latest Schiit Yggdrasil ads on Stereophile May issue

[xxie]

I am flipping the latest Stereophile May issue, and I saw Schiit Yggdrasil ads, the ads title is "slightly less obsolete", and it provide a picture from tinyurl.com/yggy-meas, the notes says:
"Yggdrasil with Gen 5 USB and Analog 2 updates showing -144dB, 1Khz sine wave output".

I did see there is -144dB at 1Khz on FFT, but there is 120Hz peak at -135dB. I am just wondering what is this ads trying to say about "slight less obsolete"? It is a multi-bit DAC. Is sigma-delta 24bit DAC cannot produce it?

 

[amirm]

The ad is saying whoever approved that ad copy doesn't know what that measurement means as you explain.

There are more distortion products than there is signal. No way a DAC can produce clean -144 dB sine wave:

This is the ad in question as posted by Ray earlier:

 

[garbulky]

I am flipping the latest Stereophile May issue, and I saw Schiit Yggdrasil ads, the ads title is "slightly less obsolete", and it provide a picture from tinyurl.com/yggy-meas, the notes says:
"Yggdrasil with Gen 5 USB and Analog 2 updates showing -144dB, 1Khz sine wave output".

I did see there is -144dB at 1Khz on FFT, but there is 120Hz peak at -135dB. I am just wondering what is this ads trying to say about "slight less obsolete"? It is a multi-bit DAC. Is sigma-delta 24bit DAC cannot produce it?

I think it means they are saying it can do it. How many obsolete dacs can do that, I guess. However, now interestingly they give an actual meaasurement here. It would be interesting to see how they compare.

 

[amirm]

One more thing: this little notation:

Means there is 40 dB of noise gain. If this is added up to the noise floor, there would be no trace of that 1 kHz tone.

 

[xxie]

One more thing: this little notation:

View attachment 12097

Means there is 40 dB of noise gain. If this is added up to the noise floor, there would be no trace of that 1 kHz tone.

Amirm, can you help me understand this? How does 16k fft points translate 40dB noise gain here? I did see your measurement has higher noise floor on the same dac. Thanks.

 

[amirm]

Amirm, can you help me understand this? How does 16k fft points translate 40dB noise gain here? I did see your measurement has higher noise floor on the same dac. Thanks.

There is a thing called "FFT gain." What happens is that when we transform the signal from time domain to frequency, we spread the noise across many bins. The more bins, the less energy in each one. The less energy in each one, the lower the displayed noise in the graph. Our main tone however (1 kHz) in this case, occupies one bin so it never spreads out and maintains its amplitude.

Also, in my measurements I am testing very strong signals (e.g. 0 dBFS). This causes both the DAC and Analyzer to have higher noise floors (called "noise modulation"). Reducing the signal to near zero at -144 dBFS as is done in their tests, reduces the noise levels yet again. Here is the same test but run on RME ADI-2 DAC:

Our main tone is reproduced at -143 dB. That one dB of error is due to windowing and accuracy issues at such a low amplitude (about 22 millionth of a volt).

As you see, unlike the Schiit Yggdrasil, there are no other pronounced peaks around power supply harmonics. And what is there is at the limits of the analyzer and at any rate, below the level of our 1 kHz tone.

 

[xxie]

As you see, unlike the Schiit Yggdrasil, there are no other pronounced peaks around power supply harmonics. And what is there is at the limits of the analyzer and at any rate, below the level of our 1 kHz tone.

thanks so much for explaining this!

 

[FrantzM]

Early in the morning and .. Thank God for ARS ...

No BSchiit seems to be able to fly under ASR radar! Excellent!!

 

[SchwarzeWolke]

There is a thing called "FFT gain." What happens is that when we transform the signal from time domain to frequency, we spread the noise across many bins. The more bins, the less energy in each one. The less energy in each one, the lower the displayed noise in the graph. Our main tone however (1 kHz) in this case, occupies one bin so it never spreads out and maintains its amplitude.

Also, in my measurements I am testing very strong signals (e.g. 0 dBFS). This causes both the DAC and Analyzer to have higher noise floors (called "noise modulation"). Reducing the signal to near zero at -144 dBFS as is done in their tests, reduces the noise levels yet again.

Ah, now that makes a lot of sense to me now! Thanks Amir for the really good explanation! What I'm asking myself is if Atomicbob and the dudes at Schiit really don't know what they're measuring and what the results indicate or if they are doing it deliberatley which would be really something... I mean, after seeing the 1kHz spike, I was wondering myself how anybody with a sane mind would see this measurements as a testament for the really good performance of the Yggy. We also discussed this in the past when we looked at the linearity and reproduction of a sine wave measurements from AB. Well, what I really don't like is that Schiit and AB are playing the "we're objectivist or at least we do measurements" card.

