AW: [802.3ae_Serial] Jitter experiments
Hi all,
Sorry for not being able to participate on the last call (I have a conflict
tomorrow too). However I want to give some interim info about tests on
jitter we did. We used a SDH test set for jitter based on the ITU
definitions. So we measured the broadband jitter generation and tolerance
in the frequency domain between 4 and 80Mhz. However as concerning the
tolerance it is not likely that a system is more tolerant to jitter at
frequencies even higher above the PLL border the results may give some
understanding of the underlying principles.
We used commercial transponders as available on the market. We measured the
generation and tolerance w/o fiber to be free of influences of fiber effects
(Which may be traded off in the future) (TP2=TP3). The Jitter generation of
random jitter of the Transponder was at the measurement limit and smaller
than the jitter comming out of the Test set. (Which gives me the
understanding that all this was clock jitter that will be further reduced by
the PLL an the parallel stream. There was no visible DJ on the scope of the
transmitted signal. Under these conditions we observed a tolerance for a BER
of about 10 to -10 of a bit above 0.4 UI P-P sinus jitter, which was nearly
equal to total jitter as stated before, nearly constant for frequencies
above PLL border frequency. We will repeat those tests for other error rates
to get an understanding how the theoretical bathtub may look like (as we
used commercial modules we have no way to shift the sampling point through
the eye) and get an understanding about the penalty induced by which jitter
amplitude. It should be noted that the penalty in the measurement we did was
real big, and so I do not believe at al that if we have at 0.4 UI P-P (of
whatever waveform the jitter follows) a penalty that eats up a big part of
the budget we see the nominal sensitivity at 0.35 UI. (So this somewhat
confirms the measurements from Intel) I will come up with more dedicated
results.
Regards Juergen
----------
Von: Lindsay, Tom [SMTP:tlindsay@xxxxxxxxxxxxxxxxxxxx]
Gesendet: Samstag, 1. Dezember 2001 01:21
An: 802. 3ae Serial PMD (E-mail)
Betreff: [802.3ae_Serial] Jitter experiments
This email is in response to my committment on the 11/27/01 serial
PMD con-call. This is still very crude, and I apologize for lack of
appropriate thought, but hopefully it can stimulate more discussion.
****
A major concern for 10G serial is instrumentation error for doing
jitter measurements. Steve's Buchheit's work clearly demonstrates this.
Ideally, instrumentation will improve sufficiently to allay this concern,
but none of us expect that to occur sufficiently in advance of deployment.
So if we are to keep the basic definition and method for jitter measurement,
then we require agreed upon means to compensate the instrumentation.
With this in mind, I simply brainstormed 5 categories of test
settings and measurements, with the hopes (dreams?) that differences in
their results can provide insight into how compensation can be achieved.
1. Measuring instruments
Scope - eye patterns w/ crossing histograms
Error detector - jitter bathtub
2. Configurations (essentially the same used by Steve)
Pattern generator to measuring instrument
Pattern generator through E/O and O/E to measuring instrument
Pattern generator through stress conditioning and E/O and O/E to
measuring instrument
Pattern generator through DUT and O/E to measuring instrument
3. Test patterns
Pattern 1
Pattern 2
PRBS31
00 00 00 00 00 00 00 03 FF FF FF FF FF FF FF FD (isolated 1,
isolated 0)
Repeating CC (square wave with 50% transition density, same as
average PRBS)
4. Signal variations
rise/fall time (at least 2 levels of slow/fast)
amplitude (at least 2 levels of low/high)
5. Stress conditioning mechanisms
high frequency ISI/DDJ
low frequency BLW/DDJ
At this point, I do not know if there is a path through this that
will lead anywhere (this is the apology again...). I hope someone else can
determine if there is really any value in here. The 5-dimensional matrix
results in a large number of test combinations - too many. A Design of
Experiments may be required to reduce the set. Obviously more directed
experiments can be defined with some risk of missing key information.
Tom Lindsay
Stratos Lightwave, NW design center
425/672-8035 x105