Ethernet Cables and Sound

[SDL]

Vee,

Thanks for sharing this video test of Ethernet switches. I watched most of it, but the audio was hard for me to judge using my MacBook with crappy speakers. The biggest takeaway seemed to be that power supplies matter a lot. The guys doing the listening seemed fairly concerned about how much the switches and power supplies cost, and their subjective opinions about sound quality followed the cost levels pretty closely.

The EE 8Switch (and its NuPrime twin) did ok for their cost as long as they had an upgraded power supply, but the Paul Pang Quad was definitely the winner overall. Al mentioned above that he has an 8Switch, and maybe he’ll peek in and let us know what he thinks of it.

The reviewers clearly like to use a fiber-optic tweak to provide additional noise isolation when necessary, but the audio with the fiber feed sounded pretty soft and there was general agreement that reproduction of piano was not very good using fiber.

Lots for me to digest as I think about my own system.

[aangen]

My English Electric 8 Switch is not an inspiration to me. I have never tried an upgrade power supply yet so before I pass final judgement I’ll try that. It’s fine, it’s just not a magic upgrade.

[Serhan]

Hi @aangen,
May be the rest of your audio chain is so good already!

[aangen]

I don’t know. Improving Ethernet has never seemed to result in magic for me. But then the last meter to my streamer is fiber optic cable. I ordered a couple of linear power supplies for my devices just now.

[Serhan]

The guy in this review of Melco S100 ethernet switch says he had to use a linear power supply! It seems to be the case with a raft of companies.
Melco S100 | The Ear (the-ear.net)
This said, I can justify the need for LPS with a piece of gear worth 250 bucks, but I am struggling to accept that when one has to pay thousands!

[rower30]

This chart is WRONG. The noise changes the square wave amplitude yes, but the “legal” high and low voltage IGNORES the noise on the signal and re-clocks a perfect one or zero at the end with ZERO noise. Exactly as digital is supposed to work.

As long as the voltage level isn’t too low or too high, it is in the “zone” to be a one or zero. This work really, really well, too. With Ethernet we have multi level signal voltages called PAM, Pulse Amplitude Modulation, encoding where the voltage steps are darn small yet we get BER of basically zero. 10G Ethernet uses an even higher voltage level PAM encoding and meets the same BER requirements as CAT5e with 10G verses 1G bandwidth.

How can this work if noise is a problem? The voltage level correction filters figure it all out and reconstruct the PAM levels perfectly over 100 meter channel lengths. Or longer with fiber opotics. We set the shannon’s law bandwidth to a needed ACR, attenuation to cross talk level,m to get the needed accuracy. And yes, noise is included in that, too. Better, the balanced nature of the signal CANCELS common mode noise. ONLY the differential noise remains and that is a small amount based on the cables unbalance (no cable is perfectly balances wire to wire in a twisted pair). But the unbalance, CUB, or capacitance unbalance, is also in the cards with a design. We know what it will do to the BER.

I’m fine with illustrations on stuff, but we need to make sure our basis of what we KNOW to be true stays that way and we don’t veer off into the made-up world of how we want something to work.

Best,
Galen

[alaint]

please listen, think later

[davidl]

Since posting my comparisons last January (post 435) in which I concluded that the SOtM cables were a good price/performance compromise, I’ve auditioned several more less expensive SOtM cables in my ‘HiFi’ LAN. Attached is a screenshot of my findings on sound quality which I’ve added to the spreadsheet. I ended up with a complete run of SOtM cables from the Wireless access point to the renderer / DAC. I also inserted a SOtM isolation transformer (ISO CAT6) immediately upstream of the renderer. The improvement wrought by each change was not major but the sum total improvement was very worthwhile, adding noticeably to my enjoyment of my music.

Of course as before, I should add the caveat that these results are for for my system, in my listening room, with my music and my ears.

[rower30]

Lots of weird stuff on CAT7 and 8 ISTP cable that isn’t really right. What’s it all REALLY about?

