Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: [802.3ae_Serial] Draft comments




Piers,

On ANSI/TIA/EIA-526-4A-1997, it should be added as a normative reference
(I'm surprised Shimon didn't catch that one in his sweep of Clause 52).

It should not be deleted from B13 because Clause 36 references it there
(which should be fixed in a maintenance change because Clause 36 uses it
normatively, but 3rd sponsor recirculation is past the point where one
should do service to humanity changes).

On clause 45/49 - This isn't open for comment - we didn't change it in D4.2.
Furthermore, the number of errors produced at the PRBS31 output will be
about the same as the number of bit errors in the received data stream
because the descrambler also induces error multiplication by 3 for isolated
errors. One cannot divide the number of errors by 3 because errors occuring
in bursts may not have been multiplied by 3.

Leave it as it is. We have clearly explained what is being counted.

Pat

-----Original Message-----
From: piers_dawe@agilent.com [mailto:piers_dawe@agilent.com]
Sent: Thursday, April 25, 2002 4:02 PM
To: stds-802-3-hssg-serialpmd@ieee.org
Subject: [802.3ae_Serial] Draft comments



As requested, my wish list in the form of draft comments for discussion on
tomorrow's serial PMD phone conference.

Clause	SC	page	line	comment / remedy
----------------------------------------------
30	30.5.1.1.4	63	39	enmeration
					enumeration

45	45.2.3.16	226	1	Draft says "The test pattern error
counter ... contains the number of errors received ....  This counter will
count either block errors or bit errors dependent on the test mode (see
49.2.12)."  But 49.2.12 says "When an isolated bit error occurs, it will
cause the PRBS31 pattern error signal to go high three times... The test
pattern error counter shall increment once for each bit time that the PRBS31
pattern error signal is high.

Clearly, clause 45 is misleading, because the counter does not report
received bit errors, but around three times as many counts as received
errors.  You can't call the output of the checker "bit errors", that's what
is at the input of the checker; the signal coming out of the checker is not
an error or in error, but deliberately created.  It has to have a different
name.

It would be a disservice to anyone trying to write MDIO software and report
received errors, without taking time out to understand the detail of the
other clauses, not to tell him that he may need to divide the counter value
by 3 to get a good estimate of received errors.

					Remedy: "bit errors" with "asserted
bits at the bit error checker output".  Add another sentence "In the latter
case, a good estimate of received bit errors may be made by dividing the
counter's contents by 3."

1	1.3		7	1	52.9.4 refers normatively, as I see
it, to ANSI/TIA/EIA-526-4A-1997 which is informative reference [B13] in IEEE
Std. 802.3 Annex A, which by the way should say (OFSTP-4A) not (OFSTP-4).
52.9.7 uses a "should" so maybe that's informative.
					The entry presently in Annex A needs
copying or moving to the list of normative references, 1.3, (OFSTP-4A)
replaced by (OFSTP-4) and (if moved) the [B13] removing.  Note 38.6.3 refers
to it in a way that looks normative but 38.6.3 is a variation on what
ANSI/TIA/EIA-526-4A says.

52	Table 52-18	469	12	There can't be as much as 0.5 dB
additional insertion loss allowed at 30 km because we can't know that the
path penalty will change by that much in the last 10 km.
					May have to be 0 dB.

52	52.9.6.2	474	25	Number on different line to unit
					Use nonbreaking space

52	52.9.6.3	475	9	Wrong step, as Petar pointed out
					step c)

52	52.9.7	475	17	as per
					per ?

52	52.9.7	475	45	As I understand it, measuring the average
timing of the edges away from the crossing level (waist) introduces a new
form of error, because the scope will sample a random proportion of rising
edges vs. falling edges, which then occur at different times.  For typical
sample sizes, this creates a random timing error which largely negates any
benefit of moving the expected timing to the desired place.  In the example
I looked at we were talking 1 ps.
The mask dimensions are not chosen to 1 ps precision.
Mask measurements are disappointingly inaccurate already.  This would make
it worse.
The mask is not the primary measure of transmitter quality; TDP is.
There is an industry standard way of mask alignment already.  It adds cost
and confusion to all users, on an ongoing basis, to create another way of
doing it.

