[802.3ae_Serial] Trying to measure Jell-O with a rubber ruler
From serial PMD call 4 December: topic was optical test methods for the
serial PMDs.
1. Stressed eye generation
2. Alternatives to stressed eye
3. Status of "golden PLL"
4. Floating eyes?
5. Patterns
6. 10GEA
1. Stressed eye generation
-----------------------------
Petar has created an incomplete stressed eye.
It would be useful to discuss a scope plot as an example stressed eye, to
see if we can agree an improvement on fig. 52-14.
Question of how/whether to correct for a stressed eye ISI which is not quite
as specified.
2. Alternatives to stressed eye
----------------------------------
Proposal to separate stressed receiver conformance test into two parts:
A. Jitter tolerance with fast edges but low amplitude, possible some
other high-probability jitter to bulk out "W". Could be a relative (jitter
penalty) test, like SDH/SONET. This test would screen for a combination of
receiver speed, Rx jitter, and decision timing offset. If DCD is
significant, might have to be included here.
and
B. Receiver minimum bandwidth test: use method of 52.9.13 "Measurement
of the receiver 3 dB electrical upper cutoff frequency". The idea is that
this test takes the place of the slow eye part of the stressed receiver
conformance test, both being failed by receivers which are too slow.
But this bandwidth test apparently is also not proven. Are there volunteers
to check it out? No. Also, a receiver with adequate bandwidth but poor
group delay could pass the test because the phase of the interfering sine
wave is not important in the test, but still receive slow signals poorly.
[The following week it was pointed out that the high frequency part of the
signal would be distorted also. Not sure where this leaves us.]
Can anyone do a thought experiment for a technique(s) which is accurate
enough?
Can we map test coverage onto the 5-dimensional matrix Tom mentions in
http://www.ieee802.org/3/10G_study/public/serial_adhoc/email/msg00489.html ;
or should this be, map to around five possible receiver weaknesses?
3. Status of "golden PLL"
----------------------------
How much jitter will the "golden PLL" or clock recovery unit (CRU) in the
test equipment generate? We don't know. Has anyone aware of the existence
of such a clock recovery unit? Will CRU lock to the "square wave"
pattern with its low transition density? This isn't strictly required, the
CRU could be switched out in tests calling for the square wave, but that
adds some inconvenience and cost. In general, CDR performance is affected
by pattern length. Tom would try to find a Fibre Channel document which
attempted to specify a golden PLL at 1 Gb. (Sent following the meeting:
http://www.ieee802.org/3/10G_study/public/serial_adhoc/email/msg00492.html
.)
4. Floating eyes?
--------------------
Are eyes allowed to float in time? Consensus answer: not unless the CDRs
are adaptive. Thought the original OFSTP-4 made this clear but didn't have
document to hand.
5. Patterns
--------------
Patterns: our game plan seems to be, sort out test procedures first then
check out the patterns. But time passes...
Four reasons for using traffic as our yardstick rather than fixed patterns
1. As Juergen and the ITU say, in a network it can be impractical to
get the transmitter put into a special test mode.
2. For NEM's convenience - they can also avoid a test mode and can just
look for dropped packets.
3. If all parameters are measured on the same pattern, may
simplify/speed up/cheapen testing in the transceiver factory: e.g. could
capture eye, extinction ratio, mean power and OMA from one measurement
rather than one on a mixed signal and one on a square wave.
4. Because it is what matters. We want to achieve interoperable
carriage of packets, not interoperable transmission of test patterns.
However, specified patterns may be a useful fall-back for some factory tests
e.g. transmitter jitter bathtub, for test-equipment reasons.
6. 10GEA
-----------
How could 10GEA help the optics-test/spec development? No really good
suggestions.
Piers