[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: CPCI term resistors
Perhaps using the word "termination resistor" is a poor choice of words on
the spec's part, and implies more than the 10 ohm resistors are supposed to
do. While i believe that Ted is correct that they provide some source
termination, Andy is also correct in that they minimize impedance
discontinuity across the backplane. A typical multidrop backplane (like
cPCI) exhibits localized impedance drops at every stub. That's just a
function of a wire connected to another wire - it's a transmission line
phenomena. These impedance "divots" appear capacitive to incoming signals,
and being good little capacitors, reflect a portion of the signal back to
the transmitter. The longer the stub, the bigger the capacitance and the
bigger the reflected signal. The summation of all of these little divots
across a multi-slot backplane leads to a signal that's just a mess. While
this can be alleviated by slowing down the signals (edge rate and duty
cycle) - thus averaging the impedance bumps across the whole of the
backplane - or by shrinking the backplane, the more stubs the worse it
gets. And speeding up the bus just exhibits this effect all the more. Note
that this phenomena occurs on both reflected-wave switched and
incident-wave switched systems - some are just more tolerant of it than
others. (The original PCI is more tolerant of this effect because it's a
shorter, more lightly-loaded bus.)
The solution to this problem is to modify the transmission-line structure
of the backplane, which is what cPCI did. Effectively, the 10 ohm resistors
"electrically shorten" the stubs by providing a little isolation between
the rest of the stub and the backplane. The most ideal (but impractical)
solution is to use the resistors right on the backplane to split off the
signals before the cPCI connector, but the spec didn't do that. And doing
that now would put you out-of-spec. So, the next best thing is to put 'em
on the cards, as close as possible to the backplane connector.
As far as source termination goes, this too is an advantage for the 10 ohm
resistors. When looking at the backplane wire from the point of view of the
stub (say, at the connector), the nominal backplane impedance looks lower
because the signal splits as soon as it hits the backplane wires; it has to
go two ways. The 10 ohm resistors help to bring this effective load
impedance up closer to the source impedance of the drivers.
mike nemeth
Rockwell Collins, Inc.
{Expressed herein is my opinion and not that of Rockwell Collins, Inc.}
Ted Firlit
<firlit@utmc.aer To:
sriram@cpmail.cyberpathinc.com
oflex.com> cc: pci-sig@znyx.com
Sent by: Subject: Re: CPCI term
resistors
firlit@utmc.aero
flex.com
03/21/01 09:05
AM
Please respond
to ted.firlit
K.C.,
I agree with the issue you bring up about bus lengths being longer on
cpci than pci, but as I understand it, the cpci series term resistors
are an implementation of reflected wave switching. At an incident
signal edge, the voltage level out of the resistor is lower than the
target voltage until the reflection from the end of the line gets back
to the resistor. If reflected wave switching is not used, end
termination would be required, as in a resistance to gnd, power or both.
Cpci spec mentions "stub termination minimizes the effect of the stub on
each board to the PCI backplane" in para 3.1.3.
I believe it also provides some attempt at impedance matching for source
termination of a reflected wave signal.
Ted Firlit
"K. C. Sriram" wrote:
>
> PCI uses reflected wave switching and it does not need any termination
> resistors. The length of the PCI buses are usually small limited to the
> motherboard.
> CPCI is meant to be used for the backplane and the lengths are extremely
> large. reflected wave switching cannot be used and needs the terminations
on
> the bus.
>
> -----Original Message-----
> From: Ingraham, Andrew [mailto:Andrew.Ingraham@compaq.com]
> Sent: Tuesday, March 20, 2001 4:02 PM
> To: pci-sig@znyx.com
> Subject: RE: CPCI term resistors
>
> > The CPCI specification requires 10 ohm series termination resistor on
the
> > PCI
> > lines. This minimize the effect of the stub on the backplane.
Standard
> > PCI
> > does not require termination resistors. Why is this?
>
> Standard PCI as a motherboard-only bus came first. Then came the
connectors
> and PCI cards. Standard PCI was designed without the resistors.
>
> CPCI came later, and I believe their inclusion, along with a different
card
> connector, allowed more slots to be used with less difficulty. At the
time,
> the typical bus had three, maybe four PCI slots, which was somewhat
> inadequate for some cost-effective industrial applications. CPCI allowed
> eight without a bridge.
>
> > I am not sure the exact
> > reasoning behind these termination resistors.
>
> They somewhat isolate the stub from the rest of the bus, resulting in
less
> impedance discontinuity, or less capacitive load, to the main bus at each
> stub. There is some added delay "through" the resistor, which is weighed
> against the reduced bus delay waiting for the bus to settle.
>
> This is sometimes called distributed termination.
>
> > Lastly, what would be the effect
> > of not having them ;-).
>
> I would expect you'd see worse ringing and longer settling times, but
YMMV.
>
> Andy
--
Ted Firlit (719) 594-8138
Senior Design Engineer
UTMC Microelectronics Systems
4350 Centennial Blvd., MS 1004
Colorado Springs, CO 80907