Do you need linear power supply for DACs?

[levimax]

If you are really worried about PS problems it is easy and cheap to try battery power. I tried it and could tell no difference on my SMSL M8 so went back to the stock PS. I do use battery power on my pre-amp.

 

[Snafu]

^ agree, D50 runs ~10hours with 10.500mAh powerbank. When pb runs out D50 just turns itself off, no fading or SQ chance.

 

[suttondesign]

It seems to me that if you inserted a 4-pin regulator into the defillibration output cap, that would linearize the capacitance leakage, proving that more devices in the chain is always better. And I totally know I'm talking about!

 

[solderdude]

^ agree, D50 runs ~10hours with 10.500mAh powerbank. When pb runs out D50 just turns itself off, no fading or SQ chance.

Powerbanks are just SMPS that aren't connected to mains.
The battery inside the powerbank does not feed the D50.
It powers an SMPS (DCDC converter) that delivers the regulated 5V (regardless of the battery voltage)

 

[Snafu]

Powerbanks are just SMPS that aren't connected to mains.
The battery inside the powerbank does not feed the D50.
It powers an SMPS (DCDC converter) that delivers the regulated 5V (regardless of the battery voltage)

true, but pb fixes that buzzing i mentioned

 

[solderdude]

Quite likely as there is no connection to mains so there can't be a groundloop through the PS of the DAC.

 

[Snafu]

Quite likely as there is no connection to mains so there can't be a groundloop through the PS of the DAC.

as i said buzzing is audible even when psu is plugged in but not connected to d50. it isn't ground loop.... just low quality psu

 

[Esp3n89]

as i said buzzing is audible even when psu is plugged in but not connected to d50. it isn't ground loop.... just low quality psu

Sounds like its just an inductor/choke that is vibrating. Iv got a few cheap chinese powerbricks that has this problem. As well as a bit from my GPU when fps goes crazy(loadscreen f.ex) . Some of the cheaper inductors/chokes are just glued together instead of having the plastic cast around it. The glue can hold a lot but eventually it will loose the fight against the magnet trying to vibrate inside the coil

 

[solderdude]

as i said buzzing is audible even when psu is plugged in but not connected to d50. it isn't ground loop.... just low quality psu

Ahh.. you mean an audible, mechanical hum / buzz coming from the wallwart itself and not from the speakers/headphone ?...
Just use another one I would suggest, saves you the trouble of regular recharging ?

 

[Snafu]

Ahh.. you mean an audible, mechanical hum / buzz coming from the wallwart itself and not from the speakers/headphone ?...

Noup, i mean buzz from speakers (very faint but audible). Nothing needs to be connected to this psu (it is itself quiet = no mechanical noise).
I know it sounds strange but that's how it is, psu feeds some kind of distortion back to wall (mains block) and possibly tube power amps pick it up from there. If i take power from another wall socket (not from same Supra MD06 where all amps are connected) and use extension cord = no buzz.

My point was that low quality psu (used for DAC) may cause audible distortion that can't be measured from DACout's.
Obviously it doesn't matter what device this kind of psu feeds, could be lamp nearby or something.

cheers

 

[sharpty]

Real audiophiles run their DACs from deep cycle lead acid batteries with linear regulators.

 

[Jimster480]

Well I have thought this for years.... it was even tested with SMSL M8/M8A. There was no benefit to linear power supplies.

 

[Hypnotoad]

Conclusions
It is clear that whether you use the USB power, or the supplied switching power supply, there is absolutely no audible improvement in the output of the DAC with linear power supplies. One can help himself believe otherwise by looking at the noise spectrum alone as I have shown in the last graph. But again, we don't listen to power supply wires. Those waveforms go through filtering stages even in cheapest DACs.

When I designed the AD797 Phono Stage I tried batteries, lantern batteries and SLA batteries, then I designed and made a linear power supply with LM317 voltage regulators running off a 12vac 1 amp wallwart, I couldn't tell the difference in sound quality or background noise. So it's no surprise that there isn't a difference between a switching and linear supply, thanks for confirming that.

 

[NTomokawa]

As a Topping D10 user, I don't worry about PS issues. :whistles:

But seriously though, thanks a ton for testing this out.

 

[solderdude]

Noup, i mean buzz from speakers (very faint but audible). Nothing needs to be connected to this psu (it is itself quiet = no mechanical noise).
I know it sounds strange but that's how it is, psu feeds some kind of distortion back to wall (mains block) and possibly tube power amps pick it up from there. If i take power from another wall socket (not from same Supra MD06 where all amps are connected) and use extension cord = no buzz.

My point was that low quality psu (used for DAC) may cause audible distortion that can't be measured from DACout's.
Obviously it doesn't matter what device this kind of psu feeds, could be lamp nearby or something.

cheers

Looks like the amp (or active speakers) you are using is susceptible to mains noises.
In that case you cannot measure anything coming from the DAC.

Some cheap wallwarts, just like cheap LED lighting etc. don't have enough mains filtering and perhaps won't pass proper conducted or radiated EMC tests.
Do you also hear 'ticks' when things like washing machines, lights or other devices are switched on or off ?

 

[Miska]

Interesting. With 96k sample rate the strongest images are pushed to 756k and 780k already. That made 705.6k upsampling redundant even for visual beauty.

Surprisingly high level though. And there is unacceptable junk also between. Since we know the first-stage ESS Sabre filter has -120 dB stop-band attenuation it's computations are thus accurate to about 20-bit resolution. But then there is one stronger intermediate image around 576 kHz (384 + 192) which really shouldn't be there. Overall there shouldn't those images around every multiple of 96 kHz in first place which you can now clearly see.