 

[Palladium]

Early in the morning and .. Thank God for ARS ...

No BSchiit seems to be able to fly under ASR radar! Excellent!!

How dare you offend the Schiit teapots apologia around these parts.

Well, I wish Amir has a chance to test the cheaper Benchmark DAC3 which will really show just how "slightly less obsolete" the Ygg is when a ~$400 competitor is already running circles around it.

 

[SIY]

How do they know it's the signal and not an idle tone?

 

[Soniclife]

There are more distortion products than there is signal

Their slogan really should be More Schiit than signal.

 

[amirm]

Copying this post from the Yggdrasil review thread to here just in case someone doesn't read it there:

----

You really can't blame a manufacture to present a counterpoint to what they might feel as a misrepresentation of their products.

It is a misrepresentation in the ad to imply that any DAC, theirs or others, can remotely reproduce a -144 dB signal. As I explained in the other thread, the measurement they are showing indicates lots of other noise sources that are higher amplitude than the test tone. So whatever they think of my work, is beside the point. They are using measurements that they either don't understand or intentionally hope to mislead the reader.

Let's show this in a graph where I replicate their test conditions but using the superbly measuring RME ADI-2 DAC. For now, focus on the left side only:

We see a very clean 1 kHz tone despite hugely small signal (rightmost bit of 24-bit signal). None of the extra noise products are visible that Schiit shows with their DAC in the same situation.

But now let's look at what that same "sine wave" looks without transformation by FFT on the right. We see that it is nothing but random noise. That 1 kHz tone is somewhere in there but is dwarfed by noise. So our DAC here despite being worlds better than Yggdrasil, still cannot remotely reproduce a 24-bit signal. Laws of physics simply do not allow it.

How does the FFT then show such clean 1 kHz tone? That is because of "signal processing." The FFT is able to provide huge amount of noise reduction by spreading the noise across the full spectrum. The tone however stays put (i.e. not reduced) so we see it clearly.

Conclusions
The measurements shown by Schiit are a) not that good anyway and b) are hugely misleading to readers. No DAC in the world at room temperature and with audio bandwidth can reproduce a signal at the bottom of a 24-bit PCM sample. Various noise sources in electronics including our measurement gear make that a huge impossibility.

This is why my/industry standard linearity tests stop at just -90 dB.

 

[Blumlein 88]

I'm no fanboy of the brand, but I don't really see a problem with the graph in the ad by itself. In principle a DAC can put out such a signal, which is down in the noise. I understand FFT gain and how such a signal would be down in the weeds of the noise. Amir even shows this is possible with the RME results.

Now why is it with less FFT gain than Amir AB has a graph running along at -165 db with the -144 db tone sticking up when Amir's test was showing more like -145-150 db for the same situation? It might be possible for a dithered 20 bit DAC to show such a thing, but it is a marginal possibility. Granting AB is being honest with us maybe a dialog about how he does testing and where the difference comes in is worthwhile.

I can think of ways to dummy up such a result to look halfway possible and post results. The only way to avoid that is following same or similar principles and do the tests right. Otherwise test results in isolation mean nothing.

AB said he measured 120 db SNR (or was it dynamic range) which if so would put the graph in the range of what he shows. So what did he do differently? Otherwise we are implying he cheated on the test or at best the test was bungled. Same kind of accusations fanboys have been hurling at Amir's results.

 

[amirm]

AB said he measured 120 db SNR (or was it dynamic range) which if so would put the graph in the range of what he shows. So what did he do differently? Otherwise we are implying he cheated on the test or at best the test was bungled. Same kind of accusations fanboys have been hurling at Amir's results.

Without having access to the same unit, it is hard to pinpoint what is going on. After all, I am testing an entirely different DAC which has DSP functionality which may be impacting noise at such low levels.

To be clear, I am not implying that his measurements are wrong at all. My point is that this is not a graph to show people, making them think DACs reproduce 24 bit signals. They just done.

 

[hvbias]

It seems to me like the most egregious thing is the low frequency roll off with the Analog 2 update. Was this discrepancy ever explained between Amir and atomicbob's measurements?

 

[amirm]

It seems to me like the most egregious thing is the low frequency roll off with the Analog 2 update. Was this discrepancy ever explained between Amir and atomicbob's measurements?

No, his measurements don't show this. This was enough of a surprise that I drove another 150 miles to re-test the DAC at another location. The results were the same.

Seeing how it occurs in both channels, that rules out something being broken. That leaves us with some design change that was rolled in later in production, or a mistake in part selection and assembly.