Hi Albrecht - see within.
I've copied this onto the reflector.
----- Original Message -----
Sent: Tuesday, April 26, 2005 2:57
AM
Subject: TP3 noise discussion
Hello Tom,
For today's TP3
call, Jim asked us to look at the noise
calculation again. Therfore, I
would like to share my
thoughts on this topic with you. May be I am
missing
something completely, which is the reason why I am
sharing
my thoughts with you and not sending them
directly to the LRM
reflector.
Our main concern about the TP3 SNR definition
is
that we don't see the relationship
between the link budget
and
the SNR calibration as currently defined by the Qsq
ratio
of 11.5 in the LRM
draft.
Regarding the link budget, the LRM community is
assuming a
reference SNR at the
input to the EDC of 30dBe (17+2x6.5).
This
value is used for TWDP calculations and for
PIE-D
coverage work and assumes a reference
channel without ISI
but with RIN and MN penalties.
TL - this particular calculation (for TWDP) does
not include RIN or MN. It is based on a matched filter bound receiver
that provides -13 dBm optical sensitivity to a perfect rectangular
waveform. The min received OMA is -6.5 dBm, so the difference is 6.5 dB
(and hence, the value in the equation).
Having written this, I think the value may
have a problem. Perhaps the SNR without these terms should be 32 dBe. The
latest budgets (see Nov 04) show sensitivities at ~-14 dBm (not -13),
required to accommodate 1 dB of penalties due to RIN and
MN!
For
the TP3 SNR calibration, the value of Qsq
is
derived from 0.9dBo noise power
penalty. Even Piers
pointed out in
dawe_1_0504 that the contribution from
MN of
0.5dB is pessimistic as it was defined
for
GbE where the exact value did not matter and OFL
was
used. We further noticed that a
slight change from 0.9 to 1.0dBo
will
have a large impact on Qsq (11.5 to 12) and the resulting
SNR.
TL - the penalties were set at 0.5 and 0.4. They are
independent and assume no other simultaneous noises. These
cannot be simply added when computing penalties. Mathematically,
they combine to 1.0 dB. The math is right, but if the values for the
individual terms are too high, then we should lower them (which will
reduce the combination).
The
SNR corresponding to Qsq=11.5 is 20log11.5 =
21.2dBe
Using Lew's assumption of -3dBo noise
coloring due to IFR,
this improves the SNR at the optical
receiver input to
about
27.2dBe. Assuming a reciever sensitivity of -14dBm and an OMA
of
-6.5dBm, this results in a SNR at the EDC
input of about 25.8dBe,
which is
significantly below the 30dBe reference SNR
derived
from the link budget. To
make the link budget and the Qsq approaches
match, you
would need to reduce MN+RIN or increase
IFR.
Should these match at all?
TL - I do think there is a problem. For example, the RIN
spec produces Qsq = ~29 at TP2. MN may bring it down further, maybe
~20 - not 11.5. 0.5 and 0.4 came from experience at 802.3ae, where
an overall Qsq on the order of 20 produces penalties in the overall budget
of 1 dB due to interactions with other impairments, mostly ISI closure. We
should not be starting with 1 dB, but ending there. Now, there is the
question of whether any of these are the right values for 802.3aq! Perhaps
the 29 factor is fine for RIN at TP2, and perhaps 0.4 dB for MN is okay
(at TP3, not a TP2), but the interaction with all the dispersion is not
well understood. That's the rub here. Paul Voois did some early
spreadsheet work on RIN, and that work should be revisited. See http://grouper.ieee.org/groups/802/3/aq/public/may04/voois_1_0504.pdf.
I'll try to take some time to look at
this work. Finally, I wonder why we can ignore jitter during
the stress test,when we needed to apply
it for 10G-Base-LR .
TL - if we reduce the
noise, we may need to add some jitter back in, although the added
complexity is not desirable. Currently, the noise is high, so no added
jitter is needed.
--
Best regards,
Albrecht Rommel
Acuid Corporation Ltd.
Phone +49 (0)8141 3555 813
Fax +49 (0)8141 3555 829
Mobile +49 (0)172 607 8039
albrecht.rommel@acuid.de
www.acuid.com