Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: 10/100/1000 "4578" power

Roger and all,

What I am saying is that whatever ill effects (not just noise, but rf-type
things such as parasitic phenomena such as resonances, impedance matching,
and the like) of attaching power components to the signal pairs are
intrinsically isolated from the signaling pairs by using the other pairs
for power. 

This is a basic consequence of the guaranteed properties of CAT-5 cable.

The amount of isolation guaranteed as a minimum is per the numbers quoted
in my little drama.

Even with things like RJ-45 connectors in the link, the isolation that is
obtained is at least 15 dB. You can confirm this by putting some CAT-5
cable on a network analyzer. (Maybe we will have to do that for the

What this means is that whatever dirties of ANY kind your power attachment
does to the wires you use (4-5,7-8) is isolated from the signal wires
(1-2,3-6) by at least 15 dB over the 2-100 MHz frequency range. This is a
factor of more than 5:1 and represents a risk reduction of that amount.

15 dB covers quite a lot of sins!

In the early meetings of the DTE Committee this was discussed and deemed to
be a worthwile thing.

I am not saying that it is impossible to successfully attach power to the
signal wires, OK?

But what I am saying is that if you want to take the riskier path and throw
away this 15 dB cushion, you will have to show that you did not mess up the
system, at least if you want my vote.

***** Slight topic shift *****

The present UTP universe is defined by the clauses in IEEE802.3 and ANSI
TP-PMD that I cited in the list I put on the IEEE Web site. This is a list
of tests that we use to evaluate any new technology variant. 

In particular, we (our company) have insisted that PHY vendors go through
this hoop. A subset of this same list of tests is what the UNH IOL test
suites uses for qualifying new PHY designs.

The reason for doing this is that it is possible (even probable) for a link
to still operate even though it did not meet these criteria; i.e., a plain
old functional test or even general scope pictures do not give any idea of
the margins (if any) present. If you have not been doing these tests on
your own designs and have been getting away with it, then you can thank
others who did do these characterizations so that by simply buying and
hooking up "approved" parts you got a compliant design.

Now we are going into another dark corner of the PMD ("new technology"), so
we have to go back through the whole list of parameters to make sure that
we did not inadvertently tank something.

There is a definite proof that this is so: some of the test methods sitting
in full force and effect in IEEE 802.3 will not work with powered signal
wires, and if this is adopted the existing standards will have to be
overhauled in some areas.

As an analogy, this is a little bit like the Gigabit fiber committee
(802.3z) discovering far, far into the standards process that the paucity
of modes in laser outputs caused very severe effects on some kinds of
multi-mode fiber. This was nothing new, and in fact it was in my
20-year-old optics handbooks, with pictures even. But it had gotten
overlooked. The result was that it delayed the standard by some 6 months to
a year.

So, to repeat:

You wanna power the signal pairs? Then you gotta show that this does not
goof up anything of importance, and in fact, the standard will have to
enumerate what you have to do to achieve this.



At 09:50 AM 5/15/00 -0700, R karam wrote:
> Hi Larry
>Please clarify the statement below.
>>I hope that it was shown that sending power over the signal pairs raises
>>the technical difficulty bar by the amount of isolation between pairs. (15
>>dB at least)
>are you saying that because power is on the used signal pairs, I get
>possibly up to 15db of noise injected on there, what about the supply
>filter that insures DC- Pure DC on the wire, I am not clear on this?