Jun 7, 2018 at 6:44 AM Post #8,356 of 13,822
.....I have shown this numerous times on ASR Forum. There is no question that Schiit DACs don't measure well no matter who measures them. Buy them for other reasons than engineering excellence.....

....So while there can be doubt in everything, I suggest by default you should accept the data as presented unless it is shown otherwise. Doing it in reverse means closing one's eyes to information that is purely created to make consumers more informed.

@amirm

Hello amirm. I'm a rather satisfied owner of the Yggy (V2, and originally had V1). I have read every post in your "Measurements and Review of Schiit Yggdrasil DAC" thread on the audioscience review site. Suffice it say I'm inclined to be interested as to why the Yggy measures as poorly as you have found and document. I lack the technical knowledge to offer meaningful rebuttal, so I'll take it you are correct (even about the comparison and interpretation to atomic bobs measurements).

I've posted herein this thread on headfi a few times, citing your findings in order to seek additional clarity, other perspectives, etc., and some have offered a few items into evidence as to why your data should be ignored. Nevertheless, I haven't read anything compelling that in my mind would cause me to completely ignore your words.

So given the aforementioned, can you please speculate as to why the Yggydrasil produces sound quality that essentially belies its measurements? Similarly, can you please speculate as to why the Yggydrasil produces sound quality that is essentially "better" or "preferred" to many other DACs (especially sabre delta sigma chips) in its price class?

In this context, assume that there is more to it than simply subjective listening preferences, but more objective impressions from qualified listeners (musicians and those with formal music education background) who can discern the Yggy in a blind test. And of course, this is PCM only and is irrespective of bit depth and sample rate.

Thank you.
 
Last edited:
Jun 7, 2018 at 9:40 AM Post #8,357 of 13,822
I suppose this would be a good time to talk about my experience with Yggy v2. I bought Yggy and Jot, then a PS Audio Stellar Gain Cell DAC, which uses a D/S Saber dac.

First I connected Yggy single ended to a Outlaw 2160 and LS-50 speaker system, using a iFi reclocker and Analysis Plus USB cable on the Outlaw. Yggy using just a cheap AQ usb cable.
The dac in the Outlaw is of unknown origin, but is a cheap, harsh sounding pos.
Yggy was great strait out of the box here, now I know I don't hate my LS-50s. I hate that cheap dac.

Then I compared the Schiit to the PS Audio SGCD using LCD Xs with the stock cable. Again same pc with the iFi gizmo and $$$ usb cable on the D/S dac. Yggy doesn't like the iFi reclocker, so I just stuck with a basic AQ usb cable.

Everything was burned in and I spent quite a while listening. Yggy was VERY much better, I had no allegiance to Schiit at this point, I just wanted the best sound. And the preamp in the SGCD was very good, but back went the SGCD.

I also have the OG 4490 dac in the Jot for comparison. IMO Schiit Mutibit kills the D/S dac's I have heard.:gs1000smile:
 
Jun 7, 2018 at 2:51 PM Post #8,358 of 13,822
I have shown this numerous times on ASR Forum. There is no question that Schiit DACs don't measure well no matter who measures them. Buy them for other reasons than engineering excellence.

Every time I see a post like this which starts up the same old back and forth argument yet again, I always get the distinct impression that guys like you don't actually enjoy this hobby. At least not in the way I do (or try to). This isn't meant to be an insult or attack. Rather it's just me musing on the nature of people, and how the things that motivate us can be so different it leads to perspective differences too great to reconcile.
 
Jun 7, 2018 at 4:08 PM Post #8,359 of 13,822
I attended the talk. It was mostly about knowing which FFT parameters to use and making sure you are using bit-exact path to test products. My team wrote the Windows audio stack while I was at Microsoft, and one of my professional specialities is signal processing (I managed that team at Microsoft too) so, neither was news to me. His testing and mine both use bit-exact and I have taken his measurements and replicated them on my Audio Precision analyzer showing the same problems. In almost all cases Bob's data actually matches mine. It is just that his words are far more positive and people go by that, instead of data that represents otherwise.

