Hugo M Scaler by Chord Electronics - The Official Thread

[musickid]

Very lovely and homely set up Nick. Spend money where it matters and leave the multi K anti vibration stands out in the cold.

 

[musickid]

JAZZ SAMBA, STAN GETZ, CHARLIE BYRD.

 

[birdlandbill]

A new interview from Rob Watts in four parts, check it out...

This 4 part interview is outstanding! Thank you for posting this here.

Love the interviewer's modesty and honesty. Mr. Watts is uniquely generous with his time and explanations. Was pleased to hear Matlab mentioned and also interested his mention of the interpolation filter used to transition to zero (Kaiser or Kaiser-Bessel interpolation filter). Much to digest here and a great point of departure for further exploration of the maths behind it all. I really enjoyed the interviewer's expression of astonishment that there is still not a consensus on the maths to achieve best resolution information recovery with DAC. It's true and probably all bound up in implementation costs/points of emphasis/etc. Maybe someday. In the mean time, we can enjoy Mr. Watts' particular and wondrous implementations.

 

[ruthieandjohn]

Here is my summary of the 92 minute interview of Rob Watts captured in the four-video sequence "Sound Decisions" on the previous page:

MAIN POINTS (elaborated below)

• Sampling theory holds that you can perfectly reconstruct a band-limited signal if you use a perfect sinc function, but you cannot implement a perfect sinc - hence, choices to make
• Psycho-acoustic experiments show that the brain distinguishes timing information as short as 4 usec. Brain uses this info for pitch, timbre, direction, and placement
• If the human ear only hears up to 20 KHz (or less), why do we need higher frequencies? Music listening is not about sine waves, but about timing of transients.
• (Nearly) all DACs exactly replicate the signal level at each sample time, but it is what is done between sample times (interpolation) that affects listening fidelity
• Feeding Mscaler with 44.1KHz CD signal > Roon PC-upsampled input; Qutest + Mscaler > Hugo TT2; Mscaler + Hugo TT2 more muslcal than DAVE alone (which is more transparent)

VIDEO 1

• Two kinds of DACS - R2R (resistor ladder - do nothing particularly well, but easy to design -- have a lot of "glitch" energy (error signal between actual and reconstructed signal)) vs. DS (delta sigma - hard to design noise shaper)
• Two kinds of DS - 1 bit DSD (modulator easily goes unstable, use began around 1995) vs. n-bit DSD (4 - 5 bits at once, exemplified by Cirrus, Sabre, AK - 3 MHz noise)
• DSD requires noise shaping - can use MATLAB to create coefficients, but then have to program in DSP or in FPGA (or in custom chip)
• R2R requires all resistors to have tolerance well under 0.01% (0.01% only gives 12 bit accuracy; Watts seeks 64 bit accuracy) but no noise shaping filter is required - its error signal power is constant regardless of signal level
• DSD error is proportional to signal level and hence scales down with softer sounds; also has less high frequency content than R2R noise

VIDEO 2

• Rob began work with pulse density modulation (PDM) chip from Philips in 1989 (a DS technique) - used 256 1-bit pulses to encode a 20 KHz signal - discovered warmer and better sound stage
• DSD only has 1 resistor - no need to match a ladder
• To eliminate signal switching activity, use pulse array scheme: multiple elements with some rising and some falling, leading to noise cancellation
• For example: pulse width modulation representation of a zero level could have five + signals and five 0 signals
• For 4-bit DSD on 32 bit signal, you take the top 4 bits, then use 28 bits of error feedback to compare to (subtract from) input
• If you cannot accurately reproduce small signals, you cannot accurately capture depth
• For Hugo 1, Rob needed to add crossfeed, but had used 95% of the FPGA capability and had to slim down the design, while maintaining -200 dB noise shaper -- -200 dB should not make a difference, but did, based on listening
• DAVE has -330 dB to -350 dB, still with audible differences. In fact, if there is a known distortion, however small, it shows up by listening

VIDEO 3

• (Nearly) all DACs exactly replicate the signal level at each sample time, but it is what is done between sample times (interpolation) that affects listening fidelity
• You can perfectly reconstruct a band-limited signal if you use a perfect sinc function (sampling theory), but you cannot implement a perfect sinc. Chord tries with up to 1M taps
• Psycho-acoustic experiments show that the brain distinguishes timing information as short as 4 usec. Brain uses this info for pitch, timbre, direction, and placement.
• If the human ear only hears up to 20 KHz (or less), why do we need higher frequencies? Music listening is not about sine waves, but about timing of transients.
• Example of transient error: imagine a sine wave that starts at Time 0. Reconstruct that wave from two samples per period... you will find that its onset becomes delayed, as a function of frequency.
• The sampled sinc function rings on either side of its peak (past and future time). Other designers suppress forward ringing, ending up with smeared transients.
• Various designers use various filters... minimum phase, linear phase, etc. Only Rob Watts uses sinc filters.
• Rob considers the Kaiser filter as the best second place to his WTA, and states that a 256-tap WTA sounds better than a 2,000-tap Kaiser filter.

