[802.3ae_Serial] Dubravko's model
FYI, the following notes are from a brief series between Dubravko and
myself regarding his model for budgeting jitter for the receiver.
Text beginning with "TL" is from me; his comments beging with "DB". The
YES and NO response are also from Dubravko.
---------------
TL - Given TP3 RJ and DJ values, and after some assumptions about Rx DJ
and DJ addition, an RJ generation limit can be determined for an Rx
between TP3 and TP4. Further given an eye shape (edge rate), the RJ
limit can be directly translated into a Rx sensitivity requirement.
If this sensitivity is lower in power than the unstressed sensitivity
value determined by the Agilent model, then the difference becomes an
Rx_RJ power penalty.
Is this correct? YES
Is it really this simple? YES
Any other assumptions about eye shape or sample offset? NO
DB - The limitation imposed on the RX contibuted jitter places a
limitation on the equivalent input noise of the receiver via the
edge-rate: Noise = edge-rate times contributed jitter. The equivalent
input noise determines the [RJ-induced] unstressed receiver sensitivity,
while the edge-rate depends on the TP3 risetime, signal shape, and the
OMA.
--------------
Back live - he and I have briefly discussed what assumptions to make
about Rx jitter. Rx RJ is assumed to be Gaussian and adds
root-sum-square with RJ coming in from TP3. Rx DJ between TP3 and TP4 is
assumed to add linearly with DJ coming in from TP3. Jitter tolerance at
TP4 is comprised of 3 effects - setup and hold (DJ), sampling offset
(DJ), and timing variation (assumed to be RJ caused by jitter transfer,
VCO noise, etc.). I do not know if his model breaks TP4 tolerance into
DJ and RJ terms. In talking to one serdes supplier, he thought that the
RJ portion would be about 1/3 of the total tolerance. Of course the
total TP4 tolerance is another variable.
Dubravko - this came up this morning during the serial PMD con-call.
Tom