Twisted-pair model for HSPICE

[Sapling]

Hi,I'm looking for a Twisted-pair model to do simulation.There is a Twisted-pair model I found in IEEE 802.3 standard.But i don't know it was equivalent to how long twisted-pair in real.50m or 100m or something else?
Any help is much appreciated. Thanks!

 

[FvM]

You find the answer in clause 14.4, "Characteristics of the simplex link segment". (100 m)

Please consider that the circuit is intended to model certain properties of an ethernet twisted pair in a hardware test. It's not intended as general twisted pair simulation model.

 

[Sapling]

Thanks for your answer.I extremely need a Twisted-pair model to do simulation.Do you know where I can find one?Thanks

 

[FvM]

Depends on which properties of a twisted pair you want to model. Did you read the transmission line chapters in the HSPICE Signal Integrity User Guide?

 

[dick_freebird]

If you had the actual wire manufacturer and P/N you
might find either they, or some customer has fitted a
transmission line model (whether compatibly with your
particular simulator, or not...). Seems that even wires
of the same class (like say RG-58) have a lot of detail
variations that alter the attributes (shield material and
construction, solid vs stranded core, choice of dielectric
material and diameter).

Now if you only "need to do simulation" and are not all
that worried about accuracy and matching to a real
piece of hardware someday, you can be less picky
about just how good (or even valid) the model is. Just
what (besides, say, the difference between a B+ and an
A-) do you care about in all of this?

If you can describe the cable then Googling "{cable_ref}
SPICE model" will turn up plenty for you to rummage
through. Then you get to port it to your simulator if
you're not lucky enough to find one with a pretty pink
bow on it.

 

[Sapling]

I want the properties of twisted pair model it can run like as the real one.Maybe I could build a model by using the HSPICE W-element RLGC model.I will study HSPICE Signal Integrity User Guide.Thanks.

 

[Sapling]

Hi,dick_freebird
I don't have a transmission line model fitted by some customer or anyone else.
I'm designing an Ethernet chip which has a 10M/100M PHY.I found a problem when I test it by using category 5 cabling.It work abnormal When the category 5 cabling is longer than 70m.So I need a Twisted-pair model which can work like 100m or anyother length category 5 cabling to do simulation.Then I could find out the problem of my design.Also I can use the model instead of 100m category 5 cabling to test the Ethernet chip.The category 5 cabling is worse compare with Category 6 or better-balanced cabling.If my design can work normal when using 100m category 5 cabling,It also can work normal when using Category 6 or better-balanced cabling.So I worry about accuracy and matching to a real
piece of hardware.
I had try google "Twisted-pair model" or something else,But I can't find what I need.

 

[FvM]

Guess you noticed that the figure 14.7 isn't intended to model twisted pair behavior for 100BASE-TX transmission. Another problem is that it only models the signal attenuation and respective waveform distortions but not other effects like reflections and crosstalk. In so far it possibly might not show the problems you are looking for.

You can of course design a cable model that represents arbitrary properties of a real cable but it won't help much if you don't know which specific effect is causing transmission failure.

I believe it's more promising to analyze exactly how your receiver fails.

 

[dick_freebird]

If there's a problem that is deterministically related to
length, that has to be either reflections (intersymbol
interference) or attenuation. Until you can better say
which, you need a model which supports both. The
lumped element model may give you a look at attenuation
but has nothing for reflection - there, you want a tline
model. There are several variants and most of them I've
found too difficult to work with. But maybe you start
with trying to pick at it with the lumped model for atten
and a simple primitive ideal tline for reflections and see
if either gets close to observed behavior.

 

[Sapling]

Hi FvM and dick_freebird
Now I had fabricated about 40 pieces chip and tested it by using category 5 cabling,they communicated with other company's Ethernet controller chip,some failed in 60m length cabling,some failed in 70m length cabling,some failed in 80m length.But if I used category 5e cabling to do communication test,they all can pass in 100m length cabling.It meas that if the cabling is better they can pass length is longer.So if I had a Twisted-pair model of category 5 cabling,It's very convenience to week out the chip can't pass 70m(or 60m etc) length cabling in CP/FT.And I found an equation of insertion loss in TIA/EIA-568-B.2(clause 4.3.4.7)
I think it's properties is about 100m category 5 cabling,so dou you know how can I turn this equation to circuit which formed by RCL?

 

[Sapling]

Hi FvM
Yes,the figure 14.7 can't use at 100BASE-TX transmission.

 

[Sapling]

Hi dick_freebird
I will research the tline model and how to turn "lumped model+tline model" into one model.

 

[dick_freebird]

The various CAT5_ cables ought to have the same electrical
length per physical length so the "Cat5e works, CAT5 doesn't"
would tend to discount the ISI possibility and leave attenuation
(perhaps violating the RX sensitivity / drive level requirement
at the far end device) as the likely problem.

Now what I do not see here is any attempt to quantify that,
or break the problem down into TX, medium, RX elements or
even observed electrical differences.