PS Audio DirectStream DAC Mk.2 measurements

[amirm]

Another funny post from PS Audio Forum:
wudai_e
4h
Also the thing with measurement, If I have a predetermined bias before hand, I can make the measurement results to support the said bias.
I’d take any audio measurements with a huge grain of salt unless the methology is published and can be verified by an unbiased party.

What??? How ridicules is that?

Instead of that, why doesn't he ask for the manufacturer to show measurements that demonstrate otherwise? If manufacturer can't or won't do that, then that is that and independent data remains as authoritative.

 

[AdrianusG]

Instead of that, why doesn't he ask for the manufacturer to show measurements that demonstrate otherwise? If manufacturer can't or won't do that, then that is that and independent data remains as authoritative.

Exactly that!

 

[mdsimon2]

@jkr can you make a -60 dBFS generator level THD+N measurement of the Mk2 with Antero and Massive? And also show the FFT spectra at -60 dBFS generator level? Would like to explore the low level noise performance a bit more.

Thanks!

Michael

 

[jkr]

@jkr can you make a -60 dBFS generator level THD+N measurement of the Mk2 with Antero and Massive? And also show the FFT spectra at -60 dBFS generator level? Would like to explore the low level noise performance a bit more.

Thanks!

Michael

Sorry. We already have spent too much time on this. We are swamped with paid work right now. (Orders have tripled since we left the PS Audio Forum). PS Audio should provide these measurements. I'm also tired of all the ignorant responses from the PS Audio forum members.

Instead of that, why doesn't he ask for the manufacturer to show measurements that demonstrate otherwise? If manufacturer can't or won't do that, then that is that and independent data remains as authoritative.

Agree 100%

 

[ahofer]

Another funny post from PS Audio Forum:
wudai_e
4h
Also the thing with measurement, If I have a predetermined bias before hand, I can make the measurement results to support the said bias.
I’d take any audio measurements with a huge grain of salt unless the methology is published and can be verified by an unbiased party.

What??? How ridiculous is that?

People do play games with graph scale and threshold choices, but that's still a far cry from veils lifting and PRaT. Rationalization of deeply-held beliefs is a heckuva drug.

 

[kevin1969]

Instead of that, why doesn't he ask for the manufacturer to show measurements that demonstrate otherwise? If manufacturer can't or won't do that, then that is that and independent data remains as authoritative.

I think the new myth is that DSD based DACs measure differently.

DirectStream DAC MK2 MountainTop - Mount Massive

https://www.stereophile.com/content/playback-designs-mps-5-sacdcd-player-manufacturers-comment This has been discussed before, but it seems appropriate to provide a brief reminder of why the traditional noise measurements for a DSD-based DAC can be confusing or misleading. I have provided a...

forum.psaudio.com

 

[AnalogSteph]

I think the new myth is that DSD based DACs measure differently.

I mean, it is entirely possible even for the AP frontend to become overwhelmed with ultrasonic noise (slew rate limit) if you crank up input gain:

The culprit in this case (a basically unfiltered CS4272 DAC output, giving over 0.4% of THD+N in a 90 kHz bandwidth):

Very unlikely to happen with input gain dialed back to accept full-scale output though.

This is why you always measure Class D amplifiers with a lowpass filter - they spit out a crapton of ultrasonic noise.

It goes without saying that no DAC should be generating enough ultrasonic noise to upset measurement gear even in a basic fullscale THD+N measurement (what's it going to do with consumer-level electronics then?). Passive loads are a lot more tolerant of this sort of thing.

 

[DonR]

"DSD noise is more musical." There, no more discussion is required.

 

[amirm]

I think the new myth is that DSD based DACs measure differently.

DirectStream DAC MK2 MountainTop - Mount Massive

https://www.stereophile.com/content/playback-designs-mps-5-sacdcd-player-manufacturers-comment This has been discussed before, but it seems appropriate to provide a brief reminder of why the traditional noise measurements for a DSD-based DAC can be confusing or misleading. I have provided a...

forum.psaudio.com

They seem to claim to have data adaptive filtering. That makes no sense to me. Nor have I heard of any such implementation.

 

[antcollinet]

They seem to claim to have data adaptive filtering. That makes no sense to me. Nor have I heard of any such implementation.

That seems to be a generous interpretive / assumptive view of statements from Playback Designs

I can take a generous view of it though and interpret them as saying their filters are not static and therefore the measured response to a non-musical impulse may not tell you anything useful about the way the DAC filters the waveforms of sounds we actually listen to.

Perhaps just guessing? (I haven't seen the Playback Design statement that is being "interpreted")

 

[kevin1969]

Getting exciting over there.

DirectStream DAC MK2 MountainTop - Mount Massive

For those of you interested in measurements, I am including the long-promised measurements between Antero (what the DSMKII shipped with) and Massive. You can see in the Antero measurement the ultrasonic noise mentioned in the Stereophile review. You can also see the audible noise. And to be...

forum.psaudio.com

 

[NTK]

What is really troubling is how noisy the PS Audio DAC is. Here is the measurement of the beta version of Massive from PS Audio (source).

