Review and Measurements of Schiit Yggdrasil V2 DAC

[amirm]

Alcophone had asked that I also test 192 kHz. Here are a few representative measurements:

Linearity: No change

IMD: no change

THD: no change

And finally, our dashboard. This one is different because I expanded the analyzer bandwidth beyond audible band to 90 kHz:

There is a drop in SINAD and increase in THD+N. That is expected because we are now integrating 4X more noise into those figures. Otherwise, nothing untoward has occurred at 192 kHz.

Note that both sampling rate and a day of warm up have gone on and no impact on unit's performance. Those people who think "everything matters" should think twice.....

 

[amirm]

Something somewhere somehow is causing a difference between these measurements and the other testers.

Why don't you put your thinking cap on and hazard a guess as to what may be going on.

 

[amirm]

Out of interest - is there much/any difference in the Topping when using the ASIO drivers vs the ASIO4ALL?

No, and there isn't with any other device I have tested. Just to be sure, I just did a comparison:

The ASIO graph is in green on top of the blue one from yesterday. As you see, they are exactly the same.

And oh, the Topping was cold and powered off and it still made no difference from yesterday to today.

 

[gvl]

I do wonder if these static measurements tell us the the whole story. Music is somewhat more complex than a sine wave. Is there anything else that can be captured and compared? Impulse response? After trying various things over the course of couple of years I still can't decide between OS and NOS, and I do know that even the best NOS measures worse than the Yggdrasil.

 

[Blumlein 88]

I do wonder if these static measurements tell us the the whole story. Music is somewhat more complex than a sine wave. Is there anything else that can be captured and compared? Impulse response? After trying various things over the course of couple of years I still can't decide between OS and NOS, and I do know that even the best NOS measures worse than the Yggdrasil.

Lots of people think so.

This powerpoint on Spectal analysis of musical sounds is worth reading. It's done as an easy to understand lecture.
http://astro.pas.rochester.edu/~aquillen/phy103/Lectures/D_Fourier.pdf

Yet remember Fourier. Any signal can be decomposed into an equivalent group of sine waves.

And look at something like this twin tone IMD signal. It is going from maximum to minimum 1000 times per second. It is something of a torture test signal beyond what you'll see with music.

And for comparison here are some recordings of large cymbals being struck loudly and recorded with a wide bandwidth microphone. I think it was an Earthworks omni. Also notice the second one has been downsampled to 44 khz and upsampled back to 176 khz. The others are the direct very loud recording of struck cymbals. That should be transient enough for you shouldn't it.

Even something like a cymbal isn't the instantaneous smashing signal people imagine. What you see is a period of time from being struck until the energy can travel across the brass cymbal, reflect and eventually build into a resonance, then slowly decay. So even these cymbal crashes build, reach a peak and decay well within our ability to record the result. And within the ability of good gear to follow the signal with room to spare. None are as stressful as the full max twin tone IMD signal.

 

[Blumlein 88]

I do wonder if these static measurements tell us the the whole story. Music is somewhat more complex than a sine wave. Is there anything else that can be captured and compared? Impulse response? After trying various things over the course of couple of years I still can't decide between OS and NOS, and I do know that even the best NOS measures worse than the Yggdrasil.

I also think of the post of Ray's. He captured the step response of his speakers. Compared it to the calculated impulse response done by REW using a slow sine sweep. They were identical which is what the math says should happen. This sort of thing really works.

 

[RayDunzl]

I also think of the post of Ray's.

Posts 1 and 9 - https://audiosciencereview.com/forum/index.php?threads/impulse-response.1765/

 

[Alcophone]

Hmmm. I see that the frequency counter is now showing the correct value at 1.00000 kHz. I wondered if the USB input is running too fast so reconnected that and measured one more time:

Yup, it is back to 1.00008 kHz. So the USB clock is running fast. If you like your life in the slow lane, use AES (or S/PDIF?) inputs.

Interesting! I'd be curious if the same happens with any other DAC connected to a Schiit Eitr - since that is also USB gen 5.