-) ISTP, Individually Shielded Twisted Pairs, have a braid outer shield and an outward facing foils on each pair that are grounded to the outer braid,
-) The shield provides about 100 dB shield isolation from ingress OR egress (shields work the same way either direction).
-) UTP design cables, that use 100% EM filed rejection, provide about - 65 dB RF isolation.
-) Standard UTP cables have a marked difference in the pair lays to set-up an “average” field cancellation between all pair combinations. None can be ideal or the others go way up (ninety degree wire cross directions).
-) ISTP cable DOES have to have varying lays as the LOW FREQUENCY coupling is less effective through the shields. This can be seen in a decides RISE in the coupling as frequency DROPS, and a braid/foil shield is less effective.
-) This is 4 MHz and below for ISTP shield effectiveness changes. No sweat, though, we add a set of four LONGER lays than needed on UTP to clean-up the low-end.
-) UTP’s market lay differences (shorter lays) cause a SKEW/Delay TIME difference between the pairs that limit ASIC circuit signal timing. This pair time limit from wire length difference has improved over the years.
-) ISTP COULD leverage the better TIME alignment between pairs at a set 1 GHz bandwidth, but it doesn’t do this as a primary design goal. It uses thew higher ACR, Attenuation to Noise Ratio, to improve the Shannon’s law bandwidth instead.
-) We gain about 40 dB of S/N with the shield and SOME advantages with the 22 AWG wires size (larger circumference improves high frequency attenuation). This isn’t as significant as the shield’s Shannon law contribution, though and it is expensive.
-) ISTP at shorter lengths and a lower bandwidth of 1 GHz and less is no more BER perfect the UTP.
-) Where ISTP works is at above 1 GHZ to10 Gig where the PAM 16 encoding (16 voltage levels) verses PAM 5 that requires low pair-to-pair cross talk and the individual shields mitigate that where a UTP, again, needs shorter and shorter lays to mitigate alien internal pair to pair NEXT. It can be done with UTP and with the right designs.
-) ISTP has, and has had, great bandwidth at a high initial cost but…it has been future proof all the way to 10G and more. Great for early adopters and long term fixed installations.
-) The BER is not an advantage UNLESS you run the system well to the end of the spec length values where ISTP will retain a bandwidth error margin over UTP.
-) Run less than 100 meter / 328 feet and the IEEE system error will be in spec with bit perfect transmission.
-) Would I use CAT& or 8 over good CAT5e even for 1 meter audio? No.
-) If your system is poorly grounded, you may make the system WORSE with shielded cable from the ground differential current noise coupling. And yes, that also impacts the internal noise rejection as now you Are INJECTING noise.
-) A shield ATTENUATES noise traveling through it. This requires ZERO shield current that offsets the transfer impedance (in Milli-ohm/meter) performance through the shield.
-) If you think using an antenna SPG, Single Point Ground fixes the poor ground resistance references you are now injects noise into the grounded shield end. Not at all what you want to do. All you do is add a ton of noise to one end or the other. Not good. FIX YOUR GROUND!
-) If you use shielded cable, a proper ground is mandatory.
-) CAT7 and 8 is about maximum Shannon’s law bandwidth at the same spec BER. We use the better noise / lower attenuation to up the bandwidth, and even if we did extend the bandwidth it is immaterial to short reach applications.
-) Short reach cables actually can have TOO GOOD a performance! Yep, the end reflections can be too larger if a cable has poor geometry control (higher RL, return Loss signal reflections) and impact link performance. 15 meter or less Ethernet permanent link channels do far better with true patch cable at the ends to ADD attenuation. The added attenuation damp signal reflections. We have PLENTY of signal level with short reach systems. What we need at the extreme is the opposite what we need at the shortest reach links. We have to concurrently manage, both requirements though.
-) ISTP cable is easily damaged and more fragile with foamed dielectric and the deformation of the shield adding RL resonance points. UTP is far more robust to abuse with solid dielectric and no shields and even BONDED pairs to lock-in critical center to center distances.
-) The closer a shield is to a pair, the capacities effect that sets impedance is algorithmically multiplied. ISTP is as bad as it can get! Any shield or insulation or wire dimensional changes are magnified greatly when or after the cable is made.
-) A 26 or even 28 AWG UTP patch uses attenuation to manage RL that will be higher with small AWG wire. The smaller wire and dielectric bump up against the limits of the extrusion precision and the wire draw process precision to keep minimizing variations as a percentage of the target values as stuff gets smaller and smaller. Small wire does not mean the machines variation gets smaller, too!
-) A bonded pair 24 AWG uses low RL (far better center to center stability) to manage reflections with lower RL.
-)Is CAT7/8 great cable? Yes it is if you use it for the bandwidth enhancement, no question. But for one meter audio?

