Just got a Yggdrasil!

[esldude]

1 hour ago, crenca said:

 

This is because the quantization noise is being pushed way up higher than the audible band (i.e. noise shaped), but a the wide band amp is working hard to amplify that signal right?  Which reminds me to ask what is the point of a wide band amplifier?  Anyone know the theory that argues for their advantage(s) or point me to a link?

Depends on what you are calling wide band.  Conventional transistor amps will have a first order, 6db per octave roll off.  That means you have to be -3db down at 200 khz to manage being only .1 db down at 20 khz.  So most good solid state amps should have response to 150khz or 200 khz.  You might wonder about filtering, and that could be done.  However once you get more than a first order rolloff you have the chance of resonance and more so using feedback could become problematic.  It also is problematic in class D amps as the output filter necessary for those interacts with the loudspeaker load.

 

Now Demian Martin who used to post here and sometimes still does designed the Spectral amps early on.  He said they worked very hard to get an amplifier that let go of a signal or transient very quickly.  That it was an unappreciated difference you could hear. The near megahertz wide bandwidth according to him was a side effect of getting an amp that let go as quickly as it ramped up on a transient, and not a design goal.  Apologies if I am misrepresenting that, but that is it according to my memory.

[esldude]

For those of you worried about the glitch, and I do think it poor and unnecessary in a design, I wonder if it makes it to the speaker terminals and more so to the speaker output.  Or do the limited bandwidth of speakers filter it out so the low level glitch never actually makes a sound in room? Measurements and understanding are much better than name calling invectiveness.

[esldude]

4 hours ago, audiventory said:

Middle and right oscillogramm looks like positive and negative voltage management transistors (don't know English term) issue.

Maybe the term crossover distortion?  When the transistors cut off and on going from positive to negative near the 0 volt level. Usually that causes a flat spot at that point rather than a spike, but could go either way.

[Jud]

5 minutes ago, esldude said:

Maybe the term crossover distortion?  When the transistors cut off and on going from positive to negative near the 0 volt level. Usually that causes a flat spot at that point rather than a spike, but could go either way.

 

That was the term I misremembered as switching noise.

[crenca]

39 minutes ago, esldude said:

 

 

2 hours ago, semente said:

 

Maybe this: http://www.learnabout-electronics.org/Amplifiers/amplifiers14.php

 

39 minutes ago, esldude said:

Depends on what you are calling wide band.  Conventional transistor amps will have a first order, 6db per octave roll off.  That means you have to be -3db down at 200 khz to manage being only .1 db down at 20 khz.  So most good solid state amps should have response to 150khz or 200 khz.  You might wonder about filtering, and that could be done.  However once you get more than a first order rolloff you have the chance of resonance and more so using feedback could become problematic.  It also is problematic in class D amps as the output filter necessary for those interacts with the loudspeaker load.

 

Now Demian Martin who used to post here and sometimes still does designed the Spectral amps early on.  He said they worked very hard to get an amplifier that let go of a signal or transient very quickly.  That it was an unappreciated difference you could hear. The near megahertz wide bandwidth according to him was a side effect of getting an amp that let go as quickly as it ramped up on a transient, and not a design goal.  Apologies if I am misrepresenting that, but that is it according to my memory.

 

Thanks guys.  The link semente posted proposes the harmonic argument (i.e. ultrasonic harmonics effect audioble band because they affect the "shape" of the sound).  However, since learning about how the vast majority of recordings are done with microphones that do not capture anything above 20hz, and even the ones that do only seem to go up to 25hz or so - that begs the question of not only ultra wide band amps but digital formats above 48 or so sample rate.

 

However, what Demian Martin is talking about (is it really anything more than slew rate?? - perhaps measured at the particular transistor level??) is bandwidth as a side effect of another design/sound goal...

[esldude]

20 minutes ago, crenca said:

 

 

 

Thanks guys.  The link semente posted proposes the harmonic argument (i.e. ultrasonic harmonics effect audioble band because they affect the "shape" of the sound).  However, since learning about how the vast majority of recordings are done with microphones that do not capture anything above 20hz, and even the ones that do only seem to go up to 25hz or so - that begs the question of not only ultra wide band amps but digital formats above 48 or so sample rate.