 

[amirm]

I do wonder if these static measurements tell us the the whole story. Music is somewhat more complex than a sine wave.

Well, a sine wave can very well be music. So distorting that more in one device than another ought to be a bad thing, not good.

Is there anything else that can be captured and compared?

Of course! This is audio science review. We can do anything.

We can try a square wave. For those of you who don't know what a square wave is, read Don's excellent article: https://audiosciencereview.com/forum/index.php?threads/composition-of-a-square-wave-important.1921/

In a nutshell, you need infinite series of odd harmonics of the square wave frequency in the form of sine waves to make that signal. If we feed an ideal, 100% linear system a 1 kHz square wave, we would expect to see 1 kHz, 3 kHz, 5 kHz, 7 kHz, etc.

Let's see how our two DACs do on such a signal (-30 dBFS, 1 kHz, Square wave):

We have a problem here. While Topping DX7s produces exactly what we expect in the form of odd harmonics in blue, our Schiit Yggdrasil is doing the same but then throwing a bunch of even harmonics in between!

No way, no how can we say the Yggdrasil sounds better given addition a spray of unwanted distortions not in the original recording. These distortions cannot even be euphonic. It is just bad stuff.

Fortunately our hearing is poor in these regards so likely in controlled testing, we can't tell these two apart. But the distortions are there whether we test a smoothly varying sine wave or a pulsing square wave. And signal processing theory is proven over and over again (that you can decompose a signal like square wave into component sine waves).

So there is no escaping of the truth here.

 

[gvl]

Digital filters ring differently, that ought to have some effect on transients, no?

Lots of people think so.

This powerpoint on Spectal analysis of musical sounds is worth reading. It's done as an easy to understand lecture.
http://astro.pas.rochester.edu/~aquillen/phy103/Lectures/D_Fourier.pdf

Yet remember Fourier. Any signal can be decomposed into an equivalent group of sine waves.

And look at something like this twin tone IMD signal. It is going from maximum to minimum 1000 times per second. It is something of a torture test signal beyond what you'll see with music.
View attachment 13528

And for comparison here are some recordings of large cymbals being struck loudly and recorded with a wide bandwidth microphone. I think it was an Earthworks omni. Also notice the second one has been downsampled to 44 khz and upsampled back to 176 khz. The others are the direct very loud recording of struck cymbals. That should be transient enough for you shouldn't it.

View attachment 13529

Even something like a cymbal isn't the instantaneous smashing signal people imagine. What you see is a period of time from being struck until the energy can travel across the brass cymbal, reflect and eventually build into a resonance, then slowly decay. So even these cymbal crashes build, reach a peak and decay well within our ability to record the result. And within the ability of good gear to follow the signal with room to spare. None are as stressful as the full max twin tone IMD signal.

Let's not forget there is a digital filter in the path and they do tend to have some variations in how they work. How about we level match 2 DACs, feed it this same digital signal to both, overlay the resulting signals in time domain and see if there are any meaningful (or not) differences? This won't necessarily tell us which one is better but would be interesting nevertheless.

 

[Alcophone]

No way, no how can we say the Yggdrasil sounds better given addition a spray of unwanted distortions not in the original recording. These distortions cannot even be euphonic. It is just bad stuff.

Fortunately our hearing is poor in these regards so likely in controlled testing, we can't tell these two apart.

So... it can't sound better, even though we likely don't hear that bad stuff? That doesn't make sense to me

 

[Blumlein 88]

Interesting! I'd be curious if the same happens with any other DAC connected to a Schiit Eitr - since that is also USB gen 5.

It is really not specific to this DAC. When running asynch USB, you are using the free running clock in the DAC. When using SPDIF/AES3 you have to use a PLL or PLLs to synch the free running clock with the embedded clock of the digital source. Clocks vary among themselves. So you should see SPDIF clocks match the source. And all USB asynch connections will vary from the source a small amount.

Among those I have on hand, picking my Antelope Audio as the reference, one is 59 ppm fast, one is 3 ppm slow, and one is 15 ppm fast. That isn't unusual. 80 ppm is on the high side, but by itself of no consequence.

