Layout of Ethernet Isolators

Yes, I know. I was talking about jrdmedford's SFP transceivers.

Aha, sorry that happens with side steps in threads! BTW what brand was the other ES9023 DAC besides the hifimediy?

The latter may be too simple as well. One can go to great lengths to engineer out all of the little tiny audible problems one can identify in a dac and still find the sound is flawed in some as yet unmeasured way. Ultimately then, they are all not properly designed; its just a matter of to what degree.

I think you would also have preference for either one in the diagram peufeu showed. These differences were not tiny. In the case of ES9023 the circuits are in general minimal so it possibly is a matter of spending a little more time to make it better. Peufeu's statement that "a lot of HF garbage in the output indicates the product was not properly designed, which will probably correlate with bad sound quality" is IMO a valid one. I have yet to see the opposite. No product I had on my bench deteriorated when trying to get down the numbers.
 
Last edited:
what brand was the other ES9023 DAC besides the hifimediy?

"home brewed" brand lol, secret sauce is ground plane + NFM21PS106 decap, 1µV noise MCLK regulator (50cents BOM cost qty1), shunt reg on AVCC supply (also 50cents BOM cost qty1).

proto.png

multitone2-44k-a=0.4.png
 
...No product I had on my bench deteriorated when trying to get down the numbers.

Okay. Were you measuring the numbers to get down the noise floor modulation? Just wondering if there might have been something you missed?
I ask because I suspect that sometimes one can reach a point where there is a trade off between minimizing noise modulation, and minimizing distortion by way of high loop gain...
 
The presence of HF noise in the output is not the direct cause of the bad sound.

Rather, it indicates that the layout and decoupling are trash, which means some critical nodes didn't get the care and love they need, which is the cause of the bad sound.

I've re-read my notes about the hifimediy ES9023 DAC. The bad layout had some amusing effects. For example, no ground plane and no shunt reg, therefore the current from the SPDIF optical output LED emitter flowing in the ground trace added a substantial amount of signal-correlated noise (ie, distortion) in the output, which got better by removing the LED. So the presence of the noise indicated a layout problem. Take-home point: shunt regs that make power supply current constant are a good thing. And if you make an optical SPDIF output, use a current-steering driver, so it draws constant current from the supply, and dumps constant current into ground, but it either sends this current through the LED or dumps it to GND on the LED's GND pin.

Another issue was no ground plane, so the analog reference decoupling capacitor's ground pin was referenced to a noisy ground, which made the reference noisy. Since the analog output is the product of the analog reference voltage with the DAC sample code value, this does not raise the noise floor when outputting a digital zero. And the usual way of measuring noise floor is with zero output, so this is not measured, and completely ignored. While playing signal though, the signal is multiplied by the analog reference, so you get noise modulation.

Since everyone measures noise floor with zero samples, this problem is usually missed, thus not addressed, thus it sounds bad.

Also, with zero values, the SPDIF output signal is not the same as with actual signal, so the LED issue above would manifest differently.

If you output a constant digital full scale value, or half-scale to make sure nothing clips, then the reference voltage noise is apparent in the output, and it can be easily measured. Then you can check for noise on the analog reference voltage simply by looking at the output, no need for probes, fix the problem if there is one, and then it sounds a lot better.

For example there was this guy with an ES9018 and the FFT showed it had some jitter, so he wasted his time optimizing the clock. I look at the FFT and tell him you have no jitter, you have 100Hz ripple on analog reference (AVCC) so it shows up in the output as a multiplication of two sine waves, f1+f2 and f1-f2, so you get two spikes spaced 100Hz from your test tone on the FFT. He fixed the ripple, and the "jitter" that wasn't jitter went away.
 
Last edited:
Noise on the output is something that can be measured by FFT even when too low level to easily see on a scope. The perceptual result in that case may not be 'bad sound' exactly; its perceptual effect might be masking of some low level musical details that should not be masked. In such a case, a listener may not be aware that anything is missing, so the sound may not be perceived as 'bad.'
 
Another example of measurement.

