Review and Measurements of Soekris dac1421 Multibit DAC

[restorer-john]

Power supply issues are present as Amir's graph, although a bit lower.

except for the power supply stuff, which he says were not present in his measurements.

You are clearly in a 50Hz country like me.

A-weighting will drop the power supply issues at 60Hz by about 27dB and at 120Hz by about 7dB or so. But we are going on the manufacturer's unweighted S/N spec of 120dB which we are nowhere near.

Comparing unweighted S/N numbers with the RME ADI2 is futile as the reference level is not the same @ 2 volts.

 

[Blumlein 88]

I thought this was true, but then soekris's description sounded like the high level signal wasn't notched. And that made me look up the official audio precision definition of an s/n test, and they also seem to indicate no notch, just residual noise with no signal. Am I wrong on this?

But then you're right about me overlooking the dbv scale. rme was definitely going with the tougher measurement that included noise floor modulation.

It can be done both ways. A signal at max level or 1% distortion to get a max level reference. No signal to get the noise floor. Ratio of the two.

Or you can notch it to get the ratio of the signal vs noise with signal.

DNR is more commonly done as I described with -60 db tone and notching it out.

I suppose Soekris used the more flattering method of SNR.

 

[soekris]

I get plenty of residential noise at my place

Signal to noise is full signal (your rated 2v) to the residual noise in V, expressed in dB. No-one is confusing anything here.

Unless you are deliberately muting the output when you get runs of zero data (which is essentially cheating), there is no way on earth you can get noise levels of less than <1.4uV in an active stage that can swing 2.0V RMS from 20-20KHz.

Amir's APX555 has a residual (20-22K bandwidth) of 2.8uV rated, and typically 2uV. Are you honestly telling us that a US$28,000 industry standard analyzer has a greater residual noise level in its D/A driven analogue generator than your entirely unshielded, single board discrete resistor D/A converter?

Sorry, I don't buy your numbers at all. It's a pretty looking board and congratulations on being brave enough to build a discrete D/A, but until proven otherwise, I'll take the (measured) mid 80s of S/N(+D) as being accurate and that isn't remotely SOTA is it?

The noise of the core R-2R network itself is just the resistor noise, the impedance there is 640 ohm so the noise there is 0.45 uV at 20 Khz bandwidth, with a signal level there of 1.22V....

That would give 129 dB S/N ratio, but then there is the output buffer, which add some noise, resulting in the 120 dB S/N spec. If I have chosen to use a low noise op amp like most others, t.ex. an opa1612 have input refered noise down to 0.15 uV at 20 Khz bandwidth, I would have ended at 128 dB S/N, ratio.... You can never have lower noise than the source impedance noise. And no, I don't do any funny tricks, no need to.

The APx555 have other design requirements than just low noise, like a 100K input impedance and protected inputs, that's why it have more noise. It's common when measuring noise to have an ultra low noise preamplifier in front of your analyzer.

You're welcome to look it all up.

 

[Wombat]

The noise of the core R-2R network itself is just the resistor noise, the impedance there is 640 ohm so the noise there is 0.45 uV at 20 Khz bandwidth, with a signal level there of 1.22V....

That would give 129 dB S/N ratio, but then there is the output buffer, which add some noise, resulting in the 120 dB S/N spec. If I have chosen to use a low noise op amp like most others, t.ex. an opa1612 have input refered noise down to 0.15 uV at 20 Khz bandwidth, I would have ended at 128 dB S/N, ratio.... You can never have lower noise than the source impedance noise. And no, I don't do any funny tricks, no need to.

The APx555 have other design requirements than just low noise, like a 100K input impedance and protected inputs, that's why it have more noise. It's common when measuring noise to have an ultra low noise preamplifier in front of your analyzer.

You're welcome to look it all up.

Some basic references would be helpful.

RebbiP, maybe you can help, here?

 

[restorer-john]

The noise of the core R-2R network itself is just the resistor noise, the impedance there is 640 ohm so the noise there is 0.45 uV at 20 Khz bandwidth, with a signal level there of 1.22V....

That would give 129 dB S/N ratio, but then there is the output buffer, which add some noise, resulting in the 120 dB S/N spec. If I have chosen to use a low noise op amp like most others, t.ex. an opa1612 have input refered noise down to 0.15 uV at 20 Khz bandwidth, I would have ended at 128 dB S/N, ratio.... You can never have lower noise than the source impedance noise. And no, I don't do any funny tricks, no need to.

That's all fine and dandy, but you are quoting 'your' theoretical output noise of the D/A converter resistor noise and completely (again) ignoring the (measured) noise that the rest of your design in its entirety contributes to the output. Quoting a 'resistor noise' uV when there are three close proximity transformers, a whole bunch of high speed logic and switching, little evidence of shielding or other countermeasures is a bit rich.

You cannot quote an unweighted S/N figure of 120dB with respect to 2V and then dismiss the measured noise component like it doesn't matter- it does.

The APx555 have other design requirements than just low noise, like a 100K input impedance and protected inputs, that's why it have more noise. It's common when measuring noise to have an ultra low noise preamplifier in front of your analyzer.

I am talking about the D/A converter driven output (analogue signal generator) in the APX555. It has no 'input impedance' or a 'protected input' or an 'ultra low noise preamplifier in front' as it is an OUTPUT. It is state of the art at -120dB (typical), below 2V.

You're welcome to look it all up.

I have. You can email AP for details on their residual noise in uV on the APX555 generator outputs.

