Ethernet Cables - which are the most important?

[r_w]

Because they focus on the signal path and assume not much else is of such importance... and of course most of them are incorrect in those assumptions - as we are finding out for them.

 

 

 

If GE/Siemons/Teledyne/ etc don't require a device like the Baaske, it's because they have taken greater care to the physical implementation of their network connection.

 

Which leads me to the obvious point: Why aren't manufacturers of supposed high end audio devices doing the same?

[jabbr]

Just sayin'

 

Fair enough. I use medical as an example because that is an area where actual measurement and real engineering has been done. Regarding audio, having personally experienced problems with ground loops, I tend to think that many cable effects:issues are due to the presence of loops and leakage currents -- and that the proper solution lies in eliminating the leakages rather than buying $$$ pieces of wire. But that's just my opinion.

[plissken]

Fair enough. I use medical as an example because that is an area where actual measurement and real engineering has been done. Regarding audio, having personally experienced problems with ground loops, I tend to think that many cable effects:issues are due to the presence of loops and leakage currents -- and that the proper solution lies in eliminating the leakages rather than buying $$$ pieces of wire. But that's just my opinion.

 

I believe a lot of these issues go away with balanced systems. Star grounding, using the same wall outlet, or best yet going CMNR power.

 

I've seen people spend thousands of dollars on single ended systems, to spend just as much on single ended cabling in attemp to banish gremlins, when they could have taken a structured approach and engineered a solution.

 

Too many audiophiles think that you can't engineer good sound. That you have to play esoteric component roulette to obtain it.

[JohnSwenson]

Too much stuff to quote individual posts.

 

First off, Ethernet PHYs

...

Well I guess that is about it for now.

 

John S.

 

Here is some more on Ethernet PHYs and cable interactions.

 

One topic I did not cover in the first post is package, this can make a big difference. The package encapsulates the die and provides electrical connections that can be soldered onto a PC board. The "pads" on a die are way too small to be soldered onto normal PC boards and frequently made from a material that won't solder well.

 

As with everything else there are trade-offs here. Inexpensive packages have lots of inductance between the package pins and the die pads, on both the signal and power connections. This causes significant degradation in both signal integrity (SI) AND power integrity (PI). Expensive packages have vastly lower inductance for signals and have internal planes (just like on a PC board) for the power and ground. The have vastly lower SI and PI degradation, BUT they cost a LOT more. A cheap one might cost 12 cents and a really good one might cost 12 dollars. It is kind of hard to use a $12 package when the whole chip is earmarked to sell for $1.79

 

Here is a free tip, if you are a streamer or network DAC designer, go for your Ethernet chip in a BGA package. It will usually have vastly lower SI/PI degradation VS TQFP or QFN packages.

 

With a cheap package you can actually wind up with most of the SI/PI degradation happening in the last mm. There have been arguments along the lines "I'm only using 10 feet of Ethernet cable, how can that possibly put the PHY into a mode where it has to work much harder, 10 feet couldn't possibly add that much SI degradation". That is true, but this argument ignores the significant degradation that can be happening in the package. That alone can put the PHY past its "cruising along" mode into a working harder mode no matter what cable you use.

 

I worked on one program where the customer wanted very high performance but simply could not afford a good package, so they had to add multiple regulators on the die and live with the extra power consumption. Other customers have gone with higher cost packages so they don't have to deal with the higher complexity on the die. There are lots of different trade-offs being done.

 

This preoccupation with low power has produced chips which actually turn off most of the circuitry between packets, so every time a packet comes along the chip power draw may go from 2mA to 50mA and back when the packet is over. THIS can wreak havoc with power supply noise and in particular ground plane noise. Because this is happening at packet frequency not symbol frequency normal board decoupling schemes simply do not work at dealing with this.

 

One thing I hope you take away from this is that PHYs are very different from each other, something that might significantly affect one PHY may not affect another. Thus attempts to make things sound better in YOUR system may not work on someone elses or actually make it sound worse.

 

How cables interact with this can have interesting aspects. The chunks of hardware in the PHY are designed to deal with different types of distortion to the signal. Thus you can have the situation where one type of distortion causes a chunk of circuitry to be turned on that has significant affect on PG noise. Thus if a cable does something to decrease that particular distortion that chunk of circuitry may not be turned on. BUT frequently what it takes to decrease that distortion increases something else, but since the circuitry to deal with THAT distortion causes less noise, it is a win for that cable -- with THAT PHY. On a different PHY, it may make things worse.

 

I think this is happening in "audiophile" network cables. They try a bunch of different things that worked for analog cables or maybe S/PDIF cables and by accident wind up lowering one type of distortion that helps with a particular PHY, but winds up increasing a different type of distortion. Then they start throwing money at it (silver wire, special connectors etc etc.) which don't actually do much. So you wind up with a very expensive cable which does give some improvement with some equipment that uses the same PHY as they used in their tests. But on other systems it doesn't improve anything or makes it worse.

