[jrfmd]

In spirit yes. A big part of the Chord "secret sauce" is their algorithm to reconstruct as closely as possible the original analog wave form from digital data. As you go up the Chord product line from Mojo to Hugo to Hugo2 to DAVE, the sophistication of that filter goes up (as represented by "taps"...at a 100k' level more taps == better reconstruction)

The Blu2 takes the secret sauce algorithm which is built into the DAVE/Mojo/Hugo/etc. and executes it in a separate box. It is many times more powerful than even what's in the DAVE natively (1 million taps vs 150k taps) Chord has branded the 1M tap processing as "mScaler". mScaler applies to both digital data from the Blu2 CD player, as well as the USB and other digital inputs to the Blu2.

When connected to the DAVE, it essentially transforms the DAVE from a 150k tap device (which is already world class) to a 1M tap device (which at least to my ears, is completely transformational)

thanks, I appreciate your time
jeff

 

[jrfmd]

I would phrase it differently. See if this is helpful:

What does a digital PCM file give you? Essentially amplitudes (heights of the waveform) at fixed intervals, say 44,100 times a second.
If you draw this out, you won't get the original waveform back. You get a series of rectangles, 44,100 of them per second.

To render this into a waveform, you need filtering -- to smooth out the rectangles if you will.
This can be done entirely in analog (and NOS DACs do this), but getting high precision out of analog is hard.
So most designs upsample to a higher rate -- i.e. get more rectangles per second -- to make the final analog stage easier.

But how to do this upsampling? From Whittaker-Shannon interpolation, we know that at any point in the timeline can be calculated, assuming the original waveform that was sampled was < 1/2 the sampling rate (known as the Nyquist rate). This calculation requires convolution with an infinite sinc function. That's the challenge -- it is hard to do a calculation to infinity (except as a mathematical expression on paper)!

So inevitably, every actual realized upsampling process uses an approximation. And every approximation has some sort of error -- sometimes people trade off one type of filter for another bec they prefer the sound of one type of error over another.

The approach of DAVE, Blu2, etc. is to try to approximate the sinc function as closely as possible by using longer and longer filters. As the filter gets longer, it gets closer to the ideal infinite filter. The contribution of the Blu2 M-scaler is to add a really big 1 million tap length filter compared to the state-of-the-art before the M-scaler. The M-scaler outputs a 705.6/768kHz upsampled stream calculated with the 1MM tap filter.

So this is the problem preventing me from accepting this theoretical explanation of the blu m-scaler:

(BTW if they showed me a reconstructed sine wave with and without the m scaler I could see if there is any difference but on theoretical grounds the m-scaler doesn't make sense to me)

According to the nyquest theory, you cant "upscale" since you are not adding information when you double the rate of a sampled segment. This you simply creates pairs of identical rectangles for each single rectangle you had when you sampled the analogue signa at the original sampled rate. this doesn't reveal anything about what was happening to the analogue signal between samples (which is the whole point of the Nyquist theory=you have to SAMPLE twice as fast as you wish to reproduce the analogue signal (simply duplicating each sample you've already sampled is a waste of time)

I appreciate your message since it made clear to me what the blu is supposed to be doing. It may indeed sound much better than any other d/a converter but the explanation seems to be incorrect (at least to me) and I understand that a lot of smart people have found that it actually works in their system. My only problem is it seems a black box to me.

Jeff

 

[Crgreen]

My only problem is it seems a black box to me.

Jeff

It also comes in silver.

 

[rayl]

So this is the problem preventing me from accepting this theoretical explanation of the blu m-scaler:

(BTW if they showed me a reconstructed sine wave with and without the m scaler I could see if there is any difference but on theoretical grounds the m-scaler doesn't make sense to me)

According to the nyquest theory, you cant "upscale" since you are not adding information when you double the rate of a sampled segment. This you simply creates pairs of identical rectangles for each single rectangle you had when you sampled the analogue signa at the original sampled rate. this doesn't reveal anything about what was happening to the analogue signal between samples (which is the whole point of the Nyquist theory=you have to SAMPLE twice as fast as you wish to reproduce the analogue signal (simply duplicating each sample you've already sampled is a waste of time)

I appreciate your message since it made clear to me what the blu is supposed to be doing. It may indeed sound much better than any other d/a converter but the explanation seems to be incorrect (at least to me) and I understand that a lot of smart people have found that it actually works in their system. My only problem is it seems a black box to me.

Jeff

This may help with your interpretation:

Upsampling should *not* be performed by replicating the same sample value 16 times to go from 44.1 to 705.6. That is the point of the interpolation filter. (A very basic zero order hold filter would perform the replication, but outside of a class lab, such filters are not used in DACs...) The interpolation filter determines how to calculate the 16x value.

Actually one project I've always wanted to tackle is to create an illustration of going from 44.1 to 705.6 with each of the filters published for various DACs (though I don't know whether SABRE publishes their filter specs!) vs various lengths of sinc.... Alas, I no longer have access to MATLAB, etc. to whip this up easily. Maybe someone else can carry the torch!

(There is another way to view upsampling, in the frequency domain where sinc maps to a perfect brickwall rectangle, but I find that discussing it that way confused the non-EE audience more, so I've stopped offering that explanation...)

 

[Triode User]

It also comes in silver.

Also, it has rounded edges.