 

However, what Demian Martin is talking about (is it really anything more than slew rate?? - perhaps measured at the particular transistor level??) is bandwidth as a side effect of another design/sound goal...

That was more or less what I asked Mr. Martin was he designing for high slew rate.  That was advertised by Spectral and is a result of wide bandwidth.  Mr. Martin talked about fast settling time and thermal tails of transients at the transistor level.  So a circuit that had a fast settling time was also wide bandwidth.  The high slew rates were not a goal however. Just a side effect of fast settling time.

 

Rick Fryer at times in interviews had talked about them going to wide bandwidth so they could use the cable, specifically MIT, to ensure a nice orderly roll off of a first order variety.  That without that an amplifier and cable could create audible resonances in response.  My impression was Mr. Fryer was doing market-speak once they were working with MIT.  Mr. Martin wasn't aware of that being part of it at all.  Perhaps having the MIT prevented unwanted oscillations that sometimes occurred with other gear. I believe I recall some Stax ESL speakers (not headphones) could simply oscillate until destruction of the amp and speaker at some improbably high frequency.

 

I had a Meridian 563 DAC when I owned a Spectral amp.  JA noted in testing it had a 1.1 mhz idle tone.  I connected it to an oscope and sure enough it did.  You could directly connect it to the Spectral input and see the same tone at the amp output.  I did this with a resistor load for the amp.  I think it was around 20 watts of 1.1 mhz output at 8 ohms.  So one does wish to be careful about such oddities with those amps.   Had I also been using a Spectral preamp, it would have passed that right onto the amp.  Fortunately I was not (only because I couldn't afford one). 

 

As for microphones, even some inexpensive condenser mikes respond pretty well to 30 khz or so.  They are not flat at that point, but do still respond only a few db down in level.  Dynamic mikes on the other hand likely don't make it to 20 khz to any real extent. 

[jabbr]

4 hours ago, semente said:

Since going down the custom designed electronics road I have learnt that such artifacts, inaudible as they may be, may affect the overall performance of the system downstream; you've probably heard that some wide-band amplifiers and even speakers can be affected by aggressive noise shaping (e.g. the designer of my amplifier has advised me not to use HQ Player's NS9).

 

I'm not sure I understand this. The DAC is supposed to have an LPF which should effectively eliminate the high frequency noise.

[Jud]

26 minutes ago, jabbr said:

 

I'm not sure I understand this. The DAC is supposed to have an LPF which should effectively eliminate the high frequency noise.

 

Good thought.

[Jud]

1 hour ago, esldude said:

As for microphones, even some inexpensive condenser mikes respond pretty well to 30 khz or so.  They are not flat at that point, but do still respond only a few db down in level.

 

Some good quality mikes do too.

 

http://www.earthworksaudio.com/wp-content/uploads/2016/03/M30-Data-Sheet-2016.pdf

[esldude]

39 minutes ago, jabbr said:

 

I'm not sure I understand this. The DAC is supposed to have an LPF which should effectively eliminate the high frequency noise.

Yes, but it was there nonetheless.  I don't know why JA at Stereophile was even looking at that frequency.  It wasn't something he normally would do. 

 

I have seen other sigma delta DACs that have idle tones though not normally at such high frequencies.  So I don't know how that happened with the 563.  It was an early sigma-delta design. 

[esldude]

7 minutes ago, Jud said:

 

Some good quality mikes do too.

 

http://www.earthworksaudio.com/wp-content/uploads/2016/03/M30-Data-Sheet-2016.pdf

 

Here is one of their recording mikes the QTC50.

 

 

I have one of their SR20 mikes. It is one nice sounding microphone.  If I were a pro I would have several more of their microphones. 

 

 

[jabbr]

6 minutes ago, esldude said:

Yes, but it was there nonetheless.  I don't know why JA at Stereophile was even looking at that frequency.  It wasn't something he normally would do. 

