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

Re: Working with Larry!

Hello, Roger,

Your reply has brought up a few question for clarification. Please see my
comments/questions below:

At 11:56 AM 5/10/00 -0700, R karam wrote:
>Hi Larry
>  Thank you for asking- I am encouraged.
>Cisco's proposal has the following A, B sections.
>A- Use the Signal pair 1,2 and 3,6 to deliver power on a new switch.
>B- Use unused pair (4,5 and 7,8 ) to deliver power for mid-span.
We have seen your press releases and are somewhat confused. Your Web page
(4/24 post) talks about a Catalyst 6000 Family Inline Power Field Upgrade
Module. This sounds to me like a mid-span powering unit. However, under
Specifications/Inline Power Specifications it says Pin Assignment 1,2,3,6.
In another place pins 4,5 and 7,8 are mentioned. Just curious. Will your
Ethernet Phones support being powered either way?

>Powering from the center tap is a classic approach, All right, we are all
>about the phys doing bad things to us, how many phys can we test and under
>what conditions, please note that I can have a phy on my board now that has
>300ps of transmit jitter and can receive up to 130m over cat-5 with the far
>end transmitter doing 1.2ns peak-peak jitter on the far end.  quite a few
>phy vendors have gotten good at this. 

>Crosstalk and noise are another issue, well, we filter the power supply
>(we sleep better doing so for EMI reasons...)
>And some phys have good crosstalk performance- No Spec was present/
> This does not guarantee you that
>every phy will be happy, then again did you ever think that this would be
the case
>without power?
>To this, I did present the MLT3 waveform with and without power via the
>tap, and overlaid the results on the scope.  We did not see any difference.
>it would be nice to measure rise time, and every other spec.  but I don't
>think at this point it is worth my time when my slides showed no change
>in the signal for up to 250ma or so.  Magnetic vendors claim 300ma....
Well, actually, here is an area where I think that just using a normal
signal is inadequate. I don't seem to be alone in this, beause IEEE 802.3d
(MAU Conformance Test Suite, on your CD) was deemed necessary to test for
proper implementation, and it was for this reason that I put in in my List
Of Requirements (Have You Looked).

To put it in a nutshell, these tests look at the ability of a an Ethernet
link to deal with bursts of out-of-band interference, minimum-level signals
and the like. Having magnetics operating near saturation may affect the
ability of the system to cope with such things. 

(And, along this line I might point out that your favorite PHY vendor
flunked some of these tests due to an improper implementation of the
receiver input filter specified in IEEE 802.3. This was not a show-stopper,
but shows that these tests do a pretty good job of fully characterizing a

When you are running 100TX you have 20 mA signals, but up to 8 mA of
allowed DC unbalance. 10BASE-T is DC balanced, but you have 50 mA signals.
So the two schemes sort of help each other out in the magnetics department,
but is that still true with a big DC bias present? (I don't know-- but this
is the kind of thing that will have to be sorted out, as well as the prior
Ethernet clauses that deal with such things.)
>    I did spice  the Nortel bandpass approach and the series resistance of
the CAT-X cable
>    seems to lower the Q of the DTE  BP filter no data was presented on
your scheme
>    either- at least none that I have seen.  Could you please fill me in
on how this is done?
I think you are talking about an earlier Level One variation on the B/C
scheme. We are not looking at common-mode tuned circuits, precisely because
of the de-Q'ing you mention. Does this happen with your differential scheme?
>2- Powering over center taps, is a classic approach (exclude the 100BT)
>    that are worrying a lot of people- and where we trust our magnetic
vendors to
>    do a good job and a good characterization on these components, along with
>    the one we have done... 
>3- a general statement here,
>    I am not sure that I owe the group every detail and my schematics at
this point
>    when we are going in with the attitude that only midspan or unused
pairs is the way
>    to go,  leaving the burden of the proof on me alone....
Well, please yourself. We intend to go through the full characterization,
if only to uncover any flaws in what we are presenting. As to trusting the
magnetics vendors, we have no quibble about what they do characterize, but
I am always amazed at how different models of magnetics from the same or
different vendors --all supposed to be alike, and in fact they look alike
in a passive test fixture-- vary in performance on Real Boards. Considering
how good they are for the price and looking at how they are made I am also
amazed that they are so consistent unit to unit. Anyway, we don't take this
area for granted on any of our designs.

>please see a reply to your questions below,
>>I have a few questions to start with:
>>1. In the slideware, we always see a 4-pole relay at the DTE equipment end
>>used to switch out the discovery filter when power is applied.
>>a) Are you actually using a relay?
>Yes, we are currently using a DECENT relay, but the relay and the filter
>have a 90% chance that they will disappear in the future (will be inside
the phy)
>Please note here, that we are not out to discuss the millivolts, the
number of
>components,  to start do we as a group beleive that this loopback appoach
>where differential signaling is employed with magnetics- something we 
>know how to do, handle, and deal with will be robust over a current pulse
>from the center tap approach.  Also please note that this detection
>scheme may have good alternatives, and they may be your's or Mr X...
>so Let's keep in mind that.  On the other hand my argument is that
>differential signals, magnetics, and phy integration are realistic, familiar
>and with some improvements!!! lower risk?  what am I missing?
>even if this was your approach that would still be my engineering opinion...
>enough said.
Whee-ew! I just asked, because it was not clear if the relay was a
conceptual thing or really used. Most people seem to like to avoid
electromagnetic devices these days as I'm sure you know. Your partner in
this has been treating everything like an NDA, so I thought I might as well
ask The Man Who Knows.

