Investigation Into Effects Of PC load On DAC Analogue Output

[March Audio]

Prompted by discussion in another thread I decided to take a look at the effects of PC load on a DAC analogue output.  The assertion is that PC load (CPU/Disk etc) increases noise in the DAC output. The assertion doesnt specify what type of noise or how this noise manifests, so I will be looking for any differences in a DAC output between an unloaded PC at idle (baseline activity), and when at 100% CPU load with disk read/write activity.  This is going to be purely a measurement based exercise and not using subjective listening.

 

PC used is one I recently built myself.  It uses standard off the shelf components, nothing "specialised" for audio.

 

DAC and measurement ADC is a RME ADI2 pro FSR BE.  One of the best technically performing DACs / ADCs around.  Used in loopback.

 

 

First test is to simply use an FFT to look at the noise and distortion in an unloaded situation.  PC load was as you see in the picture above.  The first FFT is a low resolution 8k length.  The input signal is from Roon playing a 1kHz tone at -60dBFS.  I have kept the signal low as to not confuse the inherent DAC and ADC distortion with the noise I am searching for.  Averaging is also used to smooth out the noise floor.  This will make any very low level consistent tones or spuria stand out.  Note this measurement will not show any very short term blips or noises.

 

 

So even with a low resolution FFT we can see down to about -150dB blow the DACs maximum output level.  To put that into context, if a DACs max output level is 2 volts RMS (typical) we can see signals that are as low as 0.06uV!

 

However we can do even better.  If we increase the FFT resolution we split the frequency band into smaller chunks.  Each chunk contains less energy (noise) so the apparent noise floor drops and we can see even smaller signals.

 

This is a 4M point FFT of the same signal.

 

 

The apparent noise floor has dropped to around -180dB.  This has exposed that the DAC and ADC do actually have some odd harmonic distortion components that were previously hidden below the noise floor.  BTW -180dB on a 2 volt output DAC is about 0.002uV or 0.00000000002 volts.

 

So what happens if I load the PC.  To do this I used dBpower amp to perform flac to AAC conversion.  This maxes out the cpu and also read/writes to the disk at the same time and increases memory usage.

 

 

 

No significant difference in the noise floor and spuria between idle and full CPU load.

 

 

As mentioned the above test will not see transient noises, for that we need to look at the waveform.  To do this I will use @pkane2001 Deltawave utility which can examine the waveform (and other parameters) for differences.

 

As deltawave will "see" transient noise we need to consider that any two measurements will have a degree of random noise. So to show what you might expect I performed 2 measurements with the PC at idle and ran them through Deltawave.

 

We see the same odd harmonics in the spectra, although due to the lower spectral resolution we have a higher apparent noise floor.

 

 

This shows the spectrum of the difference between the two measurements.  No spuria stand out, its just the noise floor. -160dB.

 

 

This shows the difference of the spectra.  Just random noise.

 

 

This shows the differences in the waveform.  The RMS differences are -120.24dB, which is essentially the performance limit of the DAC and ADC.

 

 

 

Below are the same plots except this time the reference is the unloaded measurement against the loaded measurement.

 

 

 

 

 

So no significant difference between the PC unloaded and loaded.  The RMS difference was this time -120.17dB

 

 

Final test is to look at jitter.  Jitter will appear as sidebands on an FFT, with close in phase noise appearing as a widening of the base of the fundamental tone.

 

Below is a 12kHz test signal with the PC unloaded.  The RME has excellent jitter levels.  We can see the only sidebands are at +- 50Hz from the fundamental which indicate the tiniest amount of mains noise is getting into the dac clock/dac.  The level is -174dBFS.  Obviously insignificant.

 

 

This is with the PC loaded.

 

 

 

So to conclude, with this PC, load makes no significant measurable difference to this DACs performance.

 

Please note that these tests do not account for other potential issues that may be encountered such as ground loops in single ended RCA systems.

 

[March Audio]

38 minutes ago, SoundAndMotion said:

Bold added.

 

Thanks for your efforts.

 

So now I can conclude that some computers with some DACs do show measurable activity-related noise (I'll find links, if anyone is interested), but not all (as you've shown)!

Thats why I phrased it the way I did 😀 .

