Is Digital Audio Transmission Analog? [video]

[antcollinet]

Haven't said anything about that this has something to do with sound differences.

If one would deal a little bit more intensively with the matter, then one would know that apart from the differential signal also common mode interferences are transmitted via Ethernet. The galvanic isolation is only effective in the very low frequency range.
I have already made many measurements here and know what I am talking about. These common mode interferences get into the ground system of the receiver and are also passed on by the streamer via the USB ground to the DAC.

It is not only zeros and ones that are transmitted.

Do you have any data or measurements that show how much of this noise is able to bypass the galvanic isolation? Audio frequencies are all very low compared to RF interference.

Plus - again and again, in measurements, we see that none of this noise finds its way onto the analogue output of the DAC. They are pretty damn good at ignoring it - if it is there at all.

 

[VonGoethe]

FFT 0-125MHz
Blue FFT is USB Ground without Ethernet Link.
Pink with Ethernet Link.
White is difference, so the noise added by Ethernet.

And here in the audio frequencies:
FFT 0-12.5kHz

 

[antcollinet]

View attachment 248014
FFT 0-125MHz
Blue FFT is USB Ground without Ethernet Link.
Pink with Ethernet Link.
White is difference, so the noise added by Ethernet.

And here in the audio frequencies:
FFT 0-12.5kHz
View attachment 248015

Nice - What are the horizontal and vertical scales on those plots - I can't determine it from the pictures.


Also - are you sure that is noise transferred from ethernet, and not noise added to the ground by the activity of the receive device processing the incoming data?

 

[VonGoethe]

Horizontal scale: 12.5MHz for the first plot, 1.25kHz vor the second.
Vertical scale for both: 10dB

For validation here a plot with a self constructed cable and Ethernet with filter effect.
No, these interferences don't come Frome the streamer (Stack Audio Link II) itself.

 

[antcollinet]

I'm really impressed that you already have the measuremments to answer my questions. This is really interesting.

Sorry to be a pain - for the vertical scale, dB relative to what. I'm interested in the absolute level of the noise in volts.

 

[Ken Tajalli]

Haven't said anything about that this has something to do with sound differences.

If one would deal a little bit more intensively with the matter, then one would know that apart from the differential signal also common mode interferences are transmitted via Ethernet. The galvanic isolation is only effective in the very low frequency range.
I have already made many measurements here and know what I am talking about. These common mode interferences get into the ground system of the receiver and are also passed on by the streamer via the USB ground to the DAC.

It is not only zeros and ones that are transmitted.

From where does this noise originate from?
- From the transmitter ?
- picked up by cable?
- how does common mode noise affect differential data lines?
Thanx

 

[Ra1zel]

I have already made many measurements here and know what I am talking about. These common mode interferences get into the ground system of the receiver and are also passed on by the streamer via the USB ground to the DAC.

It is not only zeros and ones that are transmitted.

Something tells me your transducers, the stuff that actually does sound, are not very concerned by that

 

[antcollinet]

I'm really impressed that you already have the measuremments to answer my questions. This is really interesting.

Sorry to be a pain - for the vertical scale, dB relative to what. I'm interested in the absolute level of the noise in volts.

Oh yes, also: Is this with shielded or unshielded ethernet?

 

[VonGoethe]

I'm really impressed that you already have the measuremments to answer my questions. This is really interesting.

Sorry to be a pain - for the vertical scale, dB relative to what. I'm interested in the absolute level of the noise in volts.

The interferences added by Ethernet can be over 100mV.

I've been researching this for a while. I hear differences myself with different switches and cables, but I can't accept all the explanations, like Ethernet jitter, for this so far.
If you are interested, here is a Google Translate link to some of my measurements and findings:

Messungen von Ethernet-Infrastruktur - Switches (Nur Lesen) - open-end-music-professional

Eric@Ohrwurm (https://www.open-end-music.com/member/6904-ohrwurm) wird hier seine Untersuchungen zum Netzwerkteil einer audiophilen Streamer-Kette vorstellen. Inhaltverzeichnis: 1. Meßaufbau und -equipment...

www-open--end--music-com.translate.goog

The common mode noise is produced by the sender switch and asymmetries in the signal path, which causes mode conversion of electrical power from differential signal to common mode.

The reason this ends in the grounding system is the Bob Smith termination, which is common standard in ethernet devices.