 

I have seen other sigma delta DACs that have idle tones though not normally at such high frequencies.  So I don't know how that happened with the 563.  It was an early sigma-delta design. 

Maybe it had this type of filter:

[gmgraves]

5 hours ago, manisandher said:

 

You may not hear anything untoward either, but other people certainly do:

 

 

 

 

And I think this is the real issue: Harley, Stoddard and others on this thread are mistaking the Yggy's distortion for "bold incisiveness" and "better clarity & focus".

 

Mani.

 

How can one hear that which is un-hearable? Because Yggy's distortion is below the level of audibility. I am pretty sure that what Yiggy detractors are hearing is caused by a lack of synergy between the Yggy and the rest of their system.

[clipper]

6 hours ago, manisandher said:

 

 

 

I took a few captures of the Yggy's analogue output and got others to share their thoughts blind. This is how they described the Yggy:

Yggy:

- Too tizzy (HF noise?)

- Sharper leading transients

- More sizzle (increase around 5kHz?)

- More (artificial?) detail?

- more sharply etched

- "crisper"

- not as "full"

- "detailed, dry, analytical" [type of systems]

- better clarity & focus

 

 

 

 

It sounds to me like these others are describing your recording of the Yggy rather than the Yggy itself.

 

I have a Yggy and have heard it in 4 other completely different systems.  To me it sounds nothing like the way you keep describing it.

 

Maybe there was a mismatch with some of your other equipment.

 

Maybe there was a mismatch with some of your cables.

 

Maybe you had a broken Yggy.

 

What was your source?  Was it computer based?  What playback software?  What were the settings?  

 

USB input?  S/PDIF?  AES?

 

 

 

[audiventory]

10 hours ago, Jud said:

 

Switching noise?

 

6 hours ago, esldude said:

Maybe the term crossover distortion?  When the transistors cut off and on going from positive to negative near the 0 volt level. Usually that causes a flat spot at that point rather than a spike, but could go either way.

 

Yes. I meant term "crossover distortion" (https://en.wikipedia.org/wiki/Crossover_distortion). There at picture shown 2 push-pull transistors.

I suppose, "switching noise" is more general term. Though shown in the thread kind of distortions looks like appear in the moment of switching between push and pull transistors.

At Wikipedia's picture shown other waveform of the distortions.

I still can't imagine exactly, how the switch may work way like picture in the thread.

 

 

[esldude]

24 minutes ago, audiventory said:

 

 

Yes. I meant term "crossover distortion" (https://en.wikipedia.org/wiki/Crossover_distortion). There at picture shown 2 push-pull transistors.

I suppose, "switching noise" is more general term. Though shown in the thread kind of distortions looks like appear in the moment of switching between push and pull transistors.

At Wikipedia's picture shown other waveform of the distortions.

I still can't imagine exactly, how the switch may work way like picture in the thread.

 

 

I may not have this right, but it is the picture I have.  Imagine your varying sine wave.  The top is approaching the 0 voltage point.  It has been encoded with 24 bits.  It is just barely positive at near 20 bits.  There should still be 4 bits between there and reaching zero to encode with.  However, it has been truncated and the next sample drops straight to zero. The following sample then jumps straight to the negative 20 bit level without the 4 bits to encode it either.  It would be as if a short term glitch of higher frequency was injected for those two samples. You'll get mostly odd harmonic distortion.  As you see in the 24 bit distortion plot you get higher odd distortion with lower even distortion spikes in the 24 bit result.  Which is why JA thought truncation instead of rounding.

 

It would have been nice to see the glitch in the waveform at both 100 hz and 10,000 hz in this particular case.

[manisandher]

1 hour ago, audiventory said:

I still can't imagine exactly, how the switch may work way like picture in the thread.

 

Yuri, the Yggy uses a pair of AD5791 chips per channel. Here is their 'raw' glitch energy:

 

 

Schiit have obviously done some work to reduce these glitch errors down to around 45µV in the Yggy, but they're still obviously there. JA referred to these as "significant errors at the signal's zero-crossing points".