In short, Amir measured SINAD in the mid 80dBs with respect to 2V. You haven't once addressed this poor performance in real terms. Every other parameter is buried in an S/N that is that bad and you know it. It doesn't matter how many bits you have, how many SMD resistors you have soldered in, or how pretty the box looks on your HiFi rack. S/N is the arbiter of High Fidelity. It's where noise dominates signal and at 85dB or so, that's decidedly poor isn't it?

 

[March Audio]

The noise of the core R-2R network itself is just the resistor noise, the impedance there is 640 ohm so the noise there is 0.45 uV at 20 Khz bandwidth, with a signal level there of 1.22V....

That would give 129 dB S/N ratio, but then there is the output buffer, which add some noise, resulting in the 120 dB S/N spec. If I have chosen to use a low noise op amp like most others, t.ex. an opa1612 have input refered noise down to 0.15 uV at 20 Khz bandwidth, I would have ended at 128 dB S/N, ratio.... You can never have lower noise than the source impedance noise. And no, I don't do any funny tricks, no need to.

The APx555 have other design requirements than just low noise, like a 100K input impedance and protected inputs, that's why it have more noise. It's common when measuring noise to have an ultra low noise preamplifier in front of your analyzer.

You're welcome to look it all up.

Just to clarify this, you only consider resistor thermal noise, there is no noise from the preceding power supply and switch of the resistor matrix?

 

[Veri]

View attachment 14593

So either I'm doing something wrong (likely!) or the DX7s is pretty disappointing.

While the Soekris is very similar to what amir has showed it seems your DX7 unit measures particularly bad! 0.02% THD, more than 10x below claimed performance. I would think something to be up with either the set-up or the device, especially considering the MANY distortion spikes present in the DX7 but not the Soekris.

 

[soekris]

I can't really comment on Amir's measurement, and don't really care that much. I of course base my specifications on my own measurements and are very satisfied with the performance of my DACs.

The APx555 Analog Source or D/A converter output noise doesn't really matter, it's the A/D part that is used for the measurements.

 

[soekris]

Just to clarify this, you only consider resistor thermal noise, there is no noise from the preceding power supply and switch of the resistor matrix?

When there is no signal then all the resistors in the R-2R network are connected to signal ground, per Sign Magnitude DAC principles....

 

[Blumlein 88]

When there is no signal then all the resistors in the R-2R network are connected to signal ground, per Sign Magnitude DAC principles....

So what kind of result do you get for the dynamic range test with a -60 db signal present? Notching it out afterwards and seeing what is left. Eyeballing it and assuming you have no power supply pollution it looks like it might be 112 db or so.

It also would appear the way Amir measures isn't the limit in this case, as several other DUT have lower noise than the Soekris is showing. Though again the main issue isn't noise in Amir's tests so much as higher levels of THD. The noise looks at a respectably low level though not down to -120 db from what is being shown.

@amirm could you do the test with -60 db 1 khz tone to show us the noise floor measurement done that way?

 

[DonH56]

I have noticed a lot of times that SNR (signal-to-noise ratio) as reported by various tests is really SINAD (signal-to-noise-and-distortion). You can notch out the harmonics using digital processing, but then you either notch out a tiny bit of noise or add back in a little bit inferred from the noise around the distortion spurs. And THD measurements all too often also include noise, particularly at low signal levels where the actual distortion spurs are buried in the noise floor and more difficult to extract.

There are a lot of noise sources beyond just resistor thermal noise, and of course many other sources of "noise" from power supply noise to crosstalk to clock noise etc., but it hardly seems worth the effort to divert into all that in this thread...

 

[dc655321]

So in the example above the noise floor is bouncing along at -134 db which with FFT gain would be equivalent to -100 or 102 db.

Isn't FFT gain approximately, 10*log(M/2), for an FFT of length M?

Using an FFT of length 2^12 (4096) gives a gain of ~33dB.
AFAIU, these measurements use an FFT with significantly more length... What am I missing?

 

[amirm]

@amirm could you do the test with -60 db 1 khz tone to show us the noise floor measurement done that way?

I won't be home for the next few days so can't run new tests.

 

[gvl]

I won't be home for the next few days so can't run new tests.

We should crowd-fund your clone so that we don't have to wait while you're away.

 

[Blumlein 88]

Isn't FFT gain approximately, 10*log(M/2), for an FFT of length M?

Using an FFT of length 2^12 (4096) gives a gain of ~33dB.
AFAIU, these measurements use an FFT with significantly more length... What am I missing?

Yes that's right. I took it Amir is using 32k fft which would be a gain of 42 db. And the graph I keyed on was with +12 db signal.

 

[Thomas savage]

We should crowd-fund your clone so that we don't have to wait while you're away.

Ok but let’s make some tweaks to the genetic code...

 

[gvl]

Ok but let’s make some tweaks to the genetic code...

I agree, the gene responsible for the urge to install manufacturer-provided drivers needs attention.

 

[DonH56]

Judging by his new avatar, I would be very leary about letting Sir Thomas tweak The King's DNA...

 

[dc655321]

Yes that's right. I took it Amir is using 32k fft which would be a gain of 42 db. And the graph I keyed on was with +12 db signal.

Ah, I missed the +12dB. Thanks.

 

[DuxServit]

I can't really comment on Amir's measurement, and don't really care that much. I of course base my specifications on my own measurements and are very satisfied with the performance of my DACs.

As an audio consumer I care about Amir’s measurements and pay close attention to the response of a vendor to these measurements.

A great example was John Siau in the Benchmark DAC3 thread. He responded very well to the questions surrounding Amir’s measurements of the DAC, and showed his deep engineering/design skills. I very much respect that. As such, Benchmark is on my “good vendors list”.