 

Problems can occur when noise currents flow to the dac via the USB shield wire.  Its a variation on the classic ground loop.  Its also why SMPS leakage currents can end up causing hum and higher harmonics. Its also a problem that can be exacerbated by single ended RCA connections.  These currents end up flowing in the RCA interconnect shield, which is signal low, creating potentially audible and easily measurable noise.  RCA is a really bad idea WRT signal integrity.  However its always been solved in my experience by proper galvanic isolation.  You just break that link. 

 

However to be clear this isnt an issue of "the PC activity" and load causing noise as I am discussing above.  Its a connection issue.  PCs have noisy ground planes.  That noise doesnt stop because you have disabled a few processes.

[March Audio]

1 minute ago, SoundAndMotion said:

It seems to be activity related. Sure, ground loop, but still activity related ground currents.

No its not. Its a problem created by a problematic way of connecting equipment together.

[March Audio]

Just for info here is some info on galvanic isolation from Mathias Carstens who is the founder of RME.

 

Hello Mr. Stämmler,

the isolator arrived this morning. I tested it right away:

- It’s fully compatible with all RME USB interfaces, regardless of whether they’re working in interrupt or isochronous modes. The test was carried out with the maximum channel count. Also, I've flashed the device with a new firmware via USB. No problem at all.

- The power supplied to the device end is surprisingly high. Our bus-powered interfaces Babyface and Babyface Pro overload the isolator anyway, so you’ll have to work with it there on an external power supply. This was to be expected, because at 550 and 800 mA we’re drawing considerably more power than originally planned of course.

- The isolator does exactly what I had hoped it would: it eliminates all problems arising via USB during unbalanced test setup.

I’ve attached two screenshots for you. Two USB audio interfaces were connected to a notebook here (one wasn’t from RME). On one device the analysis software emitted a test signal (1 kHz sine wave), on the other device it was analyzed at the input. The analog connection Out to In is unbalanced. Once with the isolator, once without.

I’ll share this information and the images with you. So if you want to show somewhere that RME has found the isolator to be fully compatible, or want to use the measurements on the website or as visual aids in your presentations, then you’re welcome to do so free of charge.

Best regards,

Matthias Carstens

RME
 

 

[March Audio]

19 minutes ago, SoundAndMotion said:

I guess you didn't read it. Same connection, different software made the problem disappear.

First search delivered pages of hits which I am not going to search thru.  Second produced nothing. So I couldnt.  This is not a software issue.

[March Audio]

9 minutes ago, SoundAndMotion said:

I guess search doesn't work on your computer.

 

https://www.audio"science"review/forum/index.php?threads/computer-activity-can-impact-dac-performance.22/

Sorry but just looked at that and see nothing relevant to PC activity (ie load) causing a problem.  Its a broken dac design, do you see the jitter issues over SPDIF?

 

We also have no idea how Amir set that test up, direct sound, wasapi etc etc.

 

[March Audio]

7 minutes ago, SoundAndMotion said:

The OP shows it's software related. 

 

If you hadn't read it, why say with such certainty that it's not software?

No it doesnt.  It shows he got different results with 2 different software which could be down to any number of reasons.  What Amir states:

 

Its brochure brags about "isolation" but the word "galvanic" is not there. This means that it is not isolated from the PC ground and that is feeding into the oscillator causing those correlated jitter spikes.

 

So Amir specifically state its a connection issue as I describe above.

[March Audio]

14 minutes ago, SoundAndMotion said:

You've missed the point. I already mentioned that the ISO Regen improved the results, so galvanic isolation can block the noise. But changing the software, also changed the noise! So it appears the noise is generated by the computer activity and it travels along the connection. It's activity AND connection, not OR.

 

And you were still willing to state your opinion (in Objective-Fi?) without investigation of the other measurements.

And we don't know all the details of Amir's test. Good point. Do we know all those details from your post? (Direct sound, wasapi etc.?)

I havent missed any point and it wasnt just opinion..  I looked at what Amir wrote and it confirmed what I said.  Amir said it was due to exactly the same reasons I did.

 

The fundamental problem is not PC activity.  A PC is "active" all the time.  The problem is specifically due to the lack of isolation in the USB connection.

 

ASIO from Roon.  Measured using REW via ASIO.  No I didnt use an isolator on the USB connection to the DAC.  Anything else you would like to know?  Happy to fill in  any blanks 😀

 

Also I will perform the same tests with a laptop tomorrow.