 

And this is what the designer of the AD5791 thought about his chip being used for audio purposes:

 

"... 

I always assumed the code-change-dependent glitch energy would make THD unacceptable to audio buffs."

 

[Though in fairness, he does go on to say that he's glad that someone is at least trying his chip for audio purposes.]

 

Mani.

[PeterSt]

When I was designing with that same chip (being in the stage of a ready DAC board), I was looking and looking for a stupid reason that I could not attenuate to zero. Back at the time, at first I dedicated it to the LSB being stuck or something. And mind you, I was looking at an FFT of it ... Don't pin me down on it, but envision that I attenuated to 20 bits all right, but that beyond 13 bits or so nothing did not want to drop any more, including the harmonics of the signal. Thus, all looks normal, unless you start to digitally attenuate.

 

Can you see what I am getting at already ? if not, hold on.

 

The chip itself exhibits 3uV or so (peak-peak) of noise (I do all from the top of my head so a disclaimer should be in place). This means that the glitching, assumed this is 45uV peak (not peak-peak), is massively present in the middle of the normal LSB's. So we have an LSB coming up (which is just a square which holds for the intended period of time (like 22us) but right out of that peaks the glitch. I didn't do the math now, but if the 20th bit implies 5uV then the glitch extends to 45uV on top of that. And worse, because it goes the adverse way compared to the slope of the signal. Thus, 5uV plus for the signal (going upwards) and 45uV minus for the glitch (going downwards), which 45uV is relative to zero. Difference opposed to the signal : 50uV.

When the signal goes down (negative) the glitch goes the other way around (it goes upwards).

 

Notice that to see this better - and especially to understand this better, we must look at a lower level of the signal itself. And with the proper bandwidth you see how "dirac" this glitch is (say of infinite rise time).

 

Now on to the plot :

 

The characteristic of this glitching is, that it exhibits a modulation of the signal itself on to that same signal. If you look again at the plots exhibiting those glitches, you will see it. It happens at the exact frequency of the signal. Thus, when the signal itself is low enough, knowing that the glitching is of constant amplitude, the glitching energy starts to be prevalent. This is how I could not attenuate further under a certain (bit chop off) level.

 

The (dirac like) glitches themselves imply infinite frequency (investigate dirac pulses) while the signal itself is heavily polluted by itself and the amplitude is disturbed once the output level of the frequency is low enough (say under 12-13 bits as per my own example above).

 

Lastly, try to envision that this ONLY occurs when the signal crosses zero which is guided by the lowest frequency at first. I should draw a picture of it, but notice that all frequencies right on he lower ones. Thus, have 80Hz and the zero crossing happens twice per cycle. But, have a small amplitude of e.g. 200Hz on top of that and at a certain stage, when the 80Hz starts to approach zero, the 200Hz starts to cross zero first, And back and forth and back etc. Until the 80Hz is sufficiently under (or above) the zero line to the amplitude of the 200Hz can not reach the zero line any more, and then all is normal again. Until the next half cycle. And because there will not only be 200Hz but also many other frequencies, the zero crossing for a larger time will be polluted by the glitching, depending on the amplitude of the lowest frequency and of the amplitudes which ride on it.

 

All implies a huge mess and besides you can reason that a 20bit DAC becomes way less because of this although it better had be disappearing in the noise instead of being correlated to the music in some strange fashion (which most certainly is not random).

Can't you hear this or do you perceive this as "better" as such, well, ... I have my ideas about that. ;-)

 

Peter

[manisandher]

3 hours ago, clipper said:

It sounds to me like these others are describing your recording of the Yggy rather than the Yggy itself.

 

The recording of the Yggy was taken at 24/48 with a Tascam DA-3000, straight into its XLR input. To my ears, the Tascam isn't totally transparent, so you're right to raise this as a potential issue. And the playback chain was far from perfectly optimized too.

 

But it was the same playback and recording chain that I used for capturing my regular DAC, and people heard the difference between the two DACs quite easily.