 

[voodooless]

To what reference are these things measured? One end goes to ground? Where does the other end go?

 

[VonGoethe]

To what reference are these things measured? One end goes to ground? Where does the other end go?

One end to USB ground an the other to measurments ground plane, solid aluminum plate 100x50cm that is grounded.
streamer is ungrounded and isolated on the aluminum plate.
I make no claim that the measurements are correct in absolute terms, but in relation to each other.

 

[voodooless]

One end to USB ground an the other to measurments ground plane, solid aluminum plate 100x50cm that is grounded.
streamer is ungrounded and isolated on the aluminum plate.
I make no claim that the measurements are correct in absolute terms, but in relation to each other.

Well, you're measuring potential differences between grounds.

More interesting would be to actually look at what comes out of the DAC when playing silence (or a test tone) in both cases. I bet you won't see much...

 

[VonGoethe]

Well, you're measuring potential differences between grounds.

You are right - this is what EMI test measurements are doing

 

[antcollinet]

One end to USB ground an the other to measurments ground plane, solid aluminum plate 100x50cm that is grounded.
streamer is ungrounded and isolated on the aluminum plate.
I make no claim that the measurements are correct in absolute terms, but in relation to each other.

Ah - OK, I see a problem then. Have to go out now but will describe the problem as I see it (fwiw) later.


EDIT - not so much a problem as an explanation as to why this is not relevant for the audio performance of the streamer or downstream components.
Disclaimer - I've made some assumptions about your test setup which might invalidate the discussion below if they are wrong.

I've I understand correctly - you are measuring the difference between noise on the streamer ground and on a separate ground plane - the two have no reference to each other (you say the streamer is isolated from the measurement plane). I'm guessing (assumption 1) you are grounding the external plane to mains earth - you might as well say that the earth plane has the noise on it (it is shared with all the power devices in your building). What for example would you measure on that plane if you referenced it to a copper rod hammered into the ground.

Whatever noise in the streamer relative to that plane is irrelevant for the audio performance.


Looking in more detail - you have a common mode noise source (referenced to the earth plane), on an Ethernet cable, and are connecting that to the streamer. The only coupling mechanism (assuming (2) unscreened Ethernet cables) is via the parasitic capacitance in the Ethernet transformer. This is probably of the order of 6-7pF so is going to be very high impedance at audio frequencies.

Does quick calculation....

That is just over 1Mohm at 20kHz between the noise source, and the streamer ground plane. Impedance is higher at lower frequencies.
(note - with that level of coupling impedance - it is quite possible that your scope probes are influencing the measurement.)

But the streamer has no current path to the earth plane - so no current through the impedance - so the full noise voltage will be added to the streamer ground plane.

I think in your measurements you are getting around 400mV of noise on the streamer ground plane (relative to the earth plane).

For the audio system this is is of no relevance whatsoever. What matters is voltage differences between different parts of the audio circuit. So lets look at that.

Either the ground plane of the streamer will be earthed by connection to an earthed component downstream (EG DAC, preamp, amp)
Or - they will all be non earthed.

If they are are non earthed, then the non earthed grounds of al the devices will be connected together, there will be no ground current, and all grounds will be at the same potential - no impact on audio.

If any of the components are earthed, then there will now be a low impedance path back to your measurement earth plane via the interconnect. Lets say that path is as high as 1 ohm at 20KHz.

So now you have a 400mV noise source, a 1Mohm or higher noise source impedance, and 1ohm to your earth plane. So now you are going to have 400mv*1/1000000 = 400nV on the ground plane. This is not going to be an issue even if the full noise voltage is applied to the analogue output of the DAC into the amplifier (it is at -130dB compared with a 2V signal)

 

[VonGoethe]

Hi tonycollinet,

you might be right about your grounding explanation.
But also think about demodulation of HF.

The whole problem with this is that I think this is the only valid explanation of how different CAT cables or switches can affect the sound - as mentioned above, I accept that and can here differences myself, but I'm not a Snake Oil fanboy . I rule out Ethernet jitter, although there is a theory that jitter is noticeable in the receiver as noise in the low frequency range.

 

[antcollinet]

Hi tonycollinet,

you might be right about your grounding explanation.
But also think about demodulation of HF.

The whole problem with this is that I think this is the only valid explanation of how different CAT cables or switches can affect the sound - as mentioned above, I accept that and can here differences myself, but I'm not a Snake Oil fanboy . I rule out Ethernet jitter, although there is a theory that jitter is noticeable in the receiver as noise in the low frequency range.