[March Audio]

12 minutes ago, SoundAndMotion said:

What Amir and you say is less important than what's in the data. The noise changed when he changed software. What is the source of the noise? Grounding problems allow noise into places it shouldn't be, but what is the source of the noise?

You need not continue hunting down a single problem, and then announcing success, when clearly multiple problems exist. Activity AND connection. Not OR.

Its been explained. You are going round in circles.  Your semantics are irrelevant.  *ALL* pcs are noisy - period.  Thats just how they are.

 

The problem in that case is a non isolated USB connection.

[March Audio]

18 minutes ago, SoundAndMotion said:

As I understand it, the raisin d'être for this thread is given here:

and here:

 

And I believe you have begun to convincingly demonstrate that those who "hear differences" where you believe they can't, likely won't... not with your computer and your software/configuration choices and your hardware/DAC.

But your goal cannot be achieved with your single setup. Indeed, trying multiple setups, one at a time is very inefficient. IMHO, you can achieve your goal much more efficiently by convincing those who claim to hear a difference to replicate your effort above, or a similar measurement-oriented approach. Many will beg off, but if you convince only one or two, you'll be further along toward your goal than testing one setup, then another, then another... 

All this assumes I understand your goal.

 

Did I not say I was going to try another laptop tomorrow?  I also have an additional laptop and an additional PC to try.  I also have several DACs to try.  I never claimed this was anything more than an investigation, never claimed it was conclusive proof of a certain conclusion which you are making assumptions about.

[March Audio]

15 minutes ago, idiot_savant said:

Well, this is going downhill fast...

 

There is an interesting question here : To what degree is "sensitivity" a good thing? 

On the one hand, you want a low noise floor and low distortion, good filtering etc in a DAC and an ADC to reveal as much as possible.

You also want this system to be as immune as possible from external influences ( mains, EMC, etc ).

 

And yet - I often see it stated that a product is "sensitive" because it reveals some difference in cables, processing or suchlike that can be simply explained by being marginal engineering. For example, I could transmit a high speed clock between two devices ( think a transport to a DAC via I2S ), but I accidentally fail to terminate it properly. The cable is now very important to the performance because of the mismatched ends, to the extent you can probably make it crackle by bending the cable. The DAC is now very "sensitive" to the cable, so you can "improve" it by cable tweaks. However, the *best* it can ever be is the *same" as the same design properly implemented ( i.e. properly terminated ).

 

Which would we prefer?

 

@March Audio has been fairly clear that he is demonstrating a system that is based on off the shelf components, with no tweakery, using components he regards as well engineered, and can see no differences. This *isn't* to say he won't ever find any differences as he refines his measurements, but so far there is nothing to see.

 

Comparing this system to one where the reviewer explictly says it is poorly engineered and not galvanically isolated and exposing a problem merely shows the problems you can get with marginal engineering. 

 

As boring as it may sound, good engineering will remove a lot of the "randomness" that people seem to enjoy tweaking around. I don't know about anybody else, but I'd rather listen to a system that sounds the same irrespective of the phase of the moon,

 

your friendly neighbourhood idiot

Well it will get moderated if people insist on going round in pointless circular arguments 😏

 

You make a very good point.  Its *not* a virtue for a DAC to be sensitive to what a PC is doing. Quite the opposite.  As is shown here its quite feasible to engineer a DAC that isnt sensitive.  Although this particular DAC is expensive, it doesnt have to be expensive to do so.

[March Audio]

This is just a start.  No reason to refrain from taking it into controlled listening tests.

 

I read your posts.  It is you who is not listening.  I dont agree with you idiot_savant doesnt agree with you, we have said why.  Repeating yourself wont change that.

 

Yes repeating the tests with different PCs and DACs may not uncover anything untoward, but the whole point of investigating is to see if we can.  If nothing untoward is discovered then it still tells us useful information - that the problem isnt as wide spread as some seem to think.

 

I have no intention of suppressing views.  You have had your say.  What I wont allow however is a sensible discussion to turn into some of the stupid s**t fests we have seen recently with circular pointless arguments that dilute and distract from the point of the thread.

 

 

[March Audio]

More data.  Same PC and ADC.  This time a Gustard X16 DAC driven from Roon with WASAPI through XLR outputs.

 

No PC load.  Note a very small amount of mains pick up at 50Hz, 150Hz and 250Hz.  From the weight this DAC has a linear psu internally with a transformer.  Probably from that.