I have shown this numerous times on ASR Forum. There is no question that Schiit DACs don't measure well no matter who measures them. Buy them for other reasons than engineering excellence.


Yeh but give me credit for having more Audio Precision gear than anyone! :D There are no less than three of them in this picture (with one just peeking from the corner):

index.php


Seriously, you are right that it is easy to make mistakes and knowledge level it takes to test mixed-signal products like DACs can be pretty large. I like to think that as an owner of such measurement gear for 20+ years, and professional experience related to everything here from analog to digital and signal processing, I know what I am doing. But if I am not, I am open to others showing otherwise. So far, that has not happened.

So while there can be doubt in everything, I suggest by default you should accept the data as presented unless it is shown otherwise. Doing it in reverse means closing one's eyes to information that is purely created to make consumers more informed.

@amirm, it seems to me you have a personal agenda with regard to Schiit Audio. Some points:
  • Schiit Audio has been very forthcoming about the fact that their multi-bit DACs will in general not measure as well as delta-sigma DACs (theirs or others). They've been very upfront about this, and have repeatedly stated that they simply feel their multi-bit DACs sound better, and a lot of their customers agree. I have purchased most of their multibit models, understanding they won't measure as well as their delta-sigma counterparts, just as I also know a simple CMOY portable amp probably will measure better than all of my favorite tube amps.

  • The sensationalism of your posts appears to be clearly intended to have your readers believe that you somehow found them out in an "ah-ha!-look-what-i-found-guys!" manner. Yet they've been quite frank about this discussion since the first version of the Yggdrasil was released (and with every multi-bit DAC they've released since).

  • You seem dead set on interpreting the results (especially where Schiit's concerned) in a manner that is consistent with what I (again) feel is clearly a personal agenda or bias with regard to Schiit Audio. And your interpretation of the linearity error plot -- using your +/-0.1 dB threshold -- is one I have not specifically seen anywhere. I'm not saying nobody else is using it, only that I've not seen it used elsewhere. I am genuinely wondering where this standard originated -- I'm assuming that the +/-0.1 dB threshold on that type of measurement references a specific standard, so I'm wondering where I can read more about it.

  • When it comes to similar tests conducted elsewhere, it seems the narrative can be quite different. Let's look at John Atkinson's measurement of another multi-bit DAC (the HoloAudio Spring DAC "Kitsuné Tuned Edition" Level 3), starting with this plot in it (below):

    518HoloSpringfig08.jpg
    • John said (about the above plot):
      A relevant issue with resistor-ladder DACs is the linearity error: Will a digital signal at, for example, –80dBFS be reproduced at the outputs by an analog signal the same 80dB down from full level? However, the Spring performed well in this respect (fig.8). When I examined linearity, the error was negligible down to –60dBFS, and remained below 1dB down to 90dBFS.
    • I find this difference in narrative (versus yours) very interesting.

  • Here's an additional plot from another DAC measurement by John Atkinson (below):
    618Aquafig05.jpg
    • John said (about the above plot):
      A relevant issue with resistor-ladder DACs is the linearity error: Will a digital signal at, say, –80dBFS be reproduced at the outputs by an analog signal the same 80dB down from full level? However, the Aqua performed well in this respect. When I examined its linearity (fig.5), the error was negligible down to –80dBFS, and remained below 1dB down to –104dBFS.
    • Again, I find the difference in narrative for this type of measurement very interesting.
Also, with respect to your measurements, you only show the Yggdrasil2 measured from its unbalanced outputs, and I'm sure you noticed that the Yggdrasil does perform considerably better from its balanced outputs, since you did post a frequency response plot from its balanced outputs (but nothing else). Since the Yggdrasil2 is described by Schiit Audio as a "Balanced Upgradable DAC," I thought you might at least consider also including other measurements from its balanced outputs.