I was measured the output jitter of WM8805, by feeding WCLK into my soundcard and looking at the FFT. There was quite a lot, so I put some cotton in the soundcard, and it got better, but still not satisfactory.

So I slapped an oscillator on a bit of copper clad pcb with two AAs, taped some demakup cotton ball on top, feed the output of that and wolfson oscillator output to a XOR gate, acquire output with soundcard, look at FFT. It was pretty bad, unworthy of wolfson, who make pretty good chips.

Moving around the WM8805 crystal gave results. I drilled some holes, put it on the other side of the board to shield it, it got better. It was extremely sensitive to layout, as you would expect from a crystal. It appeared it would be tricky to get top performance.

So I grabbed 50MHz from the ES9023 oscillator, divide by 2 with a picogate, feed that to WM8805, adjust I2C registers so it knows what to do with it. Got much much better results, much lower jitter on WM8805 output.

All easily measurable with a soundcard and a hacked ghetto reference clock on a bit of pcb.
 
It is already off topic so ....Peufeu, could it be that you once measured a PSU I designed via a member called Phil? He seems to have vanished as he is not posting here anymore AFAIK.

I had a hard time decyphering technical french 😀 I ask it because I recognize the style of soldering. Maybe I asked before as it was years ago but I wonder how Phil is. He was quite active here and a Nice chap in communication.
 
Last edited:
Why not just use an SFP Optical MultiMode fiber modules... Just add a transceiver chip.

Fiber Ethernet is so cheap I use it in my house.... It lets me galvanically isolate the PoE switches that drive my outdoor cameras from the rest of the network - in case of lightning or sabotage LOL! Most of the time, the best use of Fiber is not to endpoint devices, it's use to connect distant switches, or outdoor APs, back to the core routing fabric - providing some level of galvanic isolation and security.

SFP Modules
– Ubiquiti Inc.

There are people with fiber converters in front of the streamer modded with 1000€ clocks and they use also an isolator between the converter and the streamer because there is an improvement by the isolater. It is really strange.

Just a summary of my experience with this forum:

I registered here because I thought that technical problems regarding audio could be discussed openly here.

My question related to how to properly design a product that is well established in the audio scene.

Here are just a few examples that an isolator is a common product for audio:


I have not touted anyone the isolators and said that this is a miracle weapon. i just wanted to discuss the correct implementation of this.

Since nobody has given any thought to this and apparently is not open minded enough and has only tried something like this for his "hammer looking for suitable nails" (which I have not criticized in any way), there is no serious discussion here.

Instead, apparently to distract from the lack of knowledge about the matter, only my audio setup, which I did not want to communicate here at all and was not in the focus, is ridiculed.

I hope this is not the general case in this forum.
 
Last edited:
"home brewed" brand lol, secret sauce is ground plane + NFM21PS106 decap, 1µV noise MCLK regulator (50cents BOM cost qty1), shunt reg on AVCC supply (also 50cents BOM cost qty1).

That is too much manual work for me 🙂.

I have used the Murata 3 terminal caps in an RF design, they are definitely good. Curious as to what benefit you saw here and if you've compared them to X2Y caps? Hard to get the best out of these without a really low inductance GND connection.
 
Peufeu, could it be that you once measured a PSU I designed via a member called Phil?

Oh, yes, I remember the audiophile power supplies he threw at me. Some were hilariously bad.

Was yours the one with the common mode filter? This one worked well. I probably have the test results archived somewhere, if you're interesred just ask, I will dig.

Maybe I asked before as it was years ago but I wonder how Phil is. He was quite active here and a Nice chap in communication.

I don't see him that often because he moved to a different city, but he's still alive.

No ground plane? Tell me this is a bad joke! 😱

Well, no, it's not a joke.

I still have an ONKYO TX-NR-905, a huge top of the line in its day AV receiver. The DAC board has 8 channels of nice PCMwhatever chips from burr brown. Two layers, no ground plane. I stuck scope probes in it, and I felt really sorry for these fine chips, there is so much noise everywhere, it's unbelievable. That's the default state of things.

I have used the Murata 3 terminal caps in an RF design, they are definitely good. Curious as to what benefit you saw here and if you've compared them to X2Y caps? Hard to get the best out of these without a really low inductance GND connection.