 

 

With PC loaded.

 

No significant difference.

 

I will perform the Deltawave check on this DAC tomorrow.  I can also try the DACs RCA single ended output.

 

[March Audio]

There is greater potential to see some issues with single ended RCA dacs, but they will show problems with the PC unloaded or loaded.  It's not quite as simple as "broken" dac designs.

[March Audio]

6 hours ago, SoundAndMotion said:

The whole thing? If you mean the whole 2 page thread, I did. And other DAC threads too. Did you? Amir made assumptions: he didn't locate the noise source; he read the brochure which said "isolated", but not "galvanically isolated", so he assumed it was defective. It may be, but I don't accept that type of assumption without evidence.

 

There is a common technique of stating that all "competent" or "properly-designed" DACs behave [the way you say - whatever you want], with the convenient response to any counterexample as being "broken" or "defective". You're guaranteed to always be right, since you made the definitions.
There was a $1000 bet on another site (not the recent one on ASR), where the person proposing the bet would allow the I-can-hear-a-difference claimant to pick any 2 "competent" DACs. He warned of DACs that were deliberately or unintentionally "defective". He said pick any 2 that had been tested with SINADs over "say... 100dB". I noticed that of the many DACs that Amir had tested, the median was about 100. So by the challenger's admittedly random cutoff, nearly half the DACs tested were defective. You didn't make that bet, but I wonder how useful a term like "defective" is if it applies to too many examples. The rule-of-thumb that's intended to use it may be weak, maybe a rule-of-pinkie.

 

The fact is that a "defective" DAC may function well with a low-noise PC, as will a "competent" DAC with a noisy, but within spec, PC. What about PCs that are nearly in spec, but not quite (noise on USB lines slightly too high)?

 

@March Audio has shown that his PC with (now 2) DACs performs certain tests well. Great, and I appreciate his effort. My concern is that if the goal is to convince members who "hear a difference" in whatever (don't think the motivating threads were simply about DACs), or to convince them they have defective hardware, this method may not work. And there are better, more efficient, methods. @firedog do you think this can convince the target group? I don't think convincing people who are already convinced was the goal. I do not wish to dissuade @March Audio from continuing... so I'll probably continue with the issue of convincing skeptical people in another thread.

The goal is to objectively investigate and show what's actually going on as opposed to just guessing.  You wont convince believers with entrenched views of anything, I have no illusions about that.  However there are still plenty of people our there who will be interested in the investigation.

[March Audio]

6 hours ago, PeterSt said:

 

Since I can't measure myself the differences which can easily be heard by all those who are open to it, including myself, I don't care much.

All those who are not open to this and (think they can) measure, obviously won't see any "significant" difference.

And the definition of "significant' is indeed ...

You haven't actually demonstrated using controlled techniques that people can hear a difference.

 

Later I intend to perform controlled listening tests.

 

 

[March Audio]

3 hours ago, semente said:

Since these measurements don’t show the effects of “PC load”, which other measurements could be performed?

Peter has already suggested multi-tone IMD. Anything else?

Multitone is no problem to try.  Will do later.  However I can tell you now that it won't make any difference.  IMD is directly related to THD.  If the PC load is having no effect on thd (as already seen), it won't effect IMD.

 

I'm happy to field other suggestions for tests.

[March Audio]

40 minutes ago, fas42 said:

Turns out that it's quite easy to see differences in the waveform, where it counts, which is the sound waves in the room -  @manisandher recorded what was happening when he altered settings of Peter's player; which had clearly audible variation, and this was quite distinct in the recordings when examined with DeltaWave. We were promised a more thorough version of this exercise, to overcome those who went into a mad thrash about microphone technique and everything else they could think of - but this has not eventuated.

 

Interesting how all investigations into things like this either bog down, or die a slow death - an instinctive reluctance not to rock the boat too much, perhaps ... ?

As discussed elsewhere recording with a microphone in a room is hopelessly insensitive and will pick up all sorts of irrelevant random background noise.

 

With all respect to Mani, unfortuntely Its a hopelessly flawed method.  Of course you will see differences in the waveform.  However these differences will have nothing to do with the PC or audio system.

[March Audio]

6 hours ago, PeterSt said:

 

Since I can't measure myself the differences which can easily be heard by all those who are open to it, including myself, I don't care much.

All those who are not open to this and (think they can) measure, obviously won't see any "significant" difference.