Later on, I may post more measurements, but let's take a look at a few of my measurements of the Yggdrasil2 (using the Audio Precision APx555) compared to some of the measurements you made that were key to your conclusion(s):

(NOTE: All of my measurements below were made using the Yggdrasil2's balanced digital input (XLR).)


Frequency Response


Your plot (below) and your comments (quoted below):

Schiit Yggdrasil DAC Frequency Response Measurement.png

The top line is the Gen 2 analog board in balanced mode. All is well there. But if you look at the two curves at the bottom, both of which are for unbalanced output, we see problems. The Gen 1 board has a small roll off < 20 Hz which we could ignore. But Gen 2 board starts to drift down at some 300 Hz and by the time you get to 10 Hz, it is down by half a dB.

Controlled listening tests by Toole/Olive show that low frequency deviations that are this broad have a threshold of hearing of 0.5 dB. So this is right at threshold of hearing...

Here are my measurements of the unbalanced outputs of the Yggdrasil2, from 10 Hz to 20 kHz, first as a continuous sweep (below):

RMS Level_SWEEP_Ygg2_digi-bal_ana-unbal_24-192_10Hz-20kHz_0dBFS_1s.jpg

The deviation on this plot is +/- 0.139 dB for the left and +/- 0.136 dB for the right from 10 Hz to 20 kHz. Within the audioband (20 Hz to 20 kHz), that deviation is +/- 0.058 db and +/-0.056 dB. So, whether from 10 Hz or 20 Hz (through to 20 kHz), the deviation is well below the threshold of hearing you cited.

Here's the frequency response (below) as a stepped frequency sweep measurement (100 steps):

RMS Level_STEP_Ygg2_digi-bal_ana-unbal_24-192_10Hz-20kHz_0dBFS_100 stp.jpg


The deviation on this plot is +/- 0.155 dB for the left and +/- 0.151 dB for the right from 10 Hz to 20 kHz. Within the audioband (20 Hz to 20 kHz), that deviation is +/- 0.058 db and +/-0.056 dB. Whether from 10 Hz or 20 Hz (through to 20 kHz), again, the deviation is well below the threshold of hearing you cited.

Additionally, your Yggdrasil2 frequency response measurement (from the unbalanced outputs) shows an unusual concave sag <300 Hz. As you can see, I did not get a similar feature from the same measurement (swept or stepped).

Since you now also own an Audio Precision APx555, perhaps you should consider re-doing the measurement.

Again, from its balanced outputs, the Yggdrasil 2 performs better.

Here's the frequency response from the balanced outputs, swept (below):

RMS Level_SWEEP_Ygg2_digi-bal_ana-bal_24-192_10Hz-20kHz_0dBFS_1s.jpg


The deviation on this plot is +/- 0.059 dB for the left and +/- 0.057 dB for the right from 10 Hz to 20 kHz. Within the audioband (20 Hz to 20 kHz), that deviation is +/- 0.058 db and +/-0.055 dB. Whether from 10 Hz or 20 Hz (through to 20 kHz), again, the deviation is well below the threshold of hearing you cited.

Here's the frequency response (below) from the balanced outputs, stepped (100 steps):

RMS Level_STEP_Ygg2_digi-bal_ana-bal_24-192_10Hz-20kHz_0dBFS_100 stp.jpg


The deviation on this plot is +/- 0.058 dB for the left and +/- 0.056 dB for the right from 10 Hz to 20 kHz. Within the audioband (20 Hz to 20 kHz), that deviation is still +/- 0.058 db and +/-0.056 dB. Whether from 10 Hz or 20 Hz (through to 20 kHz), again, the deviation is well below the threshold of hearing you cited.


THD+N Versus Frequency


Let's look at your THD+N versus frequency measurement (below), and your comment (quoted below):

Schiit Yggdrasil DAC THD+N Measurement.png

Whoa! What went on here??? Your guess is as good as mine. The new board seems to be much worse the lower the frequencies get. And regardless, both are shown the door as compared to the Topping DX7 and Exasound E32.