It's basically a 10µF feedthrough cap, it needs proper ground vias to ground plane on layer2 of 4 layer board to work (layout is very important)... if you look at it from the output, it acts as a low ESL [~300pH) decoupling cap:

NFM21PS (refl).png

But since it's also a feedthrough cap, it prevents noise from the device from going into the power supply, and also the other way around, much better than a normal cap. My network analyzer maxes out at 60MHz, and it's not enough to measure the ESL of this thing. It's very effective. The other NFM18PS feedthrough caps from Murata are much less effective.

NFM21PS (trans).png

I put the regulator on the other side, so it doesn't see the HF noise from ES9023.

X2Y is also nice, but they didn't have a 10µF cap, and I needed the capacitance for smooth transition to the bulk cap. Too little capacitance in the ceramic cap, and the ESL of the bulk cap would cause antiresonance problems, so I'd have to use a special snowflake expensive Low-ESL SMD tantalum polymer cap, which conflicted with the spirit of an ES9023 DAC, which can be summarized as "cheap".
 
Instead, apparently to distract from the lack of knowledge about the matter, only my audio setup, which I did not want to communicate here at all and was not in the focus, is ridiculed.

This is normal.

If you drive a car and you can feel every bump in the road being transmitted with high fidelity to your butt, you would probably think the suspension needs some work and go to the mechanic to get it fixed.

But audiophiles would call it "transparent"!

Likewise, if you have a badly designed piece of equipment that is very sensitive to some kinds of noise of EMI or jitter because the designers didn't bother to solve the associated problems, then this equipment will be temperamental and react positively or negatively, sometimes in a very strong way, to various noise mitigation measures. But this really points to a problem about said piece of equipment being a special snowflake.

What audiophiles call "transparent" really means the device is sensitive to things it shouldn't be sensitive to if it was properly designed.

The hobby that consists of making such devices perform is known, in engineering circles, as "turd polishing".

Do not get caught in this trap! It is frustrating and very expensive. Please preserve your bank account and sanity.

I am interested in finding out why your device is so sensitive to an ethernet isolator, while it should not, because ethernet is already isolated, so it is a bit of a paradox that adding an extra isolator improves things.

But the reason I am interested is simply... to understand whatever design mistakes are in your device, so I can avoid them, simple as that. Pure self-interest on my part, I confess.

> I have not touted anyone the isolators and said that this is a miracle weapon. i just wanted to discuss the correct implementation of this.

You will not get an answer here, because it depends on the ways your device is sensitive to noise, which we know nothing about.

With your device on a bench, and proper test equipment, then perhaps an answer could be found. And it would most likely be completely unrelated to what you or anyone else thinks the problem is, because this is how these things work. Test, measure, find something unexpected, investigate, solve problem. No-one can do that on your device over the internet.
 
This is normal.

If you drive a car and you can feel every bump in the road being transmitted with high fidelity to your butt, you would probably think the suspension needs some work and go to the mechanic to get it fixed.

But audiophiles would call it "transparent"!

Likewise, if you have a badly designed piece of equipment that is very sensitive to some kinds of noise of EMI or jitter because the designers didn't bother to solve the associated problems, then this equipment will be temperamental and react positively or negatively, sometimes in a very strong way, to various noise mitigation measures. But this really points to a problem about said piece of equipment being a special snowflake.

What audiophiles call "transparent" really means the device is sensitive to things it shouldn't be sensitive to if it was properly designed.

The hobby that consists of making such devices perform is known, in engineering circles, as "turd polishing".

Do not get caught in this trap! It is frustrating and very expensive. Please preserve your bank account and sanity.

I am interested in finding out why your device is so sensitive to an ethernet isolator, while it should not, because ethernet is already isolated, so it is a bit of a paradox that adding an extra isolator improves things.

But the reason I am interested is simply... to understand whatever design mistakes are in your device, so I can avoid them, simple as that. Pure self-interest on my part, I confess.

> I have not touted anyone the isolators and said that this is a miracle weapon. i just wanted to discuss the correct implementation of this.

