|
Ali,
Thanks
Ali, I think I better understand your concern now.
As I
recall, the return loss equation was simply the 10GBASE-CX4 return loss equation
(a -12dB plateau from 0.4 to 2GHz) extended to 15GHz and was a fairly arbitary
placeholder. Channel data presented to the task force has shown that this
is an unrealistic expectation of the channel and must be corrected. I
believe that Joel has identified this as a work item for the channel
model ad hoc and given the volume of channel data presented from multiple
parties, I believe it is one that we can close.
Do you
have a specific proposal for the return loss mask?
Best
Regards,
-Adam
Adam
Please see my comments
below in red.
Healey, Adam B (Adam) wrote:
Hi
Ali,
I
do agree with TP-1 and TP-4 definition.
They are located at the "component edge" of the transmitter and
receiver respectively. The definition of "component
edge" includes any external termination components required by the
transmitter/receiver and the AC-coupling capacitors, when used.
This implies that the backplane connectors (and mezzanine connector,
when used) are part of the "channel" specification. Channel
return loss requirements would need to be satisfied looking into the channel
at either TP1 or TP4. Transmitter return loss requirements would be
defined looking into TP1. Receiver return requirements would be
defined looking into TP4.
Channel return loss, based on the conventions that we are using, will
likely be dominated by the first mated connector as you point
out.
To
include the interaction between the transmitter/receiver and the channel, a
cascaded system model is required. This can be accomplished a variety
a ways (cascaded T-parameters for example, refer to OIF). I suspect
that this is something the signaling ad hoc will need to address. The
charter of the channel model ad hoc, like the cabling ad hocs 802.3 has had
in the past, is to define the channel return loss looking into TP1
and TP4. I use the word "channel" rather than backplane, as it
now should be clear the our definition of TP1 and TP4 is not only the
backplane but the trace on the node/hub (line, daughter, whatever
naming convention you prefer) cards as well.
I
apologize if you already understood this. Of course, you may not agree
with the above methodology, which is fine. Let's discuss it on
the reflector. It sounds like you are advocating that we define
the backplane connector, which is something I think we have been trying to
avoid. Please clarify. Adam, I was not advocating specifying a backplane connector and
looks like based on the the definition
of goergen_03_0704
you already have connector as part of the channel. The part I am puzzled
is the return loss
associated with current channel model "-14 dB up to 15
GHz". Based on my data and experience specifying such
low return
loss is not practical unless you introduce significant loss on the
daughter-board. With 14 dB return loss at
TP1/TP4 toward the
channel you are practically ignoring the most significant effect "double
reflection between the connector and the IC"!
Thanks,
Ali
Thank you,
-Adam
Joel
Joel Goergen wrote:
Ali,
We have been focusing
on this for several months now. I think Steve has presented
nothing new that hasn't been published and, at a min, agreed to in some
form of straw poll.
I am
not disagreeing that these were agreed in the straw poll, I just didn't
participated in some
of earlier meeting.
I would like to offer a
thought ...
For the channel ... we use the SMA and SMA foot print
that allows for a clean launch. At both ends of the channel.
This allows us to see the SDD11/SDD22 without doing a major de-embedding
and still allow for freq to 12.5Ghz.
How can you define a channel "backplane" but not define a
minimum attribute for the connector.
There is no reason to de-embed the
connector as the connector is part of the channel. A
compliant
channel must meet an specified transmission and reflection
property which include connector.
For the tp1 and tp4 ...
we then let the tx (tp1) handle the BGA and via, the rx(tp4) handle the
via, both cap pads, both via and BGA pad. It will be easier to do
the de-embedding and specify the SDD11/SDD22 as seen by the chip from
the channel.
The
methodology specified currently in BP is suitable for chip to chip
applications but not for backplane.
Where do you handle connector
effect and multiple reflection between the connector-IC? In 4Gig FC
we addressed some of these issues and do specify the channel which include
the connector.
Thanks,
Ali
Just my own thoughts
.....
-joel
Ali Ghiasi wrote:
Steve
Looking at your channel model, I see the return loss up to 15
GHz is
better than 12 dB.
How do you get -12 dB return loss at
15 GHz with connectors, what kind
of connector
are you
assuming? The most challenging effect are the primary
reflection between
silicon and the connector at this speed.
Thanks,
Ali
Stephen D. Anderson wrote:
All:
I
updated the synthesized S-parameter files that I presented in
Portland. The new revision provides a much better NEXT
file. The Thru
file was also improved so that SDD11
magnitude has ripples. You may
recall that the previous
SDD11 magnitude was flat over most frequencies
of interest and
that this was viewed as unrealistic.
These are
intended to match the NEXT and Thru templates that we have
thus
far agreed on.
If you see anything blatantly
wrong with the files, please let me
know.
Regards,
Steve A.
|