Review and Measurements of SMSL SU-8 DAC

[andreasmaaan]

Exactly. But I thought those are not called "harmonics".

Ok, let's be clear: if you send one tone that tone will be played along with its harmonics. If you play 2 tones those 2 tones are gonna be played along with their harmonics and with some new IM crap. If you play 3 tones those 3 tones will be played along with the harmonics of all 3 tones and some more IM crap. I believe we are all speaking of the same thing so let's not waste time on this.

Yeh we definitely are I just wanted to be clear because in the standards these are specifically called IM harmonics and they are numbered 2, 3, etc.., and there are disagreements between the various standards as to which IM harmonics should be counted in the IMD specification.

 

[Krunok]

You can make a very confident prediction about THD at -60dBfs by looking at the IM graph. You can make an educated guess about THD+N by eyeballing the IM graph and then adding [100 (level of PS spuriae) + level (in -dBfs)] for the unbalanced outputs, or by adding [120 (level of PS spuriae) + level (in -dBfs)] for the balanced outputs.

I agree, that would be a reasonable assumption, but I still think only measurements at -60dBfs would provide a final say.

 

[andreasmaaan]

To (maybe) wrap up this little discussion lol, my prediction is that if you use the volume control on the DAC and don't have a really ridiculous gain structure, there won't be an audible difference between balanced and unbalanced outputs

 

[Krunok]

To (maybe) wrap up this little discussion lol, my prediction is that if you use the volume control on the DAC and don't have a really ridiculous gain structure, there won't be an audible difference between balanced and unbalanced outputs

I agree. They would both probably measure pretty much the same as well. And that is why the unbalanced output linearity issue is still puzzling me.

 

[andreasmaaan]

I agree. They would both probably measure pretty much the same as well. And that is why the unbalanced output linearity issue is still puzzling me.

You don't think this is simply attributable to the power supply spuriae? It seems to make perfect sense to me.

 

[Krunok]

You don't think this is simply attributable to the power supply spuriae? It seems to make perfect sense to me.

No, I don't think so - that seems too little for that to me.

Beside, balanced was unaffected because of its "balanced" nature?

 

[andreasmaaan]

No, I don't think so - that seems too little for that to me.

Beside, balanced was unaffected because of its "balanced" nature?

Have a look at post #32.

The power supply spuriae are at a bit under -100dBfs on the unbalanced outputs, which means that, at -90dBfs, it is only about 10 dB below the signal.

The power supply stuff is at about -120dBfs on the balanced outputs, which means that it is around 30dB below the signal at -90dBfs.

So of course, linearity is far worse on the unbalanced outputs.

This doesn't explain what they did in the design to make it this way though.

 

[Krunok]

Have a look at post #32.

The power supply spuriae are at a bit under -100dBfs on the unbalanced outputs, which means that, at -90dBfs, it is only about 10 dB below the signal.

The power supply stuff is at about -120dBfs on the balanced outputs, which means that it is around 30dB below the signal at -90dBfs.

So of course, linearity is far worse on the unbalanced outputs.

This doesn't explain what they did in the design to make it this way though.

But look at the shape of the red line. Why didn't the power junk add up to the signal? How exactly is linearity measured?

 

[andreasmaaan]

But look at the shape of the red line. Why didn't the power junk add up to the signal? How exactly is linearity measured?

More info here: https://www.audiosciencereview.com/forum/index.php?threads/understanding-audio-measurements.2351/

I think Amir would need to answer your question in full though. I don't fully understand how he plots the line in the linearity measurement.

 

[SIY]

But look at the shape of the red line. Why didn't the power junk add up to the signal? How exactly is linearity measured?

It's a standard function in the AP software. Basically, the DAC is fed with a series of sine waves (Amir uses 200 Hz) starting at 0 dBFS and stepping down 1 dB per data point until -120 dBFS is reached. It can also be run the other way, starting at -120 dBFS and ending at 0 dBFS. The analog output is then measured and ratioed against the theoretical output.

A bandpass filter is applied around the test frequency which excludes a lot of noise from the measurement (but as my recent post showed, there's still a strong noise influence at very low levels).

 

[Krunok]

.

A bandpass filter is applied around the test frequency which excludes a lot of noise from the measurement (but as my recent post showed, there's still a strong noise influence at very low levels).

Ok, I see, thank you. And how narrow is the bandpass filter? Noise peaks from power supply seem to be around 180Hz and 250Hz.

 

[SIY]

Hz dB
100 -50
190 -0.102
200 -0.102
210 -0.102
300 -50

 

[Krunok]

Hz dB
100 -50
190 -0.102
200 -0.102
210 -0.102
300 -50

In that case I don't think the noise from power supply influenced the result, especially taking into account odd shape of the red curve (partly lower and partly higher than 0dB).

Btw, if that linearity test is a single tone amplitude sweep than it is not very informative. It shows the same narrow picture as a THD figure measured on a single frequency.

 

[andreasmaaan]

Btw, if that linearity test is a single tone amplitude sweep than it is not very informative. It shows the same narrow picture as a THD figure measured on a single frequency.

Have a look at the THD+N vs frequency graph. I know we can only infer, but it does show that THD+N is extremely consistent across the whole bandwidth (apart from when the balanced outputs begin saturating).

 

[SIY]

Btw, if that linearity test is a single tone amplitude sweep than it is not very informative. It shows the same narrow picture as a THD figure measured on a single frequency.

I don't think so. The linearity test is measuring how even the steps are between amplitude values. Nonlinearity from other causes (e.g., rising distortion with frequency at the analog output gain stage) is a separate measurement, which shows up in THD vs frequency sweeps- the DAC chip itself doesn't have steps that change with frequency, so this test does a good job of determining what the actual conversion is doing.

 

[andreasmaaan]

@SIY I don't understand your answer to the question about how narrow the bands are. I'm used to expressing filter bandwidth in either Q or order, and I can't work out how to convert your expression in dB into a format I can understand. Would you mind helping?

 

[SIY]

Q is only applicable to 2nd order filters.

If you plot the values, you can easily see the filter shape.

 

[Krunok]

I don't think so. The linearity test is measuring how even the steps are between amplitude values. Nonlinearity from other causes (e.g., rising distortion with frequency at the analog output gain stage) is a separate measurement, which shows up in THD vs frequency sweeps- the DAC chip itself doesn't have steps that change with frequency, so this test does a good job of determining what the actual conversion is doing.

But wouldn't the true picture only be revealed if linearity test is done through the whole frequency range?

The same goes for THD - wouldn't the true picture require 3 dimensions: THD measured for frequency range x output level (amplitude)?

It is not the performance of the DAC chip that concerns me but the electronics behind the DAC chip (I/U converter, filter and output buffer).

 

[andreasmaaan]

But wouldn't the true picture only be revealed if linearity test is done through the whole frequency range?

The same goes for THD - wouldn't the true picture require 3 dimensions: THD measured for frequency range x output level (amplitude)?

It is not the performance of the DAC chip that concerns me but the electronics behind the DAC chip (I/U converter, filter and output buffer).

Yeh, but we can never hope to have the complete picture, just good enough data to make educated inferences about the complete picture, surely?

 

[SIY]

The D to A conversion doesn't care about frequency, within reason. So you get no more information by running linearity at multiple frequencies. What you do by that is to conflate D to A nonlinearity with the analog linearity. You're best using a relatively low frequency ( like 200 Hz, or 1 kHz, or...) to determine D to A linearity, then a separate sweep at a fixed amplitude (preferably near the max) of THD vs frequency. That keeps the two mechanisms separate.