The ability of you NIC and DA and AD circuit blocks to RF are what we hear, not the bit perfect data. You can’t hear that UNTIL we add ANALOG artifacts as distortion and in the time base mostly. Slight magnitude error can be there, too, but is hard to hear over time shifted distortion. What we hear isn’t the cable, it is the circuit design. Do we see RF at the DA or AD input stage? Does it change anything when removed we can measure? Is the circuit properly grounded at both ends?

There is some of what is really going on. I use a good bonded pair CAT6 UTP PATCH for the above reasons. Lots of stuff to explain a simple answer.

Best,
Galen Gareis

[EGH]

Galen, thanks for that detailed explanation; really helpful.

[markzaiser]

I agree with the basic premise, but it also can work quite well as a needed switch to isolate an audio network. In my case: Etherregen B > Ubiquiti 24 port switch> Optical > Router > Internet. Etherregen A > Optical > OpticalRendu, Eth1>NAS, Eth2>Sonictransport i9, Eth3 > Grace Digital Streamer, Eth4> N/C.

This isolates my audio rack (physically and galvanically) from the rest of my network. Video rack is optical from the 24 port switch to an Etherregen switch connected on the B side to Roku TV.

Just tossed some relatively fresh Tele 6922’s in the BHK Pre - they truly make it all new again. Well, it took away my baby, You know it aint comin back…

[gazjam]

There is some of what is really going on. I use a good bonded pair CAT6 UTP PATCH for the above reasons. Lots of stuff to explain a simple answer.

Best,
Galen Gareis

Great explanation Galen.

Do you use Cat6 or Cat6A?

thanks

[Somppsa]

Davidl found generic cat6 cables to sound the worse in his trials. I have no problem to believe his results and many have said the same.
Also it is always interesting to read well done technical aspects of the matter. But for some reason data and listening experiences don´t match with cat cables. Do we all have some sort of ground issues as Rower says in his article above because we happen to like Cat7 and 8 more than basic cat6 ? I really don´t know why but I surely can hear the differences between every cat cable I have tried.

Right now my current favorite is Inakustik Referenz cat7 because it sounds best. The only reason why I use it

[Vmax]

My results show the same in my system of Inakustik Cat 7 sound best. Galen’s data and post lays out the variables wire gauge. Insulation, spacing, grounds twist etc. i know every CAT cord doesn’t sound the same. One thing Inakustik addresses is vibration from acoustics. I would surmise that is not normally addressed in a lab testing a data bit perfect cable.

It should boil down to design optimization for usage environment in audio not just data transfer.

[Somppsa]

Let us know how you like it

[rower30]

I use a PS6+ patch cable. Not 10G. Curiously, 10G UTP has the exact same internal electrical limits as CAT6 except it is extended for 250 MHz to 500 MHz to meet 10G bandwidth reguiriements. The requirement that is tough, is the ANEXT external cable-to-cable cross talk spec that is layered onto all what the cable needs to meet.

1G Ethernet does not use PAM 16 and can’t take advantage of the extra cable spec requirements so there is no real need to get it. Streaming Music isn’t all that hard to do bit perfect.

If you want to get the highest Shannon’s law PAM5 type Ethernet cable bandwidth, use a 4.8 GB/sec rated cable that is WAY over the 5e 1.2 Gig/sec requirement for 1.0 G Ethernet. Curiously, you can make a better 1.0G Egthernet cable INTERNAL spec limit cable for 5e than 10G. 10G UTP shifts some of the INTERNAL performance to get the INGRESS EXTERNAL requirements up.

CAT7/8 use a shield(s) to gain external 10G requirements so internal are now isolated and don’t cross-over in physical deign parameters. But with no PAM16 requirements on the ASIC PC board’s, the tech isn’t used in any way at all. ISTP cable can uncover problems you didn’t know you had, like poor ground reference issues across your stuff. That’s not the cable’s fault you have more than one ground potential making ground loops all over.

Do we ALL have ground loops? Yes, since we add wire between all our stuff but the task is to keep the differences LOW with good DCR shields and Good AWG power cords that go to the wall plate and set the ground to that point back to the circuit box. This is why some report a large AWG power cable works better than a small AWG. It isn’t the “power” it delivers, but the lower ground differential.

A P20 or like can re -reference and consolidate your grounds at ONE closer spot. A big help.

Best,
Galen

[rower30]

When you test an Ethernet channel like an Industrial 2PR CAT5 with massive noise and EM all about that are far, far worse than the FCC limits for home use limits, we test ZERO BER! So yes, we test Ethernet every permanent link installed in real world apps and zero BER is the requirement to pass and this based on an internal cable measures like you get from Blue Jeans Cable with your assemblies.

