[802.3ae_Serial] SJTP: Minutes from this week's conference call
All,
Attached, find the minutes of the last conference call.
John Ewen,
There are 3 requests/comments mixed into the minutes.
Search for your name and you should find them.
All,
In the presentation I released to David Law and Jonathan,
I forgot to add a comment that we're still looking for a
way to determine the BER for the WAN patterns. I've now
added that to the presentation.
See you all next week at the meeting. I'll send a note
to the reflector if there is any need to continue this
ad-hoc.
Thanks,
Ben
--
-----------------------------------------
Benjamin Brown
AMCC
2 Commerce Park West
Suite 104
Bedford NH 03110
603-641-9837 - Work
603-491-0296 - Cell
603-626-7455 - Fax
603-798-4115 - Home Office
bbrown@amcc.com
-----------------------------------------
Conference call of the Serial Jitter Test Pattern ad-hoc
03-July-2001, 1:00pm EDT
Attendees:
Ben Brown
Jennifer Evans
Tim Warland
Tom Lindsay
Don Alderrou
Piers Dawe
Bill Gintz
Proposed Agenda:
Review the latest WAN test pattern proposal from Tim.
Discuss the seeds and waveforms that John Ewen sent out.
Discussion:
Bit errors can be counted using B1/B2/B3. It might be
additionally useful to have information that the error
actually occured in the frame and not the Bx bytes.
Perhaps this is less useful.
Information required is the fact that we're in sync and
a BER.
How do we convert B1/B2/B3 errors into a BER?
Given random errors, you can make a reasonable estimate of
the BER if it is not very high. There is likely a limit on
the BER that can be reflected.
Even number of bit errors are not counted in B1/B2/B3.
Error rate reported is half of what is reported for high
BER. For low BER?
LAN Patterns:
We either use these individually (same value for both
seeds) or get another LF pattern.
Patterns chosen based on the stress on the particular
features such as disparity, baseline wander & transition
density.
These should be checked to see which is more stressful/useful.
Question for John Ewen:
What time constants were used for
Transition Density and Baseline Wander?
Comment to John Ewen:
In slides, the last bit of the pattern
should be 16895 not 16896 (starts with 0)
Patterns A & B with their running disparitys of > 200 may
be more stressful than once-a-day occurances. Pat will try
some calculations to verify this before next week. C & D
seem more likely...
Results from Piers using only half the length of the pattern
(no sync bits and no inversion). The pattern seemed too short
to be effective. A longer pattern is more stressful - standard
PRBS. All 4 patterns seemed less stressful than a 2^31 (or
perhaps 2^23) PRBS on a part with a known low frequency
problem. The short patterns didn't catch it.
The test will be repeated with the full pattern, but perhaps
not before next week.
It is expected that the PRBS is always a better test but it
is not convenient to get through our hardware (the PCS). It
would also be interesting to see if repeatability helps or
hurts.
Run a test where you're sending IDLE or LF through the scrambler
without reseting it. Does the device pass some times and fail
other times?
Seed selection:
Is it premature to do this? Testing is very limited and the
testing that is done is less than promising.
It would be nice to put these seeds out there since it might
get others to start using them.
Should we "scare" those about to go to silicon by stating that
they might want to figure out a way to get longer patterns
from their silicon?
The patterns are probably long enough but we don't have the
right stresses in them.
Pattern D is a good pattern for low stress (maybe not low
enough but it's closest of these 4). Jitter is looking for a
higher stress.
Clause 52 should include 2 seeds in the text then add the
other 2 seeds in an editor's note. Ben to make this comment
before tomorrow at midnight.
Which seeds should go where? It probably doesn't matter.
Wanted 2 seeds, one typical and one stressful. None of these
look particularly good as typical.
Request to John Ewen:
We may want John to look for one less
stressful for this purpose.
Is there ever a case where the 2 seeds are different for the
same test? Transceiver testers will probably load the same
value in both seeds - Clause 52 could have any number of
seeds, one or more for each test, that are loaded individually
as each test is started. A system tester would use 2 seeds
and let them run, testing the quality of the link without
looking specifically at the optical signal.
Thanks to all for attending!
Call ended at 2:25 EDT