VIDEO 4

• We have no idea how the human separates instrument sounds or focuses on one instrument - hence, we cannot write a computer program or make measurements for doing this
• Sometimes you can hear what you measure, but you cannot always measure what you hear
• Example: Chord Qutest DAC has white/green and orange/red filter choice - both measure the same, but sound different. One applies the WTA an additional time to move from 1.3 usec resampling to 88 nsec.
• Noise floor modulation is unmeasurably low on Chord; measurable on other DACs. Noise floor modulation can corrupt a sax sound with white noise that follows the level of the sax sound - brain struggles to separate
• Rob is trying to make USB input sound as good as optical input, via galvanic isolation, etc. USB gives 10 MHz RF noise which inserts intermodulation distortion into the audio range
• Q: What is better, to feed Chord upsampling DACs a 44.1 KHz CD signal, or a signal upsampled by Roon or JRiver? A: Redbook -- Chord will upsample better than the PC.
• Q: What do you think of MQA? A: Poor - uses a minimum phase filter of limited tap length and ignores pre-ringing. However, MQA can cover the flaws of a poor DAC
  
• Q: Which better -- upgrade the Qutest plus amp by adding Mscaler, or using a Hugo TT2 (both cost the same)? A: Add Mscaler - adds a new complementary dimension.
• Q: Which upgrade to the Qutest - DAVE or Hugo TT2 plus Mscaler (both cost the same)? A: Mscaler & Hugo TT2 - more musical; DAVE will be more transparent.

I hope that my summary helps you decide whether to watch the 92-minute interview of Rob Watts.

 

[ZappaMan]

Here is my summary of the 92 minute interview of Rob Watts captured in the four-video sequence "Sound Decisions" on the previous page:



MAIN POINTS (elaborated below)

• Sampling theory holds that you can perfectly reconstruct a band-limited signal if you use a perfect sinc function, but you cannot implement a perfect sinc - hence, choices to make
• Psycho-acoustic experiments show that the brain distinguishes timing information as short as 4 usec. Brain uses this info for pitch, timbre, direction, and placement
• If the human ear only hears up to 20 KHz (or less), why do we need higher frequencies? Music listening is not about sine waves, but about timing of transients.
• (Nearly) all DACs exactly replicate the signal level at each sample time, but it is what is done between sample times (interpolation) that affects listening fidelity
• Feeding Mscaler with 44.1 CD signal > Roon PC-upsampled input; Qutest + Mscaler > Hugo TT2; Mscaler + Hugo TT2 more muslcal than DAVE alone (which is more transparent)

VIDEO 1

• Two kinds of DACS - R2R (resistor ladder - do nothing particularly well, but easy to design -- have a lot of "glitch" energy (error signal between actual and reconstructed signal) vs. DS (delta sigma - hard to design noise shaper)
• Two kinds of DS - 1 bit DSD (modulator easily goes unstable, use began around 1995) vs. n-bit DSD (4 - 5 bits at once, exemplified by Cirrus, Sabre, AK - 3 MHz noise)
• DSD requires noise shaping - can use MATLAB to create coefficients, but then have to program in DSP or in FPGA (or in custom chip)
• R2R requires all resistors to have tolerance well under 0.01% (0.01% only gives 12 bit accuracy; Watts seeks 64 bit accuracy) but no noise shaping filter is required - its error signal power is constant regardless of signal level
• DSD error is proportional to signal level and hence scales down with softer sounds; also has less high frequency content than R2R noise

