Review and Measurements of SMSL SU-8 DAC

[gvl]

Yeah, seems that way.

Although even in balanced, it cannot match the D50 performance...

Yeah, you'd want SU-8 for balanced but it is a fail there with Yggdrasil-like performance. Just stick with the D50.

EDIT: looks like I misread linearity data, balanced linearity actually looks good. SINAD is disappointing still, but overall better than the Yggdrasil.

 

[maxxevv]

Was just wondering regarding the SMPTE 'hump'. Other than the DAC chip what else might be common between them? Other than the D50, DX7s seems to have it too right ?

Did the NX4 DSD exhibit that ? (don't recall seeing that)

Could it be the dual chip implementation or an opamp they use in common that's causing that ?

 

[Krunok]

I ran it again (not shown) at -5 dB and it was still worse than unbalanced (in red) but much more manageable.

If one is to use SU-8 with volume control than maybe 0dB performance is not that critical as nobody listens music with volume crank up to the max. Would it be too much to ask that you repeat these measurements at -10dB level?

 

[Krunok]

You already know my question- is the deviation from "linearity" because of actual deviation from linearity or is it from noise?

From what I see balanced and unbalanced have pretty much the same noise level and yet linearity is not the same.

 

[SIY]

The SINAD of the two is different.

Like I posted a few days ago, the easiest way to tease out the distinction is to run the linearity sweep several times and see what the power average looks like.

 

[t129]

What about the pièce de résistance: Sound colour settings? What does that even mean and do they make a measurable difference? Thumbs up for the acces to chip dsp though...

Too bad about the linear(?) powersupply shenagians at this price range. Wallwart would have worked much better as in other SMSLs, but alas, audiophile marketing...

 

[Krunok]

The SINAD of the two is different.

But only for 4dB. Don't you think difference in THD affects linearity more than difference in SINAD?

 

[Krunok]

The SINAD of the two is different.

Like I posted a few days ago, the easiest way to tease out the distinction is to run the linearity sweep several times and see what the power average looks like.

It doesn't make sense to me: balanced output has far worse SINAD and THD and still has almost eprfect linearity, while unbalanced has very decent SINAD and THD but sucks in linearity. What is the explanation for this?

 

[andreasmaaan]

It doesn't make sense to me: balanced output has far worse SINAD and THD and still has almost eprfect linearity, while unbalanced has very decent SINAD and THD but sucks in linearity. What is the explanation for this?

The explanation is that SMSL they f***ed up the balanced outputs such that clipping occurs before 0dBfs.

This case is specific to this DAC, in other words.

EDIT: "clipping" is overstating it as it never reaches 1%. "Saturating" as Amir said, is better. Seems to start at about -6dBfs. Below this output level, non-linear distortion is slightly lower from the balanced outputs at all levels.

 

[Krunok]

The explanation is that SMSL they f***ed up the balanced outputs such that clipping occurs before 0dBfs.

This case is specific to this DAC, in other words.

That doesn't explain it - unbalanced output doesn't suffer from clipping but has linearity issue.

Btw, clipping at 0dB is not an issue in my eyes as long as it works well at lower levels. As I said, I can't imagine anybody cranking the colume to 0 dB.

 

[andreasmaaan]

That doesn't explain it - unbalanced output doesn't suffer from clipping but has linearity issue.

That agrees with my point, doesn't it? I said that the balanced output starts saturating at -6dBfs, not the unbalanced output. Up until -6dBfs, the balanced output outperforms the unbalanced one in non-linear distortion, which is consistent with it having better linearity.

I'm not sure why the unbalanced output has so much worse linearity, however. Perhaps SMSL made a whole separate mistake on that output, too?

 

[andreasmaaan]

Actually, I think it makes sense. It's probably the power supply noise on the unbalanced outputs, which is at about -100dB. This would not go down in relation to the signal, so at -90dBfs you would have power supply effects at -10dBfs and this would explain the poor low level linearity.

 

[Krunok]

Actually, I think it makes sense. It's probably the power supply noise on the unbalanced outputs, which is at about -100dB. This would not go down in relation to the signal, so at -90dBfs you would have power supply effects at -10dBfs and this would explain the poor low level linearity.

And why is balanced output not affected although it has more noise and more THD?

 

[andreasmaaan]

And why is balanced output not affected although it has more noise and more THD?

The balanced outputs only suffer from these problems at higher than -6dBfs. The key graph to look at is the intermodulation distortion graph. You can see the balanced outputs have slightly better IM below -6dB, and then it rises rapidly. This is the specific design flaw (the "saturation") Amir mentions.

The linearity test is focused on the other end of the spectrum, i.e. below -90dBfs, where the power supply effects become decisive.

 

[andreasmaaan]

Sorry, just edited my post, above. Had written "unbalanced" instead of "balanced". Now it is correct.

 

[Krunok]

We should have THD and noise measurement at -90dB for balanced and unbalanced to see the reason for linearity problem.

Although, no one is likely to listen to music at -90dB as much as at 0dB. But it would be nice to have those measurementsto get the whole picture.

 

[SIY]

It doesn't make sense to me: balanced output has far worse SINAD and THD and still has almost eprfect linearity, while unbalanced has very decent SINAD and THD but sucks in linearity. What is the explanation for this?

We don't actually know what the low-level linearity is since, the way this measurement is done, it's wrapped up in the noise. Power averaging would help distinguish.

Indeed at higher levels, there's clearly a problem in the analog implementation. No excuse- there's single chip single-ended-to-balanced-differential solutions that will greatly outperform this.

 

[Krunok]

We don't actually know what the low-level linearity is since, the way this measurement is done, it's wrapped up in the noise. Power averaging would help distinguish.

Indeed at higher levels, there's clearly a problem in the analog implementation. No excuse- there's single chip single-ended-to-balanced-differential solutions that will greatly outperform this.

I agree. But as long as this DAC is used with volume control why worry about how it performs at 0dB or at -90dB as long as it performs good in between?

I would be interested to see THD+N at say -60dB, -40dB and -20dB as that is where this device will mostly be used. Don't you agree?

 

[Blumlein 88]

I agree. But as long as this DAC is used with volume control why worry about how it performs at 0dB or at -90dB as long as it performs good in between?

I would be interested to see THD+N at say -60dB, -40dB and -20dB as that is where this device will mostly be used. Don't you agree?

You get a good view of this in the IMD test already posted by Amir. That is a test of IMD vs level.

 

[andreasmaaan]

I agree. But as long as this DAC is used with volume control why worry about how it performs at 0dB or at -90dB as long as it performs good in between?

I would be interested to see THD+N at say -60dB, -40dB and -20dB as that is where this device will mostly be used. Don't you agree?

The IM graph gives an excellent indication of how it would perform at these levels.