Here is the Sabaj A20d 2023 Amir reviewed. The noise floor of the PS Audio DAC is >15 dB higher than the distortion peaks of the Sabaj!
The PS Audio DAC has absolutely terrible signal-to-noise ratio.

 

[voodooless]

What is really troubling is how noisy the PS Audio DAC is. Here is the measurement of the beta version of Massive from PS Audio (source).
View attachment 309182

Here is the Sabaj A20d 2023 Amir reviewed. The noise floor of the PS Audio DAC is >15 dB higher than the distortion peaks of the Sabaj!
The PS Audio DAC has absolutely terrible signal-to-noise ratio.

Hold on! You can’t compare the noise floors if you don’t know the FFT sizes and averaging settings.

 

[VQR]

What is really troubling is how noisy the PS Audio DAC is. Here is the measurement of the beta version of Massive from PS Audio (source).
View attachment 309182

Here is the Sabaj A20d 2023 Amir reviewed. The noise floor of the PS Audio DAC is >15 dB higher than the distortion peaks of the Sabaj!
The PS Audio DAC has absolutely terrible signal-to-noise ratio.

The graph on their forum isn't even level adjusted to 0 dB. The 2nd and 3rd harmonics are worse than -80 dB with no multitone test shown... I can't imagine why BS Audio hates measurements. I bet the ENOB would be ~10 bits. I mistook this for the current update. Still terrible, of course.

 

[mdsimon2]

Hold on! You can’t compare the noise floors if you don’t know the FFT sizes and averaging settings.

Exactly, very easy to change the apparent noise floor on a FFT by changing FFT length.

Assuming similar FFT lengths were used between the two firmware measurements, I don't really see an audio band difference. Although it looks to me like they used more averaging on the second one to give the appearance of less noise to the uneducated.

Michael

 

[jkr]

Still SINAD at best would be 70 dB. Same as what we posted last week when I was accused of posting bias measurements and badmouthing PS Audio.

 

[mdsimon2]

I'm probably in the minority but with a design like this I really don't care about distortion. Sure, you can reduce it with better transformers but it is never going to be great.

The noise on the other will be audible and I don't see an excuse for it being so high, especially with all the talk about how the Mk2 sounds better because of lower noise.

Michael

 

[jkr]

Sure, you can reduce it with better transformers

True, however this was not our main goal. Remember the stock iron core transformer has a high frequency roll off of almost a full dB at 20khz and has a less than perfect output level and impedance.

 

[NTK]

Hold on! You can’t compare the noise floors if you don’t know the FFT sizes and averaging settings.

It is true that without knowing the FFT length and sampling frequency, you can't calculate the actual noise level. However, it is easy enough to make an estimate from the just looking at the FFT plot.

Assuming a lenient FFT frequency bin width of 6 Hz (Δf = fs/n_samples, assuming fs = 384 kHz, that would mean at sample length of 64 k samples):

• Eyeballed average grass height = -115 dB
• If reference voltage is 1 voltage unit, noise voltage magnitude of each bin = 10^(-115/20) = 1.78e-6 units
• From 20 - 20000 Hz, there are (20000 -20)/ 6 = 3330 bins
• Equivalent sum of the noise voltage is calculated using root sum of squares: V_sum = sqrt( 3330 * 1.78e-6^2 ) = 1.0e-4 units
• Signal-to-noise ratio = 20 log10(1.0e-4/1.0) = -80 dB (i.e. 35 dB of FFT processing gain)

The signal to noise ratio, with the above assumptions, is just 13 bits! It will be worse if the measurement FFT sample length was longer, which is actually likely to be the case.

 

[mdsimon2]

It is true that without knowing the FFT length and sampling frequency, you can't calculate the actual noise level. However, it is easy enough to make an estimate from the just looking at the FFT plot.

Assuming a lenient FFT frequency bin width of 6 Hz (Δf = fs/n_samples, assuming fs = 384 kHz, that would mean at sample length of 64 k samples):

• Eyeballed average grass height = -115 dB
• If reference voltage is 1 voltage unit, noise voltage magnitude of each bin = 10^(-115/20) = 1.78e-6 units
• From 20 - 20000 Hz, there are (20000 -20)/ 6 = 3330 bins
• Equivalent sum of the noise voltage is calculated using root sum of squares: V_sum = sqrt( 3330 * 1.78e-6^2 ) = 1.0e-4 units
• Signal-to-noise ratio = 20 log10(1.0e-4/1.0) = -80 dB (i.e. 35 dB of FFT processing gain)

The signal to noise ratio, with the above assumptions, is just 13 bits! It will be worse if the measurement FFT sample length was longer, which is actually likely to be the case.

Not that it matters but to be fully charitable full output voltage is 3.7 V, not 1 V, so SNR is about 11 dB better than your calculation. Hi-fi news measured a-weighted SNR at 92.5 dB so you probably aren't far off. Still see no evidence of a 10 dB improvement in audible noise between the Antero and Massive firmware.

Michael