VIDEO 2

• Rob began work with pulse density modulation (PDM) chip from Philips in 1989 (a DS technique) - used 256 1-bit pulses to encode a 20 KHz signal - discovered warmer and better sound stage
• DSD only has 1 resistor - no need to match a ladder
• To eliminate signal switching activity, use pulse array scheme: multiple elements with some rising and some falling, leading to noise cancellation
• For example: pulse width modulation representation of a zero level could have five + signals and five 0 signals
• For 4-bit DSD on 32 bit signal, you take the top 4 bits, then use 28 bits of error feedback to compare to (subtract from) input
• If you cannot accurately reproduce small signals, you cannot accurately capture depth
• For Hugo 1, Rob needed to add crossfeed, but had used 95% of the FPGA capability and had to slim down the design, while maintaining -200 dB noise shaper -- -200 dB should not make a difference, but did, based on listening
• DAVE has -330 dB to -350 dB, still with audible differences. In fact, if there is a known distortion, however small, it shows up by listening

VIDEO 3

• (Nearly) all DACs exactly replicate the signal level at each sample time, but it is what is done between sample times (interpolation) that affects listening fidelity
• You can perfectly reconstruct a band-limited signal if you use a perfect sinc function (sampling theory), but you cannot implement a perfect sinc. Chord tries with up to 1M taps
• Psycho-acoustic experiments show that the brain distinguishes timing information as short as 4 usec. Brain uses this info for pitch, timbre, direction, and placement.
• If the human ear only hears up to 20 KHz (or less), why do we need higher frequencies? Music listening is not about sine waves, but about timing of transients.
• Example of transient error: imagine a sine wave that starts at Time 0. Reconstruct that wave from two samples per period... you will find that its onset becomes delayed, as a function of frequency.
• The sampled sinc function rings on either side of its peak (past and future time). Other designers suppress forward ringing, ending up with smeared transients.
• Various designers use various filters... minimum phase, linear phase, etc. Only Rob Watts uses sinc filters.
• Rob considers the Kaiser filter as the best second place to his WTA, and states that a 256-tap WTA sounds better than a 2,000-tap Kaiser filter.

VIDEO 4

• We have no idea how the human separates instrument sounds or focuses on one instrument - hence, we cannot write a computer program or make measurements for doing this
• Sometimes you can hear what you measure, but you cannot always measure what you hear
• Example: Chord Qutest DAC has white/green and orange/red filter choice - both measure the same, but sound different. One applies the WTA an additional time to move from 1.3 usec resampling to 88 nsec.
• Noise floor modulation is unmeasurably low on Chord; measurable on other DACs. Noise floor modulation can corrupt a sax sound with white noise that follows the level of the sax sound - brain struggles to separate
• Rob is trying to make USB input sound as good as optical input, via galvanic isolation, etc. USB gives 10 MHz RF noise which inserts intermodulation distortion into the audio range
• Q: What is better, to feed Chord upsampling DACs a 44.1 KHz CD signal, or asignal upsampled by Roon or JRiver? A: Redbook -- Chord will upsample better than the PC.
• Q: What do you think of MQA? A: Poor - uses a minimum phase filter of limited tap length and ignores pre-ringing. However, MQA can cover the flaws of a poor DAC
  
• Q: Which better -- upgrade the Qutest plus amp by adding Mscaler, or using a Hugo TT2 (both cost the same)? A: Add Mscaler - adds a new complementary dimension.
• Q: Which upgrade to the Qutest - DAVE or Hugo TT2 plus Mscaler (both cost the same)? A: Mscaler & Hugo TT2 - more musical; DAVE will be more transparent.

I hope that my summary helps you decide whether to watch the 92-minute interview of Rob Watts.

Good job

 

[gnomen]

For noise treatment, has anyone tried the passive ground control devices such as the CAD (Computer Audio Design) GC1 and GC3?

 

[ObjectiveSubjectivist]

Has anyone had similar problem to me ? Sometimes I got drop out of signal for like 0,5 second then everything comes back to normal.
I'm using m scaler with TT1 via SPDIF bnc.
It happens on stock BNC cable and also on third party cable. I didn't find any pattern when this is happening. Just random drop outs.

 

[Progisus]

Has anyone had similar problem to me ? Sometimes I got drop out of signal for like 0,5 second then everything comes back to normal.
I'm using m scaler with TT1 via SPDIF bnc.
It happens on stock BNC cable and also on third party cable. I didn't find any pattern when this is happening. Just random drop outs.

I had that with mscaler out through bnc to TT1. I traced it down to the stock cable not making a good enough connection at the TT1. A different bnc cable fixed it. I see you have the problem on with other bnc cables though. It could be the connection at the TT1. Try releasing the tension on the connection if possible.