I did not get the dramatic rise (going lower in frequency) that you did. Also, the fact that your THD+N plots are all dropping off as they approach higher frequencies could be accentuating the low-frequency rise you're describing (that, again, I'm not seeing to the extent you are), and this suggests to me that your measurement bandwidth may be ~20 kHz. If so, perhaps you should reduce the upper limit of your X-axis and/or increase your analyzer's bandwidth setting to at least capture the lower harmonics of the upper frequency range of the measurement. My bandwidth for this was set to 90k (192 kHz SR).

Here are my measurements of THD+N versus frequency (below) from the unbalanced outputs of the Yggdrasil2, from 10 Hz to 20 kHz (100 steps):

THDN_100 step_0 dBFS_10Hz-20kHz_digI bal out_ana unbal in.jpg


Here's that same measurement overlaid with the dashed lines showing the THD+N when the measurement bandwidth is limited to 22.4 kHz (below):

THDN_100 step_0 dBFS_10Hz-20kHz_digI bal out_ana unbal in_also-with-bandwidth-22.4k.jpg


Here are my measurements of THD+N versus frequency (below) from the balanced outputs of the Yggdrasil2, from 10 Hz to 20 kHz (100 steps):

THDN_100 step_0 dBFS_10Hz-20kHz_digI bal out_ana bal in.jpg



Linearity

Getting back to the linearity measurements, I want to ask again where the +/- 0.1 dB threshold on the linearity error test was standardized. While others may be using it, you're the only one I've seen using it. I saw that you recently cited this article by Robert Harley in Stereophile in discussing the topic. I may have missed where Harley defined a +/- 0.1 dB threshold (I didn't read every page of that article). I did, however, search Stereophile for Harley's digital component reviews, and found this example (below):

Num94fig04.jpg

Of the above plot, Harley said:
The Karik's linearity (fig.4) was excellent, maintaining good performance to below -100dB, where the noise floor intrudes on the measurement.

In a review by Kalman Rubinson in Stereophile, I found this example (below):

NAGdaFIG03.jpg

Of the above plot, Rubinson said:
The linearity error on a dithered 16-bit 500Hz tone (fig.3) is superbly low down to almost -120dBFS, the increasing error below that level entirely due to the dither noise in the data.

In both cases (and I'm quite sure there are more), their descriptions of when the devices under test remain linear seem to far exceed your 0.1 dB tolerance. Looking at those plots, Harley seems to be describing ~1 dB deviation and Rubinson ~2 dB. I decided, then, to go with 1 dB deviation in linearity tests expressed in both a linearity error measurement and an RMS measurement (where 0 dBFS output at 1 kHz is 0 dBrA in each RMS example, to make the corresponding change in RMS (versus dBFS) easier to read).

Here is my linearity error measurement of the Yggdrasil2 (below) from its unbalanced analog outputs (281 steps, 0.500 dBFS per step, from -140 dBFS to 0 dBFS):

LIN_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana unbal in_CURS.jpg


The lowest step on the X-axis at which the plots in the above linearity error measurement remain within 1.00 dB of 0.0 is -95.000 dBFS (left) and -99.000 dBFS (right).

Here is my RMS measurement (versus dBFS) of the Yggdrasil2 (below) from its unbalanced analog outputs (281 steps, 0.500 dBFS per step, from -140 dBFS to 0 dBFS):

RMS_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana unbal in_CURS.jpg


The lowest step on the X-axis at which the plots in the above RMS linearity measurement have dBFS and dBrA levels within 1.00 dB of each other is -95.000 dBFS (left) and -98.500 dBFS (right). (Again, they're in 281 total 0.500 dBFS steps, which is why they end in either .000 or .500.)

As I've now said a few times, the Yggdrasil2 definitely performs better from its balanced outputs (which you did not post, but did seem to run measurements from).

Here is my linearity error measurement of the Yggdrasil2 (below) from its balanced analog outputs (281 steps, 0.500 dBFS per step, from -140 dBFS to 0 dBFS):

LIN_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana bal in_CURS.jpg


The lowest step on the X-axis at which the plots in the above linearity error measurement remain within 1.00 dB of 0.0 is -122.000 dBFS (left) and -117.500 dBFS (right).