And the definition of "significant' is indeed ...

I'm open, however I am looking for evidence. Otherwise it is just faith.

 

Significant?  Well my definition of insignificant is when you can't see any changes at -180dBFS.

[March Audio]

6 hours ago, idiot_savant said:

@SoundAndMotion there seems to be a bit of crossed purposes here. I myself have said on a forum, you’re unlikely to “convert” people who are entrenched, but if you let some of the more outrageous claims go unchallenged, people who are maybe unconvinced need a bit of a sensible voice to help a bit of balance?

I think the point @March Audio is trying to make *isn’t* that all PC’s with all DACs are perfect, but that with a bit of care, you don’t need a PC encased in unobtanium to get really good performance. 
 

As for specs, most well-engineered stuff tends to have a margin of error built in, and good engineering in one place *can* indicate overall good engineering, whereas bad engineering in one area tends to indicate a trend

 

your friendly neighbourhood idiot 

This is the point.

[March Audio]

4 minutes ago, fas42 said:

 

First of all, the background noise can be made to be relatively consistent - to test this, merely record the same playback, with no changes, several times to establish the consistency of the recording space.

 

Secondly, what you are looking are patterns in the waveform which consistently register as being different at certain points in the music - say, a treble crescendo, or transient. These are the "tells" that one's ears are sensitive to, and which are meaningful.

No it can't Frank.  Also before you argue, you have no basis to judge.  You have never analysed any acoustic noise so have no idea about it.

 

Background noise in your listening room will probably be around 30dB on a quiet day and quite variable.  This will mask subtle differences.

 

The tests I have performed are looking far, far below the threshold of audibility.

[March Audio]

So this is a 32 point multitone 1/10 decade.

 

RME ADI2.  PC unloaded.

 

 

PC loaded

 

 

No significant difference.

[March Audio]

29 minutes ago, fas42 said:

 

The real world is that the electronics which retrieves the source music file may impact the analogue part of the chain, by their operation - vast tracts of conversation have been laid down, in people's pursuit of finding answers to this. That particular  player just happens to make it easy to play with settings which the author has found to be effective in altering the electrical patterns which can disrupt, in subtle ways, the following components - for fussy listeners.

 

So, either all these people are delusional about hearing differences - or it may be that not all electronics are designed and implemented to behave perfectly, when faced with various forms of electrical noise ... now, which do you think is more likely?

Frank.  This is an objective thread where you need to provide more than your usual nebulous conjecture spam.

 

You have your own thread where you can think out loud and speculate as much as you like, but its off topic here.

[March Audio]

11 minutes ago, PeterSt said:

 

Maybe you or others have better ideas, but I don't think that watching the multitones themselves would bring anything (well, as we saw). So the - or my idea would be that the multitones would be the stimulus and that with other means other data is observed. Thus for example, give it those tones (the more the better) and observe possible jitter influence. I did such things myself with normal music

 

 

vs

 

 

... when jitter was high enough so I could still do these things (today it is way under the measurable level for me).

This is only about the volume of music and how a higher level (2nd shot) implies more random jitter than low level or even idle (1st).

 

But see ? while this is uncontrollable music, multitones would be controllable (repeated tests with various environmental parameters like you attempt here).

 

There is no way this is really simple, because it will require measuring clock data so (analog) system noise is avoided. And, people may see me write for the xth time that by now 3 years or so ago, I had applied measuring wires to certain in-DAC chips so I could measure USB influence (this originated in the original Lush thread), but the analyser I bought for the purpose is still unpacked (yes, I am serious).  Those wires still stick out in there. Read: so much prep time this takes, including the learning what actually to look at - how to ground the probes and so much more.

And so I will repeat:  the effort you put into this, is much appreciated and I would hope something comes from it. But it won't go easy IMHO.

 

 

 

Can you explain what "other data"?  Happy to look into things

 

Measuring jitter does not require measuring clock data.  In fact by trying to you will probably introduce additional jitter. You just need to look at the dac output signal.  Thats the only thing we are interested in and if there is jitter in the clock it will be quite apparent there.  The clock timing variations directly translate into sidebands on the test tone.  Close in phase noise appears more like a widening of the tone base.

 

[March Audio]

Oh just to add to the above, more multitones wont affect the jitter levels.  just because your audio signal is more complex it doesnt mean that the dac word clock is affected.