 

[musickid]

brilliant breakdown of all the main theory and tech.

 

[CreditingKarma]

First - happy holidays all!

I've been using the holiday to further refine my front end, because I just can't keep well enough alone. A few posts back I had spoken a buying a new streamer / server - but decided to hold back - mainly because no matter what I try, I get the best result from Toslink. (I am currently running MScaler > DAVE with WAVE BNC Cables.)

It's really strange about Toslink. Everyone hates it except those who have Chord (and potentially PS Audio) DACs - because they have good jitter rejection. I assume most modern Dacs have good jitter rejection (I have no idea if it's as good as chord, although consistently toslink sounds the best for me now).

With the added resolution from M Scaler, I feel that it's even more important to mitigate RFI interference. Toslink does that best. I'd like to hear USB outputs from an Auralic Aries G2, a Lumin U1, or a Innuous Zenith MkIII to see if they've minimized EMI/RFI so that it's not discernible. I know Innuous says the Toslink receiver generates more noise - but the USB receiver is still a more complex protocol so that has to generate noise as well, no?

Anyhow - what I've decided to do before making a huge investment is a server is to get the absolute best Toslink output I possible can - and see if I can discern any difference.

I have an Auralic Aries Femto and Auralic Aries Mini. I had been using the Mini's toslink, because my particular Femto can't get to 192khz over toslink. The Mini toslink sounds great. But it can't do DOP. This isn't a huge deal in and of itself. I convert DSD to PCM in Roon, and away we go.

But what bothered me is why it can't do DOP. Did Auralic block DOP on the mini, or is it something else. It's something else. The Auralic Aries Mini doesn't send out bit perfect data via SPDIF. I confirmed this by encapsulating a dsd album in PCM with DB Poweramp and attempting to play it back in Roon. This, admittedly, pissed me off. I have all processing off in Lightning DS (the Aries control app), and no DSP in Roon. Yes, it's not the end of the world to not output bit perfect data. But I want to control that - and I want to be able to know if my music is going to my DAC bit perfect or not. I believe this is the case with Aries streamers that have a built in DAC (altair, altair g1) - because it's applying some type of filter to the digital outputs before it gets to the DAC (likely the same filter it applies to it's internal DAC, as it's on at the same time). There is no issue running DOP with the Auralic Aries Femto (via Coax at least, because the optical can't get to 192)

So this led to a rabid search for a Toslink that can run bit perfect at 192khz. In terms of USB to SPDIF converters I currently have a Gustard U12 lying around and a BelCanto Reflink. The Bel Canto Reflink is older, but doesn't have toslink out. It has ST Fiber out, AES, and BNC. BNC works to 192, does DOP - and sounds good - but not as good as Toslink. It also seems like the gain is significantly reduced vs my other sources. Not sure why this is. Interestingly, the Lifatec website has a cable with an ST Fiber connector on one end, and toslink on the other. They say that ST Fiber and Toslink were never intended to work together - but sometime they do! I also read that some people have take an ST Fiber cable and stuck it in a toslink input - and it worked! I ordered a cheap ST fiber cable just to test if M Scaler can lock onto anything. If it does, maybe I'll buy that Lifatec cable.

So last night I listened to Aries Femto USB > Gustard U12 Optical > M Scaler. It had the same warmth / tone of Toslink, but sounded ever so slightly different. One of my favorite albums is Morph the Cat by Donald Fagen - which I know backwards and forwards. I noticed two main differences - mainly in regards to detail. First, I noticed some horn trills that I hadn't previously noticed. I can't imagine missing them before. Secondly, the vocals seems clearer, but not sharper. All of these were quite small differences so could be placebo - or could be because I was sending a bit perfect signal via toslink for the first time!

Next I listened to Beethoven Around the World: Vienna - Op.59 Nos. 1&2. It's a great recording. When listening this time through this chain, it actually sounded worse in some ways. The cellos at the beginning, previously, really had a 3 dimensional characteristic which was flattened somewhat. The violins sounded "wetter" - which is the best way to describe it - because they sounded drier previously.

I have noticed this wetter sound I've typically associated with more RFI, but I don't know how it could be getting in. From a mains standpoint, I moved power supplies for the Aries and Gustard to a different circuit anyway, as they are completely isolated from the M Scaler > Dac > Amp - since their only connection is Toslink.

I don't frequent Audio Science Review frequently, but the Gustard is supposed to be a bit jittery via coax, so I can only imagine toslink is a bit worse. Could it be that there is less jitter on my aries mini, and this overshadows it's outputting a signal that's not bit perfect?