 

The best way to do it is to improve ALL aspects, not just one. This can be done but is expensive, but not in the way that I have seen from the audiophile cable makers.

 

John S.

[jabbr]

Now I know why I like fiberoptic Ethernet Of course we can swap out PHY between Finisar and Avago and SR vs LR at least for the Intel NICs. I suspect that the LR SFP+ modules use the best components.

[One and a half]

Ethernet packet rate causes problems with power consumption with Ethernet, then, as much as USB does. With Ethernet, the base transmission can be slowed, but will that cause problems with the PHY?

 

Either way USB or Ethernet needs more boxes to clean up, de-jitter, fix, isolate, at least the conditions are known, but will DAC or streamer manufacturers implement these measures? Not without a cook book though, cause fixes are not cheap to solve when you're staring at R&D $'s.

[jabbr]

With fiber, power is relatively independent of speed -- at 10G the difference is dramatic to the extent that the world is really a better place with fiber -- fiberoptic is much cleaner than copper all around

 

 

Sent from my iPhone using Tapatalk

[One and a half]

Ahh, with Fibre, there's the leakage current issues of the Fibre-Copper adapter power supplies. Oh well more LPS-1 or MINI to sell.

[jabbr]

I go direct either to motherboard or with PCI-e card. If you must, you can battery supply the FMC

 

 

Sent from my iPhone using Computer Audiophile

[scan80269]

John, even the pulse transformers sitting between an RJ-45 connector and the PHY chip would affect the SI of the Ethernet signal as seen by the PHY, correct? Would you say the selection of Ethernet pulse transformers is as critical as the PHY package type in terms of SI?

 

I suppose the better pulse transformers should have low inter-winding capacitance, much like for power isolation transformers, along with low stray inductance?

[mansr]

Here is some more on Ethernet PHYs and cable interactions.

 

One topic I did not cover in the first post is package, this can make a big difference. The package encapsulates the die and provides electrical connections that can be soldered onto a PC board. The "pads" on a die are way too small to be soldered onto normal PC boards and frequently made from a material that won't solder well.

 

As with everything else there are trade-offs here. Inexpensive packages have lots of inductance between the package pins and the die pads, on both the signal and power connections. This causes significant degradation in both signal integrity (SI) AND power integrity (PI). Expensive packages have vastly lower inductance for signals and have internal planes (just like on a PC board) for the power and ground. The have vastly lower SI and PI degradation, BUT they cost a LOT more. A cheap one might cost 12 cents and a really good one might cost 12 dollars. It is kind of hard to use a $12 package when the whole chip is earmarked to sell for $1.79

 

Here is a free tip, if you are a streamer or network DAC designer, go for your Ethernet chip in a BGA package. It will usually have vastly lower SI/PI degradation VS TQFP or QFN packages.

 

With a cheap package you can actually wind up with most of the SI/PI degradation happening in the last mm. There have been arguments along the lines "I'm only using 10 feet of Ethernet cable, how can that possibly put the PHY into a mode where it has to work much harder, 10 feet couldn't possibly add that much SI degradation". That is true, but this argument ignores the significant degradation that can be happening in the package. That alone can put the PHY past its "cruising along" mode into a working harder mode no matter what cable you use.

 

I worked on one program where the customer wanted very high performance but simply could not afford a good package, so they had to add multiple regulators on the die and live with the extra power consumption. Other customers have gone with higher cost packages so they don't have to deal with the higher complexity on the die. There are lots of different trade-offs being done.

 

This preoccupation with low power has produced chips which actually turn off most of the circuitry between packets, so every time a packet comes along the chip power draw may go from 2mA to 50mA and back when the packet is over. THIS can wreak havoc with power supply noise and in particular ground plane noise. Because this is happening at packet frequency not symbol frequency normal board decoupling schemes simply do not work at dealing with this.

 

One thing I hope you take away from this is that PHYs are very different from each other, something that might significantly affect one PHY may not affect another. Thus attempts to make things sound better in YOUR system may not work on someone elses or actually make it sound worse.

 

How cables interact with this can have interesting aspects. The chunks of hardware in the PHY are designed to deal with different types of distortion to the signal. Thus you can have the situation where one type of distortion causes a chunk of circuitry to be turned on that has significant affect on PG noise. Thus if a cable does something to decrease that particular distortion that chunk of circuitry may not be turned on. BUT frequently what it takes to decrease that distortion increases something else, but since the circuitry to deal with THAT distortion causes less noise, it is a win for that cable -- with THAT PHY. On a different PHY, it may make things worse.

