[802.3ae_Serial] From serial PMD call 5 Feb: test fest, meeting, ghost writers, eye mask sigmas
Present
-------
Piers Dawe Agilent SPG
Adam Healey Agere
Greg LeCheminant Agilent T&M
Stretch Camnitz Agilent T&M
Peter Ohlen Optillion
Petar Pepeljugoski IBM
Jamie Allen?
Tom Lindsay Stratos
David Kabal Picolight
February Test Fest
------------------
See also Greg's email of 6 Feb. Some discussion of objectives for these two
days: I think the first objective is to establish viable i.e. practical,
reasonably reproducible, reasonably accurate, test procedures. The next
objective is to gain "typical to worst" measurements with a view to setting
spec limits. I think someone said that achieving? showing? the viability of
"plug and play" was important (apologies if I have mangled that thought in
transcription).
Around six participants are expected, all? 1310 nm. Our hosts can generate
a new style stressed eye, which can be manually adjusted, and run automated
bathtub and TDP measurements. If I understood it right, the reference
(good) transmitter is the stressed eye generator with the stress dialled
down to zero. If he finds the time, Stretch will try to send out some block
diagrams of the apparatus to be used.
Calibrating the experiments when the reference transmitter is at 1550 nm and
the DUT is at 1310 nm should be feasible but extra information is needed:
relative responsivities and sensitivities of the test receiver and the
receiver under test at the two wavelengths... However, the reference
transmitter in this case is at 1310 nm.
Petar pointed out that quantifying vertical eye closure is difficult even in
simulation. Petar can you elaborate on this?
He also pointed out that the nominally 7.5 GHz receivers or filters vary in
practice.
Peter Ohlen asked if anyone had tried the split-and-delay filter to create a
virtual TP3 for 10GBASE-S. An ominous silence! Has any reader tried it?
Plan for Wednesday and Thursday
-------------------------------
David Kabal's plan is as follows:
Now: Updated comment database exists, not sure if on web. Draft 4.01 is on
the web at time of writing but clause 52 is unchanged in it.
Friday?: Soft deadline for presentations.
Before the meeting: alternative subclauses to be drafted, Piers to
coordinate:
Piers Dawe Revisions to eye and jitter sections
Tom Lindsay Alternative stressed eye generator
Peter Ohlen TDP for all wavelengths
Have we missed anything significant?
Tuesday: Teleconference to review progress and schedule Wednesday.
Wednesday: Presentations, as in any 802.3 meeting. Also report(s) from test
fest. Discussion to agree a test methodology.
Wednesday evening: motions to give enough direction to proceed to a draft.
Want 75% consensus, best achieved with motions backed by presentations and
with written alternative texts so voters can eyeball the detail. These
should contain spec limits which for preference are near to final.
Thursday: Comment resolution. For domestic reasons the meeting will not
drag on into the evening.
Eye mask sigmas
---------------
I tried to explain a way of looking at the relevance of noise in an eye mask
which is very simple in concept but hard to explain. Let's try again.
It's a one-dimensional analysis. Consider systematic eye closure,
transmitter noise and receiver noise. At sensitivity, the two noises
combine as root sum of squares and the result adds linearly with the eye
closure to add up to the eye amplitude.
Say we can measure the distribution of the eye to its half-way point (the
systematic eye closure, or the median of a distribution), to a point where
84% of the distribution is covered ("mean+1 sigma" if a normal
distribution), 98% ("mean+2 sigma") and so on. Now which one is a good
predictor of sensitivity penalty? The result was a surprise. For the
amount of noise which we allow in optical transmitters (quite small), the
transmitter noise is largely drowned by the receiver noise at sensitivity.
Far from a "7 sigma measurement" (which we can't do on a scope) being a good
predictor, "mean+1 sigma" is much better, and "mean only" is not bad. This
is a bit like the "W" intercept value in the jitter bathtub approach.
At present we measure eyes to "2 to 3 sigma". We are over-measuring them!
Now for the bad news. Trying to reduce the "number of sigmas" in an eye
measurement by reducing the number of waveforms measured would lead to very
erratic measurements. I wonder if allowing x hits in y waveforms would lead
to a measurement that was both reproducible and a good predictor of
performance. Does anyone have a suggestion for x and y, bearing in mind
that each waveform contains many samples but one sample can make one or no
hits?
And now the complication: scope noise.
Simulations of stressed eye generator
----------------------------------------------
Petar's simulations had not-quite worked. He hoped to be able to post
results within two days.
Next phone meeting
------------------
The next PMD teleconference will be on Tuesday 12 February, at the usual
time and coordinates:
4:15 pm GMT = 17:15 CET = 11:15 am EST = 8:15 am PST
+1(816)650-0631 Access code 39209
Piers