>>b) Do you have to use a relay for some reason, or could there be some
>>solid-state analog switching equivalent. If so, what are the isolation
>>requirements, etc.?
>Again, Improvements to this scheme will place relays next to the phy
>or inside the phy delivering esd protection and 1500v-2000v isolation
>via the magnetic.  Something we have done for years...
>A lot of this can be looked at once we open our mind to the possibilities
>and think that things may not be  as risky....

>>2. What kind of margins are you asking for/getting in terms of S/N on your
>>detection pulses? 
>Larry, the detection pulse is supposed to be a low frequency pulse in schemes
>A, and B above you are looking at wider pulses that the FLP = 100ns up 10
1us, that
>could easily leave 400-500mv (again to be improved when we all get involved)
>Please note we are talking DIFFERENTIAL SIGNALS.
> when I designed the 10BT phys, I remember that
>the spec was at least 900mv for the 5mhz signal (preamble) to turn the slicer
>on, but a lot of aggressive vendors designed down to 500mv to claim longer
>cat-3 cable length performance .....  Details....
Now THAT is a non-responsive answer. The reason I asked it is that we have
gotten the impression that the reason for the relay is that the system is
kind of on the edge for noise margin. None of this is Rocket Science; the
Devil is indeed In The Details.

>>a) Do you have to cope with a certain amount of bit errors here? 
>Once finished with the detection, Detection  Pulses disappear,
>and then we are back at doing 10/100 til link fails....
>this is honest 10/100 after all, and we have done it for years,
>and per my MLT3 - power/no power -no difference slides, I can say that the
>are crude and unused doing 10/100 operation, we have not seen any BER
>degradation- again I did not test to every phy that ever existed- and every
>>b) How about some scope pix of what the detection pulses look like after
>>they go through the filter and a round-trip run (200m) of CAT-5 cable and
>>c) What are the tolerances (maybe a template) on what the returned pulse(s)
>>must look like?
>>Our SPICE models show what we have seen in the slides to be rather
>>marginal, but we don't claim to have the whole picture.
I was talking about the detection pulses, not what you get after the power
is on. I do not think you have shown anything of that, at least not yet, or
I missed it. My reason for asking is that there seems to be a certain
amount of filter de-Q'ing in your differential scheme, similar to what we
found in the common mode scheme because of the low load impedance on the
filter output.

And I think that a digital filter would be a damned clever solution to this!
>>These are the kind of questions that will have to be answered for the
>>standard even if your proposal were the only one, so you might as well get
>All right Larry,  I could   do more work here, and  present the slides for
>B, C, and D.  above.   what I would like to know though is that does
anyone believe
>that this differential scheme is worth looking into- it seems like I am
the only
>one who always show up with real measurements, are you going to show the
>details of your scheme?, is Dan doing the Same. and everyone else.
>it would be encouraging to get out of this mode where this is CISCO's scheme
>think like engineers see the pluses and minuses of everything and get to
>a decent solution.  So far all we seem to care about is mid-span power 
>unless of course I really prove that every other detail in the signal - pair 
>approach is solid and has no holes, leaving me little time to understand 
>your solution that may be superior.
Well, as I said, if you want your system to be a standard it has to be
characterized. We expect to have to do the same. We are working towards
presenting to the Committee, I don't know about others. Unfortunately, we
have had limited bandwidth to deal with this, but things are looking up...

As to the "CISCO thing", I'm afraid that you guys brought that on
yourselves and seem to be continuing to try for the Darwin Award in Public
Relations. Blind-siding the Committee and then demanding that your product
announcement consitute a "de facto" standard is not the way that the IEEE
works, though this is not the first time that this has been tried, as Geoff
Thompson has pointed out here. But I think we are all getting tired of
hearing this.

However, I do want to emphasize that the Committee was leaning against
using the signal pairs for what we considered to be very good technical
considerations BEFORE we knew of the Cisco effort. I think that if you go
back over the correspondence and arguments, and I think you were there,
there have been very specific technical statements and reasons given by Dan
Dove and the rest of us. To keep saying that we keep bringing in Science to
pick on Cisco is really kind of absurd.

And I fully expect that the Committee will closely examine and question
whatever we or others propose; that's how we get to the best solution for a
>After years and million of Ethernet ports, do we doubt our ability to
loopback a 
>differential pulse and have a differential stage detect it ????
>it may be a challenge to integrate this on a phy, but it is not impossible.
I have no doubt at all. However, as has been pointed out, not everyone is
visualizing this as being a PHY kind of thing.

Anyway, I seem to have relieved your boredom for a bit....