 

I think this is happening in "audiophile" network cables. They try a bunch of different things that worked for analog cables or maybe S/PDIF cables and by accident wind up lowering one type of distortion that helps with a particular PHY, but winds up increasing a different type of distortion. Then they start throwing money at it (silver wire, special connectors etc etc.) which don't actually do much. So you wind up with a very expensive cable which does give some improvement with some equipment that uses the same PHY as they used in their tests. But on other systems it doesn't improve anything or makes it worse.

 

The best way to do it is to improve ALL aspects, not just one. This can be done but is expensive, but not in the way that I have seen from the audiophile cable makers.

 

John S.

It seems to me that rather than chasing after super-high signal quality at the input, it would be more effective to isolate the PHY from the rest of the circuit by means of a dedicated regulator and generous application of decoupling caps.

 

Also, would you agree that using silver instead of copper in otherwise identical cables will make no difference?

 

Finally, any opinion on directional Ethernet cables?

[Jud]

But John, if what you say were true though not well understood, we'd just have both "sides" arguing with each other forever with neither being corr... - oh, wait....

 

 

Sent from my iPhone using Computer Audiophile

[plissken]

But John, if what you say were true though not well understood, we'd just have both "sides" arguing with each other forever with neither being corr... - oh, wait....

 

 

Sent from my iPhone using Computer Audiophile

 

There seems to be a large supposition that this power modulation may make it all the through the ground plane of a mother board and past noise suppression measures taken by the DAC. It looks to be 50ma at worst.

 

It this is the case then, as I've pointed out before, and much like your mono testing you are a proponent of due to echoic memory, any one here can spin up a song on their streamer or computer, and provided enough RAM applied to buffer can start the song and then plug/unplug the cable and evaluate.

 

The issue that John brings up only happens in spurts. If you have a streamer with just plebian 100Mbit connectivity and 64MB of cache we are talking 5 seconds to fill the buffer. After that the connection SHOULD go idle (because streamer designers are implementing best practices, right?).

 

If it's GBe we are talking sub 1 second.

 

It has to be taken in it's totality.

[The Computer Audiophile]

There seems to be a large supposition that this power modulation may make it all the through the ground plane of a mother board and past noise suppression measures taken by the DAC. It looks to be 50ma at worst.

 

It this is the case then, as I've pointed out before, and much like your mono testing you are a proponent of due to echoic memory, any one here can spin up a song on their streamer or computer, and provided enough RAM applied to buffer can start the song and then plug/unplug the cable and evaluate.

 

The issue that John brings up only happens in spurts. If you have a streamer with just plebian 100Mbit connectivity and 64MB of cache we are talking 5 seconds to fill the buffer. After that the connection SHOULD go idle (because streamer designers are implementing best practices, right?)

 

I know you won't like this, but I have a great friend who has been in the industry for 40 years and is on the cutting edge with digital audio. He always unplugs Ethernet cabled when possible because he doesn't like the sound of his system when the cable is plugged in. He also did some testing by putting a scope on an AES wire. He saw noise on the AES cable when Ethernet was connected. Remove the Ethernet cable and the noise disappeared.

 

I know it's not a longitudinal study repeated by third parties, but this anecdotal evidence is evidence nonetheless.

[plissken]

I know you won't like this, but I have a great friend who has been in the industry for 40 years and is on the cutting edge with digital audio. He always unplugs Ethernet cabled when possible because he doesn't like the sound of his system when the cable is plugged in. He also did some testing by putting a scope on an AES wire. He saw noise on the AES cable when Ethernet was connected. Remove the Ethernet cable and the noise disappeared.

 

I know it's not a longitudinal study repeated by third parties, but this anecdotal evidence is evidence nonetheless.

 

I don't hate it at all. You'll never find me taking a position that copper Ethernet is the best connection.

 

Noise suppression should be on the DAC and Ethernet alike in audio applications. I wonder how this applies to Ravenna implementations?

 

Why do all the high end players still implement 8P8C connectivity? Why aren't they natively going over to optical?

 

How does this affect HDBase T?

 

Is the noise due to the cabled connection to another piece of equipment or is it packet demodulation on the receiver regardless of copper, fiber, wireless?

[plissken]

Here is a thought: make a 328 foot cable and get a 6 footer and see if people can hear a difference. One cable will have the Switch port and NIC at operating limits. The other not so much.

 

Setup a LAG on the Switch and team the adapters on the client workstation. Swap cables out with out interruption of play with this configuration.

[The Computer Audiophile]

Here is a thought: make a 328 foot cable and get a 6 footer and see if people can hear a difference. One cable will have the Switch port and NIC at operating limits. The other not so much.

 

Setup a LAG on the Switch and team the adapters on the client workstation. Swap cables out with out interruption of play with this configuration.

 

That'd be interesting.

[plissken]

That'd be interesting.

 

What may be more interesting is to stack the deck: A lot of top tier PHY's can exceed the 328 foot limit. I would say birth a 360 foot cable and see if the increase in BER has any effect.