The way I check these is to run linear 0 - 22.05k sweep in a loop and keep spectrum analyzer in "peak hold" mode. That's the way for example the ADI-2 plot you reposted was made.

You can also see the noise humps between 400 and 500 kHz, such exist also elsewhere in ESS Sabre noise floor, also in audio band, and are sometimes a bit hard to catch because they wander around over time, so using a lot of averaging makes them harder to see and when you reduce averaging they are harder to see because noise floor comes up. But I have found some tricks how to catch those. They are probably somehow related to the ASRC cycle. Similar thing with ESS Sabre IMD measurements, 1 and 2 kHz peaks have their relative levels alternating over time, when one goes up the other one goes down.

 

[Snafu]

In that case you cannot measure anything coming from the DAC.

aaaannnddd..... that was my point exactly (well not 100%, but close enough to end this discussion here)

 

[bennetng]

Surprisingly high level though. And there is unacceptable junk also between. Since we know the first-stage ESS Sabre filter has -120 dB stop-band attenuation it's computations are thus accurate to about 20-bit resolution. But then there is one stronger intermediate image around 576 kHz (384 + 192) which really shouldn't be there. Overall there shouldn't those images around every multiple of 96 kHz in first place which you can now clearly see.

The way I check these is to run linear 0 - 22.05k sweep in a loop and keep spectrum analyzer in "peak hold" mode. That's the way for example the ADI-2 plot you reposted was made.

You can also see the noise humps between 400 and 500 kHz, such exist also elsewhere in ESS Sabre noise floor, also in audio band, and are sometimes a bit hard to catch because they wander around over time, so using a lot of averaging makes them harder to see and when you reduce averaging they are harder to see because noise floor comes up. But I have found some tricks how to catch those. They are probably somehow related to the ASRC cycle. Similar thing with ESS Sabre IMD measurements, 1 and 2 kHz peaks have their relative levels alternating over time, when one goes up the other one goes down.

Your graphs are from the RME right? Amir was measuring a Topping, so do you think the Topping and RME will show identical measurements when using HQP when measuring with APX555?

You also need to consider that the AP can look into the noise floor much deeper, and of course visually it will show more spikes.

Also why suddenly blame ESS for not improving the measurements in DSD, but on the other hand praise about HyperStream, just to refute @mansr 's argument?
https://www.audiosciencereview.com/...es-dsd-sound-better-than-pcm.5700/post-153837

All these differences are trivial and meaningless to me. Sure there are a lot of self-proclaimed golden ears all over the world, but I am not one of them. I have no interest in this DSD upsampling stuff at all.

To do something more practical, how about adding Wavpack support in HQP?

 

[Miska]

Your graphs are from the RME right? Amir was measuring a Topping, so do you think the Topping and RME will show identical measurements when using HQP when measuring with APX555?

Yes, but I didn't pull that graph in. However Topping seems to have the 768k images at higher level than RME. In my picture there wouldn't be any images below 768 because the digital filter used has image suppression to -240 dB.

You also need to consider that the AP can look into the noise floor much deeper, and of course visually it will show more spikes.

You could also look more into the noise floor, much deeper. But in this AP measurement FFT is 256k with 16 averages. My picture FFT is 65k and peak hold mode (so maximum peaks instead of averages).

Here you can see how the ES9038Q2M noise floor really looks like in Pro-Ject Pre Box S2 Digital:

Also why suddenly blame ESS for not improving the measurements in DSD, but on the other hand praise about HyperStream, just to refute @mansr 's argument?

I don't praise it because it is not so great, it has it's own kinks.

ES9038Q2M does improve measurements in DSD. Here's PCM input:

And with DSD512 input:

You can see that the harmonic series of 1 kHz is much shorter in DSD512 than in PCM (no matter if you use 44.1k or 768k). Overall reduction in THD.

When you bypass the on-chip digital filter, either by using 705.6/768k PCM, or DSD you also get rid of one set of jitter sidebands. So those are likely related to the on-chip DSP that runs digital filters.

44.1k input:

DSD512 input:

Point of my measurements with DACs is to find input format(s) that make the DAC perform it's best. However, nobody else here seems to be doing that, instead DACs are measured only with one format, not all the measurements extensively with all possible input formats.

All these differences are trivial and meaningless to me. Sure there are a lot of self-proclaimed golden ears all over the world, but I am not one of them. I have no interest in this DSD upsampling stuff at all.

To do something more practical, how about adding Wavpack support in HQP?

Maybe I will at some point. But why is WavPack more important than the DSP features? Nobody is selling music content in WavPack format. If you want just bit-perfect playback without any DSP from WavPack to your DAC, you can probably do that with any other software instead?

Point of HQPlayer is in it's DSP engines. For non-DSP playback cases there are plenty of other choices.

 

[bennetng]

Maybe I will at some point. But why is WavPack more important than the DSP features? Nobody is selling music content in WavPack format. If you want just bit-perfect playback without any DSP from WavPack to your DAC, you can probably do that with any other software instead?

Point of HQPlayer is in it's DSP engines. For non-DSP playback cases there are plenty of other choices.

Wavpack supports lossless DSD compression. Anyway, support it or not, your choice. But all these upsampling stuff, measurable or not, I can't hear any improvement, absolutely. So these kinds of processing are not for me. Thanks.