You will not get an answer here, because it depends on the ways your device is sensitive to noise, which we know nothing about.

With your device on a bench, and proper test equipment, then perhaps an answer could be found. And it would most likely be completely unrelated to what you or anyone else thinks the problem is, because this is how these things work. Test, measure, find something unexpected, investigate, solve problem. No-one can do that on your device over the internet.

There are many posts in other forums, that isolators improve sound also in setups 10 times more expensive then mine and with dacs which are beyond doub (but think my RME is also well engineered). Also when the network is completely isolated with fiber converters. I would like to understand why this could happen. I'm sure this has nothing to do with my setup, it's a general phenomenon. It is hard to understand, without heard the effect.

It is the same at my LS50W and also at my SBT/RME Headphone Setup.
 
Last edited:
Basically, I believe you will find conducted and radiated EMI/RFI everywhere. If it sneaks inside a piece of audio equipment, it can be demodulated in semiconductor junctions as noise modulated in with the audio signal. With steady state RF it can produce a DC opamp output offset. However, real world EMI/RFI is not necessarily a steady-state sine wave, so the offset is not necessarily constant. The explanation you are looking for is probably something like: When you have audio going through a transistor while it is demodulating RF (thus shifting its bias point), that effectively can intermodulate the audio to produce with what looks like noise on an FFT (recall that a single diode can be used as a simple modulator).

Next question might be why is it audible? According to ESS they claim to have done unpublished research showing that some humans can hear very low levels of noise modulation, that people can be trained to hear it, and that the ear is exquisitely sensitive to it.

If interested:

https://www.ti.com/lit/an/sboa128a/sboa128a.pdf?ts=1619334845094

https://www.ti.com/lit/an/snoa497b/snoa497b.pdf
 
Last edited:
Oh, yes, I remember the audiophile power supplies he threw at me. Some were hilariously bad.

Was yours the one with the common mode filter? This one worked well. I probably have the test results archived somewhere, if you're interesred just ask, I will dig.

I don't see him that often because he moved to a different city, but he's still alive.


Yes mine was with the common mode coils just as I (mis)use them in other designs. If I recall correctly you noticed what I did with them. I did this design with Subbu as a Group Buy here to make people enthusiastic about LPS with the then not so standard 3A capability and we shipped quite a few of them but almost got no feedback which I suppose is a good thing. I still use these for ethernet switches. Can you dig it up?

Glad to hear Phil is still alive.

@ TomJones: I already gave a link where the Bob Smith thingie is debunked as being BS. Pun intended. I think you're in a Sackgasse and you have something else going on that might explain the drastic change when inserting a completely redundant isolator in an already isolated situation. Maybe the wave forms are so hampered that you have managed to make ethernet behaving not as it should with cables not being as they should be in a setup that is like it should not be 🙂 Voodoo if you ask me. AFAIK you don't need termination in the situation you have created, concluding from the schematics I have dug up. Maybe I am wrong but you have 2 RJ45 plugs both with "floating" cable pairs already terminated at both sides....

In post #110 you mention "My question related to how to properly design a product that is well established in the audio scene.". Well is is not well established in the audio scene but at the sales side things go AOK. It has nothing to do with ethernet design and or general practice amongst network professionals as these use equipment to measure connections and have to deliver these reports to very demanding customers (think of broadcast companies and high availability networks). You complain about being ridiculed and that is a pity but please understand that so many things are non standard not many will acknowledge it to be a good ethernet setup and some might even go so far to call it something different than ethernet. It is a self created situation that somehow works and to your satisfaction. That is positive! Simply enjoy it and forget about ridiculed but don't expect technicians and designers to give serious answers.
 
Last edited:
Regarding my initial question, I recently came to the conclusion that you should make a Bob Smith termination on both sides of an isolator, since a cable is connected on both sides.

The article JP posted seems pretty good. They recommend a resistor value different from Bob Smith and back that up with believable measurements. So, yeah.

You could measure it, if you have a scope.

But you could also stick a big ferrite core on the cable. That has different common mode impedance characteristics than the ethernet magnetics.