If we want to play with the “sound” it isn’t the BER doing it. The Ethernet NIC cards ASIC isn’t remotely aware of what the cable is at all, it just chugs along as long as the signal is above the noise limits. And, those signal limits are -3dB and more BELOW the actual noise level! Ya, it works when the noise is LARGER than the signal. YIKES.

We aren’t even close to that Shannon’s law signal bandwidth with home networks.

BER isn’t the problem in audio but poorly designed DA and AD blocks that are non linear with noise. Each Ethernet cable, due to the internal lay design, can have varying spectral noise floors, but they ALL pass the 100 meter noise floor limits if they are legally tested Ethernet products (those pesky tests Blue Jeans supplies). The proper balanced circuits at the end of the chain are your problem if there is one. That ground FLOATS between the two wires, so RF is stripped off (it is common mode) and how Ethernet works to remove noise. It uncouples the GROUND so it can eliminate noise and that includes RF. -2V from +2V is 4 Volts. -1V from +5V is still +4 Volts. The added 1 volt of noise made no “difference” to the output of 4 Volts as an example.

Are both side of a balanced circuit “perfect”?. No, we have a CUB, Capacitance Unbalanced to Ground requirement needed for zero BER, so that’s all in the math too. And, CUB is length multiplied and we are so short that CUB isn’t even, or could be, a problem yet.

We can’t hear “bits”, but we can hear ANALOG time based EM wave changes, THAT is what our signal is after all. No change to the EM wave somehow and we don’t have a change. It is that simple to say. Now we need to find the CHANGE from set-up A to set-up B.

I use a PS6+ Ethernet with my PS Audio DS DAC to my UPnP/DLNA , and a 2 meter USB cable from the PC from my SD3100HV DAC. Neither has issues with noise or “black” backgrounds. The filters (firmware) do sound different on both DAC’s but that’s the DA conversion algorithm not the Ethernet or USB cable. The bit perfect data hasn’t changed.

I do use a P20 so my grounds are re-referenced to one point, with a stable 120V putput. Heavy BAV power cords to preserve the ground values near the same. My system is very “quiet” on one leg of the box and with grounds re-referenced through the P20.

I don’t suggest one thing to you and do another. I make changes ONLY where the data supports the effort and cost. I’m all for eliminating calculated or measured differences within a budget. The others? Not so much.

Best,
Galen

[Vmax]

Agree with P20 changing the game. Interesting in that it and better LPS implementation fed from common P20 ground also affected which Inakustik CAT 6 or CAT 7 I preferred. It sll seems to matter to affect noise floor and ability to image and create space and depth and CAT7 did it better after the P20. Whereas CAT 6 was preferable before Better power, single ground and lower impedance improvements.

[rower30]

Ethernet Ignores the noise floor by design. Only single ended analog sees and is modulated by, a noise floor. It isn’t the Ethernet cable doing this. Bit perfect is bit perfect. The AD and DA blocks are at fault for allowing RF to modulate the noise floor on the analog side. That said same noise has zero impact on Ethernet.

We need to prove that we have an RF modulated noise ground floor reference on the analog side to make this statement “true” and that different Ethernet cable allows the modulation amplitude to change in time. The noise floor modulation, if it is real, does change the EM voltage amplitude with time and what we hear.

All the talk about galvanic isolation is meaningless until we prove we have the problem, noise floor modulation from RF and changes with cable, exists to be “solved”. Second is if the changes do. indeed fix a real problem. There are two issues; define the problem, really fix the problem with a repeatable process.

Only real problems can be fixed, not ideas. We formulate a hypotheses to triangulate unknowns into knowns. The hypothesis isn’t the answer, it is a guess to trap the answer into a REAL and DEFINABLE attribute. It suggest where to look and how to meaure, not that that will provide the right answer. Usually we just eliminate another wrong one.

Best,
Galen

[Vmax]

@rower30

Galen, I have a inakustik excellence CAT6 cable that is only a meter. It has one plastic retaing clip broken but still functional electrically. I have compared it with BJC cables that cost ten times less and I absolutely like the Inakustik sound better.

I will happily mail it to you to trial listen to and compare, measure or whatever. The only thing I ask is you post what you hear and how it measured different and whether it passes your technical muster as a good data cable and whether or not it sounded better and or different. If so PM me or email me where to mail it. Just return it when you are done.