There is one other explanation - and it is the most likely - that what you hear is due to cognitive biases of one sort or another. None of us are immune to this - none of us.

The problem I see (if you haven't already done controlled tests to eliminate that) is that you've done a hell of a lot of work investigating why you hear a difference without confirming that what you hear is actually there in the output of your audio kit to be heard.

Jitter in ethernet is obviously irrelevant, since timing of the music is not in any way influenced by Ethernet clocks - any more than file transfers over Ethernet are.

 

[VonGoethe]

I've done blind tests which confirmed my hearing.
Easiest way to test differences of cables is to test two identical unshielded cable, but with one you have to strip the outer cable jacket and remove the for 100baseTx unused blue and brown pair.
With this test, my faith in the irrelevance of the Ethernet infrastructure regarding sound has wavered.

 

[Ken Tajalli]

There is one other explanation - and it is the most likely - that what you hear is due to cognitive biases of one sort or another. None of us are immune to this - none of us.

The problem I see (if you haven't already done controlled tests to eliminate that) is that you've done a hell of a lot of work investigating why you hear a difference without confirming that what you hear is actually there in the output of your audio kit to be heard.

Jitter in ethernet is obviously irrelevant, since timing of the music is not in any way influenced by Ethernet clocks - any more than file transfers over Ethernet are.

If you check the review of Hugo2 and review of Audioquest Pearl USB cable, in which the very same Hugo2 was used, you can see the effect of RFI on USB input of Hugo2 .
On one review, some RFI had leaked in (Hugo2 review), and when it didn't (Audioquest review).
Even though the RFI was most probably in way out of audio band region, yet, Hugo2 developed some inband noise and distortion .
This is due to :
- Hugo2 being sensitive to such interference .
- Hugo2 has no isolation on its USB input.
- Hugo2 is single ended by design.
Hugo2 's analogue section plays up when infected by RFI .
Such issues are documented.
How audible they are, that's another story.
On my Hugo2 , the effect is a slight bright shimmering that spices the sound.
I must emphasize , that this is not cable related, but rather a ground loop issue.

 

[KSTR]

I've done blind tests which confirmed my hearing.
Easiest way to test differences of cables is to test two identical unshielded cable, but with one you have to strip the outer cable jacket and remove the for 100baseTx unused blue and brown pair.
With this test, my faith in the irrelevance of the Ethernet infrastructure regarding sound has wavered.

Hi Eric,
Assuming that this is true and repeatable -- I have no doubts that you executed proper blind testing avoiding hidden false cues -- I think it offers an IMHO very exiting and unique opportunity to "nail things down", recording and analyzing the analog signal disturbance differences that must have been present to trigger the verifiable listening impressions.

The idea goes like this:

1. We insert a transparent real-time AD-->DA stage right in front of your final headphone or speaker amplifier and tap off the digitized signal via USB for live recording. The assumption here that the signal degradation happens before this final amp (or the AD input), no significant further demodulation etc. in their input stage.
2. Transparency and that baseline assumption must be confirmed in a listening test (btw, I tend to trust sighted listening impressions from experienced listeners with a blind-testing background, so short-cuts might be possible here). If not, game over ;-) Ideally, that buffer AD-->DA stage is syncable to the clock of the actual playback DAC clock, this is possible in many cases.
3. The next step would be to confirm (within statistical bounds) with blinded listening tests that the recordings, when played back later, are still discernible, on preferably different DACs and playback systems but at least with your system as is (just that it's not real-time source playback).
4. Once we have that, we could use test signals besides music tracks and record them as well so we have a database for analysis which should show difference especially in waveform subtractive analysis (I do have some experience in this). This would allow to isolate the characteristics of the degradition, how the root cause (like GND pollution with RF that demudulates or finds other ways to sneak into the analog signal) manifests its effect in the signal. An candidate we might expect here is more or less signal-correlated noise artifacts.

I'd gladly volunteer for a collaboration (like covering the hardware and setup of that AD--(tap-off)-->DA buffer, test signal etc) in such an effort... what do you think?

 

[tuga]

Well, you're measuring potential differences between grounds.

More interesting would be to actually look at what comes out of the DAC when playing silence (or a test tone) in both cases. I bet you won't see much...

And why not seeing what comes out of DAC when playing music?