Greg LeCheminant can elaborate.

In other words, don't re-invent the wheel.  We pay test equipment
manufacturers to do a good job, let them!  

					Remedy: Delete "measured at the
average value of the optical eye pattern".

52	52.9.10.1	477	14	50.3.8 should be a link
					activate

52	52.9.10.1	478	49	After further analysis, I think
specifying "fourth-order Bessel-Thomson" here is actually counterproductive.
There is no point tightly defining the filter and allowing very loosely
specified amounts of sinusoidal amplitude interferer (particularly) and
sinusoidal jitter.  We don't know enough to really tie down the spec for the
stressed eye generator, so let's give the test equipment implementer a
chance to do the right thing.

					Remedy: delete "fourth-order
Bessel-Thomson" here and in Figure 52-10, and "Bessel-Thomson" on next page
line 4 and 15.  Could use " A linear phase filter" on first use: the phrase
occurs later on.

52	52.9.10.1	479	9	Breaks the one shall per test rule
					must be

52	52.9.10.1	479	16	Here we could hint at the bandwidth
of the filter (around 3.75 to 5 GHz).  Did we discuss this last time?
Giving this guidance would tend to keep implementers away from extreme
values of the sinusoidal terms and make for a more consistent test across
the industry.

					Remedy: Add sentence "An electrical
bandwidth of 3.5 to 5 GHz may be found appropriate."

52	52.9.10.1	479	5	Does a filter with wide and flat
frequency response and linear phase response do anything?  Even
Bessel-Thomson filters don't have flat frequency responses.

					Remedy: replace "flat" with
"smooth".

52	52.9.10.1	479	15	These two sentences are out of line
with good practice as in 52.9.7 and G.691 and if taken literally, would give
the stressed eye generator implementer an unnecessarily hard task.

					Remedy:  Replace "The Bessel-Thomson
filter should have the appropriate frequency response to result in the
appropriate level of initial ISI eye closure before the sinusoidal terms are
added. The O/E converter should be fast and linear such that the waveshape
and edge rates are predominantly controlled or limited by the electrical
circuitry."
	with
"The test pattern generator, filter and E/O converter should together have
the appropriate frequency response to result in the appropriate level of
initial ISI eye closure before the sinusoidal terms are added.  The O/E
converter should have a linear response."

52	52.9.10.2	479	42	Ambiguous.  Also I thought we had
scrubbed this use of "peak" per a comment last time.

					Remedy:  Replace "For this test,
these two components are defined by peak values that include all but 0.1%
for VECP and all but 1% for jitter of their histograms." with "For this
test, VECP is defined by the 99.9th percentile of the histogram of the lower
half of the signal and the 0.1th percentile of the histogram of the upper
half of the signal, and jitter is defined by the 1st and 99th percentiles of
the jitter histogram."

52	52.9.10.2	479	47	Text jumps abruptly into a recipe
without enough flags for the reader.

					Remedy:  Replace "Steps:" with "In
steps 1 to 7 below, a suggested method of calibrating a stressed eye
generator is described in detail."

52	52.9.10.2	479	52	Missing word
					but this increases

52	52.9.10.2	481	3	Here is where we try to keep the
pulse shrinkage within a range.
					Insert after "at least 5 ps": but
preferably no more than 15 ps" (peak-peak of pulse shrinkage jitter).

52	Figure 52-11  481	12	P1 and P0 look like the 1 and 0
levels: if so they should be next to the horizontal lines, not the histogram
boxes. 

52	52.9.11.3	483	21	"The clock recovery unit"  Which?
This is the first time one has been mentioned in 52.9.11.3.  Also, number on
different line to unit.

					Remedies: "A clock recovery unit",
use nonbreaking space.

52	52.14.2.1	488	27	Can we go forward with
"TIA/EIA-492AAAC is presently in ballot."?

52	52.15		491-5		As Peter pointed out, each
conditionally mandatory PICS needs a N/A []

I'm out of time, may add a few more tomorrow but this is most of it.

Piers