 

Hell, here are the files again. One is the Yggy, the other my regular DAC (not necessarily in that order below):

 

DAC A: https://drive.google.com/open?id=0B0PU5LO5jVjfeGhoa3RCUTk4djQ

DAC B: https://drive.google.com/open?id=0B0PU5LO5jVjfQVpQaUxLNGZRUDA

 

[There is a 1dB difference in level between these two that really needs to be accounted for to make a fair comparison.]

 

The Yggy's sonic signature ("bold incisiveness" and "artificial detail") should be audible immediately - it is on my Ponoplayer using cheap Sony headphones. Some people may actually prefer it, but I'll just say that my regular DAC sounds much closer to the original file than does the Yggy.

 

Mani.

[manisandher]

11 minutes ago, PeterSt said:

Can't you hear this or do you perceive this as "better" as such, well, ... I have my ideas about that. ;-)

 

Well, this is what I'm genuinely interested in - can we correlated measured performance with sound?

 

Mani.

[Silly goose]

Well you guys have me convinced. I'm ordering one!! I wonder if there's a way to duplicate that glitch in DSP on an Iphone app? Audiophiles attracted to euphoric glitches rather than high fidelity should be very easy to please with basic DSP emulation algorithms. Surely the Iphone DAC has enough fidelity to transparently playback these glitches faithfully. 

[clipper]

10 minutes ago, manisandher said:

And the playback chain was far from perfectly optimized too.

 

But it was the same playback and recording chain that I used for capturing my regular DAC, and people heard the difference between the two DACs quite easily.

 

 

So what was the playback chain?

 

What software?  

 

XXHighEnd?  

 

Any upsampling?

 

 

 

[PeterSt]

18 minutes ago, manisandher said:

 

Yuri, the Yggy uses a pair of AD5791 chips per channel. Here is their 'raw' glitch energy:

 

 

Schiit have obviously done some work to reduce these glitch errors down to around 45µV in the Yggy, but they're still obviously there.

 

Ha ! ... what a timing. But :

 

3 minutes ago, PeterSt said:

Notice that to see this better - and especially to understand this better, we must look at a lower level of the signal itself. And with the proper bandwidth you see how "dirac" this glitch is (say of infinite rise time)

 

So that kind of picture I was referring to, with the notice that Mani took that picture from my own (Phasure) forum since January 20, 2014. This subscript goes with it :

Notice : This is NOT from the NOS1 nor is it from the PCM1704 D/A chip; the chip shown here carries a real high glitching energy.

The picture is not the one I intended because the one Mani showed is showing an idle signal. The characteristic of this is that now no glitching peaks go downwards. It also tells me again that the glitching is not 45uV only but way more, but as I already told : shown with sufficient bandwidth or "zooming" if you want; this is because the dirac function in order is not 100% dirac but is subject too electrical rise time and therefore as a base which is wider than one bit or better : it is analogue and analogue is not infinitely fast.

Lastly the picture reminds again of the "LSB being stuck" as it came to be at first, which really is so (I now recall). This is how it took me 2 or three months more to find out it was just glitching energy which I was not familiar with (when looking at analyzer plots). Of course, once you know, all is easy to explain. But the behavior all together is a kind of strange and not consistent (like the one way of the glitching you see happening here - I could even have troubles today to explain how that can happen and for example where the energy is taken from (be creative and you can see a few things)).

 

 

[manisandher]

9 minutes ago, clipper said:

So what was the playback chain?

 

What software?  

 

XXHighEnd?  

 

Any upsampling?

 

 

 

 

Home-built audio PC (can't remember specs because its changed since). XXHighEnd software player. Respective USB inputs. No attenuation in either case. No upsampling or any other DSP for Yggy - so using Yggy's internal filter. 16x upsampling for regular DAC (NOS1a is a non-oversampling and filterless design, requiring upsampling in playback software).

 

Mani.

[manisandher]

21 minutes ago, PeterSt said:

The picture is not the one I intended because the one Mani showed is showing an idle signal.

 

Here you go:

 

 

Mani.