Here is my RMS measurement (versus dBFS) of the Yggdrasil2 (below) from its balanced analog outputs (281 steps, 0.500 dBFS per step, from -140 dBFS to 0 dBFS):

RMS_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana bal in_CURS.jpg


The lowest step on the X-axis at which the plots in the above RMS linearity measurement have dBFS and dBrA levels within 1.00 dB of each other is -122.000 dBFS (left) and -117.500 dBFS (right).

Again, the performance from the balanced analog outputs is significantly better than from the unbalanced analog outputs (and, on these forums, as best I can recall, Schiit has always recommended using the Yggdrasil balanced). I'm not sure why you only included a balanced measurement for frequency response, but nothing else (e.g. linearity, etc.).

As for determining a DAC's resolution, when I find the time, I want to examine how John Atkinson from Stereophile determine's a DAC's resolution in bits, which is certainly different than your +/- 0.1 dB linearity error standard. When I do that, I'll show the corresponding measurements for this DAC (and perhaps others).

NOTE: Stereophile's audio measurements have been published and available for decades, their methods and results subject to peer review (they're public), and their methods occasionally helped along by other engineers in the industry. As such, their work has largely guided and informed our measurements of audio electronics, and will likely continue to. That said, I will endeavor to add interesting and novel measurements (well, novel outside of R&D labs), like examining out-of-band performance and behavior, the ability of DACs to tolerate jitter (jitter that we control to deliberately impair the signal), and more.

@amirm, you are a member of the trade (MOT), and we do not allow MOTs to criticize and/or attack other MOTs here. Again, it's clear to me that you have an agenda and strong bias, especially where Schiit Audio is concerned. The only reason I allowed your post (the one I'm quoting in this post) to remain is so that I could respond to it and maintain the context for my response.

Of course, you are welcome to do whatever you want on your website and forum, but (especially as you are a MOT) Head-Fi is not your dais to carry out your particular brand of bias. You seem to suggest that you let the measurements do all the talking, but I beg to differ. Sometimes your measurements fall silent in the shadow of your sensationalist narrative.

Again, the Schiit Audio Yggdrasil2 is a multi-bit DAC, and most of those who buy it are probably not buying it for the best measured performance, even though it does seem to me to measure well relative to other multi-bit DACs -- especially from its recommended balanced outputs. While plenty of other DACs are promoted by their makers as sporting 32-bit/768kHz DAC chips, Schiit Audio has always very openly described each of the Yggdrasil2's four AD5791 BRUZ DAC chips as 20-bit DACs. Your narrative suggests that you were perhaps the first to uncover the fact that multi-bit DACs generally do not measure as well as the delta-sigma types, as if it were a notion new to us all. It is not. Perhaps your next discovery is that tube amps don't measure as well as solid state ones? That turntables don't measure as well as digital sources?

If you want to have a private conversation via PM (or even telephone), then feel free to PM me.

All of my measurements in this post were made at Head-Fi HQ using the Audio Precision APx555 audio analyzer.
 

Attachments

  • LIN_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana bal in.jpg
    LIN_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana bal in.jpg
    163 KB · Views: 0
  • LIN_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana unbal in.jpg
    LIN_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana unbal in.jpg
    169.6 KB · Views: 0
  • RMS_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana bal in.jpg
    RMS_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana bal in.jpg
    170.8 KB · Views: 0
  • RMS_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana unbal in.jpg
    RMS_281 step (0.500 per)_-140-0 dBFS_1 kHz_digi bal out_ana unbal in.jpg
    170.7 KB · Views: 0
  • upload_2018-6-7_13-32-40.png
    upload_2018-6-7_13-32-40.png
    143.2 KB · Views: 0
  • upload_2018-6-7_13-32-46.png
    upload_2018-6-7_13-32-46.png
    143.2 KB · Views: 0
Last edited:
Jun 7, 2018 at 4:21 PM Post #8,360 of 13,822
One of my favorite posters here on HF has this in his signature and I can not agree more

"Inaudible is inaudible and you don't get extra brownie points for making it twice as inaudible as another solution".

