector should cut the peak-to-peak level exactly in half, compared with the unloaded level.

 

[voodooless]

If you come to me and say "I need you to make me something that improves sound" and I provide an upscaler it *is* fraud.

That is basically what they claim:

The Hugo M Scaler brings the unrivalled advantages of our ground-breaking FPGA-based WTA (Watts Transient Alignment) filtering technology to digitally connected audio devices, dramatically improving sound quality.

 

[solderdude]

I still think this is a possibility, in conjunction with a suspected impedance mismatch.

The coax cable doesn’t have to be long to exhibit transmission line properties at the frequencies involved here. I seem to remember that a digital cable shows transmission line properties (such as signal reflections on impedance mismatch) if the propagation time isn’t at least 8 times shorter than the signal rise time (ballpark figure). I assume the rise time in S/PDIF signals is somewhere in the 25-30ns region (guessing, happy to be corrected), which means a coax cable like RG6 becomes a transmission line at just 75cm length. Not very long at all. Signal reflections will likely shift the threshold points of the signal edges sufficiently to introduce jitter.

This all only matters if there truly is an impedance mismatch, of course. We don’t yet know whether or not there is. Using a 50 ohm connector instead of the proper 75 ohm part alone isn’t going to have a huge impact. If, however, the M Scaler output is actually 50 ohm then this could play a role in how the D70 performs.

I suppose an eye diagram at D70 end of the cable, while loaded with the D70‘s input, might show reflections if there are any. Without soldering up a tap at the BNC plug, the easiest way would be opening the D70 and probing at the back of the S/PDIF connector.

Also, measuring the actual output impedance of the M Scaler would give a clue – loading with either a 50 or a 75 ohm resistor right at the connector should cut the peak-to-peak level exactly in half, compared with the unloaded level.

One would need to use a scope at the input of the D70 after the input buffer (as to not capacitively load the cable at these high dV/dt) to see if such high risetimes do not cause some ringing that would accidentally trigger the H/L level detector. When the input level detector is designed to trigger at 0.4V differences and one overloads that circuit but with a higher than expected rise time who knows what that circuit will do at the rising/falling edges.
It could well be that when the level is reduced that same input circuit does not care about risetimes and reacts normally.

Maybe, just adding some capacitance to the input could be enough for the D70 to react properly ?

In any case this M-scaler might not deliver its promise with some DACs, possibly due to the too high level in combination with rising/falling edges in some DAC's input circuits.
For Chord products it will work fine.

Some info about using 50/75 ohm cables. It would seem that SPDIF isn't very sensitive to 50/75 ohm mismatching by itself.
Overloading an input may well be rise/fall time dependent.

I think the only way to check this without proper measuring is applying an attenuator to the input of the D70 reducing the input voltage from the M-scaler to expected levels.

The 'improved sound quality' is a typical audiophool promise.

 

[Ra1zel]

When you say "a sound has to be sampled only twice in order to be properly reconstructed" I don't think you understand what "properly reconstructed" means. It does not mean it's going to be perfect.

Trying to win against math? You pick hard battles I must say.

I'm not a sound engineer or electrical engineer and I know those two things.

Yes we can see that you are not clearly, just as clearly as the fact that you do not know those things.

 

[DonDish]

If you come to me and ask me to create an upscaler for you, and I do, and it upscales (as in this case) it's not fraud. The device does work. It upscales.
If you come to me and say "I need you to make me something that improves sound" and I provide an upscaler it *is* fraud.

Fraud is only ever defined in a claim, not in a capability. The statement said "Is Rob Watts a fraud?" The answer is clearly no, he designs several products that do what they say on the box (and some of which perform very well too). Whether or not they are overpriced garbage isn't a deciding factor as to something or someone being fraud.

Lol! Maybe he cheats on his wife That wouldnt make him a fraudster either. This product, what was electronic parts cost? 350 dollars of parts and some sloppy coding. If I bought this in good belief and learned what I know from this forum thread, i would cetainely feel defrauded.
Then again there is nothing new to this. HIFI industry has always put out scam products. The M-scaler reminds me of those Sansui home-fi reverb boxes of the eighties and nineties. They look cool, sound like shit and do absolutely nothing of value. At least they had a fancy display, the m-scaler has not.

Hold on to your hard earned cash ppl!