 

[SIY]

I do wonder if these static measurements tell us the the whole story. Music is somewhat more complex than a sine wave. Is there anything else that can be captured and compared? Impulse response? After trying various things over the course of couple of years I still can't decide between OS and NOS, and I do know that even the best NOS measures worse than the Yggdrasil.

Since the sine wave sweeps all values, it can hardly be called "static." To answer the objections of the Fourier-deniers, I run multitone measurements as well (typically 42 tones, which looks pretty nonstatic!)- to date, I have not seen that very complex signal show up any problems that aren't visible in single sine measurements.

I'd say that Amir's measurements are quite adequate to show up poor behavior.

 

[Blumlein 88]

Digital filters ring differently, that ought to have some effect on transients, no?


Let's not forget there is a digital filter in the path and they do tend to have some variations in how they work. How about we level match 2 DACs, feed it this same digital signal to both, overlay the resulting signals in time domain and see if there are any meaningful (or not) differences? This won't necessarily tell us which one is better but would be interesting nevertheless.

Eh! please not with the ringing filters garbage. What is typically called ringing, only occurs in the transition band. With CD that would be between 20,000 Hz and 22,050 Hz. With the 176 khz sampled cymbals the transition band is 80,000 Hz and 88,200 Hz. Look at the second line in my cymbal post. That is one where I took the 176 and resampled to 44 khz. There isn't that much difference.

As for the latter part where you wish to overlay signals, I've done that. You aren't going to rescue your argument from the island it has stranded you upon that way.

See the other thing about those ringing graphs you see from impulses, they are something of an illegal signal. They are for testing filters, but not really pertinent to the full ADC to DAC reconstruction. An actual impulse fed to an ADC has ultrasonics filtered out first, and then converted to digital. The reverse on upon playback. You don't see filter ringing from impulses in music when you do that the way you are imagining you would. Not even at 44 kHz sampling rates.

 

[gvl]

And all USB asynch connections will vary from the source a small amount.

Isn't a DAC-side USB async connection technically the "source"?

 

[Blumlein 88]

Since the sine wave sweeps all values, it can hardly be called "static." To answer the objections of the Fourier-deniers, I run multitone measurements as well (typically 42 tones, which looks pretty nonstatic!)- to date, I have not seen that very complex signal show up any problems that aren't visible in single sine measurements.

I'd say that Amir's measurements are quite adequate to show up poor behavior.

View attachment 13531
View attachment 13532

I want to plus one SIY here. I've done similar things. Wondering if multi-tone tests are better than tone pairs or sine sweeps. They aren't. I've yet to find something that was showing me anything using more tones than a pair. I've not done 42, but I've done 5 and 12 tone measurements. I've sent max level noise, but with 2-4 kHz empty to see if anything shows up in the silent gap where our hearing is most sensitive. Nothing.

 

[Blumlein 88]

Isn't a DAC-side USB async connection technically the "source"?

In a testing case, the AP sends a digital signal to the DAC. It then has the AP ADC listening to the analog output. This ADC presumably runs off the same clock as the AP digital signal generator. If you are doing this over USB, then the DAC clock between AP digital sourcing and AP ADC can be of slightly different speed.

 

[gvl]

In a testing case, the AP sends a digital signal to the DAC. It then has the AP ADC listening to the analog output. This ADC presumably runs off the same clock as the AP digital signal generator. If you are doing this over USB, then the DAC clock between AP digital sourcing and AP ADC can be of slightly different speed.

I was just trying to say the DAC is the clock, the AP signal generator just produces samples and sends them out according to the DAC.

 

[Blumlein 88]

I was just trying to say the DAC is the clock, the AP signal generator just produces samples and sends them out according to the DAC.

That would be correct. The DAC is the source of the timing of the clock.

 

[Candlesticks]

@amirm can you upload a project file?

 

[Thomas savage]

Quick thank you to the members contributing in this thread, more precisely the manor in which everyone has conducted themselves is really pleasing.

Cheers