 

[Mython]

So this is the problem preventing me from accepting this theoretical explanation of the blu m-scaler:

(BTW if they showed me a reconstructed sine wave with and without the m scaler I could see if there is any difference but on theoretical grounds the m-scaler doesn't make sense to me)

According to the nyquest theory, you cant "upscale" since you are not adding information when you double the rate of a sampled segment. This you simply creates pairs of identical rectangles for each single rectangle you had when you sampled the analogue signa at the original sampled rate. this doesn't reveal anything about what was happening to the analogue signal between samples (which is the whole point of the Nyquist theory=you have to SAMPLE twice as fast as you wish to reproduce the analogue signal (simply duplicating each sample you've already sampled is a waste of time)

I appreciate your message since it made clear to me what the blu is supposed to be doing. It may indeed sound much better than any other d/a converter but the explanation seems to be incorrect (at least to me) and I understand that a lot of smart people have found that it actually works in their system. My only problem is it seems a black box to me.

Jeff

The whole universe is a 'black box', which we all go through life desperately kidding each other, and ourselves, we have a chance of understanding. LOL

 

[rayl]

Also, it has rounded edges.

And of course, "rounding" the rectangles from PCM is actually exactly what we are talking about!

 

[Crgreen]

The whole universe is a 'black box', which we all go through life desperately kidding each other, and ourselves, we have a chance of understanding. LOL

If that were true, how would you know?

 

[AndrewOld]

Ironic that you call this voodoo, since it's very simple electronics. It puzzles me immensely when people reject the best solution. RF filtering with ferrites isn't just a fad on a few Chord threads, it's a technique that's used widely in all sorts of industries and products where noise would otherwise cause problems. There are ferrites for RF filtering inside many hi-fi components, including Chord's.

It doesn’t seem to me simple electronics at all. Nearly every Blu2/Dave owner has a completely different set of ferrites and cable to deal with the same problem. If it is simple electronics then Chord should supply, or at least recommend, the correct cable with the correct number of correct ferrites. Or is that too simple? A Blu2/Dave costs £17k. Is it too much to expect that the correct interface between them should be specified?

 

[Crgreen]

It doesn’t seem to me simple electronics at all. Nearly every Blu2/Dave owner has a completely different set of ferrites and cable to deal with the same problem. If it is simple electronics then Chord should supply, or at least recommend, the correct cable with the correct number of correct ferrites. Or is that too simple? A Blu2/Dave costs £17k. Is it too much to expect that the correct interface between them should be specified?

That would be to admit there’s an issue, and Chord wouldn’t do that. Interesting that so far as I’m aware, there’s been no BluDave review yet, just some comments by a reviewer to whom it was demonstrated.

 

[Mython]

If that were true, how would you know?

Obviously, I meant metaphorically, in the sense of the great unknown.

Also, it has rounded edges.

Wait, are you saying Blu2 has rounded edges, or that the universe has rounded edges?



Seriously, though, back to Jeff, I'm sure Rob will weigh-in with a proper explanation soon.

 

[flyte3333]

That would be to admit there’s an issue, and Chord wouldn’t do that.

Not correct, Rob’s explained before:

“issue we have here is signal correlated RF noise from the actual BNC data corrupting Dave's ground plane, not Blu side noise getting through.”

https://www.head-fi.org/threads/cho...-official-thread.831343/page-87#post-13696701

And I don’t consider these issues - he’s just explaining the potential technical mechanisms that may explain the things we hear
. This can only lead to future improvements.

He’s a very open and honest guy and I hope he remains that way. We all lose if he decides to stop posting his thoughts.

 

[Crgreen]

Not correct, Rob’s explained before:

“issue we have here is signal correlated RF noise from the actual BNC data corrupting Dave's ground plane, not Blu side noise getting through.”

https://www.head-fi.org/threads/cho...-official-thread.831343/page-87#post-13696701

And I don’t consider these issues - he’s just explaining the potential technical mechanisms that may explain the things we hear
. This can only lead to future improvements.

He’s a very open and honest guy and I hope he remains that way. We all lose if he decides to stop posting his thoughts.

I’m sure that’s right. Rob is a specialist engineer whose services are used by Chord. What he says on these forums is not the same as the company admitting there are interface issues. I really can’t see Chord recommending certain ferrites to combat this problem.

I’m no expert, but it does seem to be an issue with Dave’s ground plane, possibly unforeseen at the time it was developed, prior to the m-scaler.

 

[Triode User]

That would be to admit there’s an issue, and Chord wouldn’t do that. Interesting that so far as I’m aware, there’s been no BluDave review yet, just some comments by a reviewer to whom it was demonstrated.

What has been said several times by Rob is that the performance of Blu2 is extraordinary even without ferrites and that any discussion of ferrites is mere fine tuning and is not fundamental or essential.

 

[flyte3333]

I’m no expert, but it does seem to be an issue with Dave’s ground plane, possibly unforeseen at the time it was developed, prior to the m-scaler.

Another example is the USB galvanic isolation of Dave (and every other high end USB DAC on the planet seemingly)...

John Swenson has only recently shown (after developing his own custom test gear) that there are “high impedance leakage currents” that sail right through digital isolators and transformers.

According to John, digital isolators and transformers typically do a great job filtering low impedance leakge currents but high impedance stuff sails right through. I’ve posted the links a few times before.

Sure, it would have been great to have known about this 5 years ago but that’s just how it is.

Dave 2 and Blu3 will be better and so on...

Dave 2 may even have it’s M-Scaler built in... who knows.

I'm grateful guys like Rob share things openly as they go along and learn. Rob's already said Davina will be an important tool going forward too, so that will fuel more innovation and improvements.