So I've decided to up my toslink game to see if jitter can play any role. I just ordered a Matrix X SPDIF 2. This seems to be the current USB to SPDIF hotness. Raved about on CA, even ASR says it measures great. Most importantly @ray-dude uses it and likes it. If I run Aries Femto USB > X SPDIF 2 - will this give me a great toslink output? Supposed to receive tomorrow so I'll find out.

I know that Chord Dacs are supposed to be 99% immune to jitter - but maybe that 1% is making a difference - or maybe I am just a crazy audiophile. Nonetheless, I'd like to get an output that sounds as good as my Aries Mini toslink, that is bit perfect - so I think it's a worthwhile exercise.

Have you had a chance to audition the Aries G2 in you system? I picked up an Aries G1 and found it to make more improvement in my system than the mscaler. I decided to upgrade it to the G2. The G2 was an even bigger step up. More separation of instruments, deeper soundstage, blacker background, anymore realistic vocal textures.

I am not sure if I am the only person that feels this way but I owned an mscaler wit a hugo 2 last year. IMO the mscaler made more of an improvement with the hugo 2 than it did with the TT2.

 

[Ards]

Those videos were excellent. The thing that struck me, as a vinyl lover, is that the ultra important transient details that Rob is trying to recreate with the M Scaler are all there in the vinyl groove. Perhaps Rob has inadvertently found the explanation for why (all analogue pressings) vinyl sounds so good...

They also convinced me that we are correct to follow Rob's path for digital, as I have been since first hearing the Mojo some years ago. Now, if I could only fit a DAVE in my rack, but I'm all out of space :/

 

[Uncle Monty]

Interesting what Rob says about R-2R DACs - The Denafrips DACs have been well reviewed by John Darko and Steve Guttenberg and they are R-2R I believe (not that I know anything about how these things work)

 

[GreenBow]

Daft question for M-Scaler owners.

Using the OPTO-DX there was discussion about needing separated power supplies for either side of the bridge. Or battery power.

However I had an idea which is probably stoopid and not going to be good enough. Has anyone tried powering both sides of the Opto-DX with an iFi Audio DC iPurifier2.


There was relative success with the Audioquest jitterbug for example. Some folk swear by them, and others not. Some folk sing praises of the iFi Audio DC iPurifier2 too. The iFi Audio DC iPurifier2, should definitely clean the power input to the last half of the OPTO-DX bridge. I am not sure it would clean anything on the first half of the OPTO-DX.

However it's got to be worth a shot trying it. If it cleans the input power supply to last half of the bridge. It should stop noise using the PSU, or mains strip supplying two PSUs, to travel across. If that iFi Audio DC iPurifier2 is efficient and of high performance, for £100 it could be a result.


https://www.futureshop.co.uk/ifi-audio-dc-ipurifier2

 

[dmance]

@GreenBow
Not daft.
OptoDx is a galvanic-isolation air gap. By using the same PSU to power both sides defeats this. No matter how low noise a single PSU is, the fact is that a common supply provides a conductive metal (galvanic) connection across the optical bridge ..exactly what you are trying to avoid.
Optodx sips power, <100mA,so it can be used to share, via a Y split, the power for Mscaler or your DAC...but you need a separate PSU for each side.
PM me if you want to delve further...
Dan

 

[Koolkef]

Dear All,

It has been a wonderful journey reading your experiences. I am on holiday and I found an audio shop where I bought an M Scaler just from reading how it changed the listening experience on this forum.

I am currently using a used Chord Hugo 2 which is close to 3yrs old and I realise that the battery doesn't last as long as it should.

I have some questions which I would appreciate your view. Should I invest in a Hugo TT2 or get another Hugo2 (I found a deal where the Hugo 2 + an Elear) is cheaper than a brand new Hugo 2 in my country. I could then use the "old" H2 in desktop mode connected to the M Scaler and the new H2 (with a 2Go) and use it to travel with or with Roon as an endpoint like what I am doing with my MojoPoly.

Is the upgrade to the TT2 worth the investment as it’s really expensive and also since the Hugo 2 can handle everything the m Scaler sends to it (768khz)?

Another thought is just save for a good pair of cans. I am currently using an LCD2, AEON 2 Closed, Drop 6XX & Drop EE Zeus.

Thank you in advance for your kind feedback and advice. This is truly a great forum.