If you can not hear the difference in measurements then they do not matter...

:)
 
Jun 7, 2018 at 4:58 PM Post #8,362 of 13,822
...Buy them for other reasons than engineering excellence.

Do you mean like "they sound wonderful and are an amazing value"? Just checking...
 
Last edited:
Jun 7, 2018 at 5:00 PM Post #8,363 of 13,822
@jude

Interesting post of yours as rebuttal to amirm. Appreciate your allowing his post and various responses to continue within this thread. Please continue in that vain.

I'm inclined to agree with you (your posts), your measured data, and interpretation of same.

That said, do you have something to offer in response to amirm claim that there is (paraphrasing) power supply (measurable) leakage in and around 60Hz?

Thank you.
 
Jun 7, 2018 at 5:02 PM Post #8,364 of 13,822
One of my favorite posters here on HF has this in his signature and I can not agree more

"Inaudible is inaudible and you don't get extra brownie points for making it twice as inaudible as another solution".

If you can not hear the difference in measurements then they do not matter...

:)

I always liked:

"If it sounds good and measures bad, then you're measuring the wrong thing."
 
Jun 7, 2018 at 5:16 PM Post #8,365 of 13,822
I attended the talk. It was mostly about knowing which FFT parameters to use and making sure you are using bit-exact path to test products. My team wrote the Windows audio stack while I was at Microsoft, and one of my professional specialities is signal processing (I managed that team at Microsoft too) so, neither was news to me. His testing and mine both use bit-exact and I have taken his measurements and replicated them on my Audio Precision analyzer showing the same problems. In almost all cases Bob's data actually matches mine. It is just that his words are far more positive and people go by that, instead of data that represents otherwise.

I have shown this numerous times on ASR Forum. There is no question that Schiit DACs don't measure well no matter who measures them. Buy them for other reasons than engineering excellence.


Yeh but give me credit for having more Audio Precision gear than anyone! :D There are no less than three of them in this picture (with one just peeking from the corner):

index.php


Seriously, you are right that it is easy to make mistakes and knowledge level it takes to test mixed-signal products like DACs can be pretty large. I like to think that as an owner of such measurement gear for 20+ years, and professional experience related to everything here from analog to digital and signal processing, I know what I am doing. But if I am not, I am open to others showing otherwise. So far, that has not happened.

So while there can be doubt in everything, I suggest by default you should accept the data as presented unless it is shown otherwise. Doing it in reverse means closing one's eyes to information that is purely created to make consumers more informed.

We all know you manipulate your measurements. It is well documented on other sites.

Go troll your own forum where your disciples will worship the ground you walk on. I'm tired of having to read about how you and your followers are spreading your "TRUTH" across multiple audio communities, desperate for recognition.
 
Last edited:
Jun 7, 2018 at 6:44 PM Post #8,370 of 13,822
All -

I would encourage everyone *NOT* to discourage @amirm from posting herein this thread. Legitimate questions by me and others and/or rebuttal has already recently been posted, and amirm should be afforded the courtesy (IMO) by all of us to not only allow, but *encourage* his candid reply (replies). Also, keep in mind that one of the negative aspects posted on his audioscience review site is that his (amirm) posts on this site have been deleted, and that he has been "banished" strictly because (paraphrasing in his words) he has cast a negative light on the Yggdrasil. IMO, we should allow amirm (or anyone else) to respond in his own words so as to make him feel "welcome". At a minimum, that will dispel the posts on his site that he cannot offer rebuttal, which IMO lends credibility to his assertions.

If you get confused, listen to the music play :smile_phones:

http://halr.x10.mx/other.html
http://halr.x10.mx/shows.html
 
Last edited:

Users who are viewing this thread

Back
Top