RE: [802.3ae_Serial] Consistent treatment of sensitivities and ma rgins
Hi all,
I little bit of history here is necessary to understand why the stressed
receiver sensitivity was introduced. During the development of the 802.3z
standard (Gigabit Ethernet) it became apparent that some receivers had a
very low bandwidth, although their sensitivity was OK. So, when combined
with multimode links that had already low bandwidth relative to the baud
rate, the overall system bandwidth was low and that resulted in extra DJ
and lower sensitivity than expected.
So, to answer your question, if the receiver has sufficient bandwidth, the
stressed receiver sensitivity is not necessary, but if it doesn't have
sufficient bandwidth, the stressed sensitivity measurement would flush it.
So, if people think that this is not a problem, especially for the 1300 and
1550 nm PMDs, we should consider eliminating the stressed rece9iver
sensitivity and making the nominal receiver sensitivity normative. The
850nm case has to be considered more carefully.
Peter
Petar Pepeljugoski
IBM Research
P.O.Box 218
Yorktown Heights, NY 10598
e-mail: petarp@xxxxxxxxxx
phone: (914)-945-3761
fax: (914)-945-4134
"DAWE,PIERS (A-England,ex1)" <piers_dawe@xxxxxxxxxxx>@majordomo.ieee.org on
10/12/2001 04:50:13 AM
Sent by: owner-stds-802-3-hssg-serialpmd@xxxxxxxxxxxxxxxxxx
To: "'Rohit Mittal'" <RMittal@xxxxxxx>, "'802.3ae Serial'"
<stds-802-3-hssg-serialpmd@xxxxxxxx>
cc:
Subject: RE: [802.3ae_Serial] Consistent treatment of sensitivities and ma
rgins
Rohit,
The informative sensitivity is what you would get if you used a VERY GOOD
transmitter (and in this standard, it's BER = 10^-12). Stressed Rx
sensitivity is what we think would happen in real life with the worst legal
transmitter and the fiber.
You say:
"practically speaking, has anyone seen a system fail the stressed
sensitivity while passing the "informative" one"
That's a very relevant thought while we are reviewing the numbers in the
draft standard. Can you quantify it?
Thank you,
Piers
-----Original Message-----
From: Rohit Mittal [mailto:RMittal@xxxxxxx]
Sent: 12 October 2001 00:58
To: '802.3ae Serial'
Subject: RE: [802.3ae_Serial] Consistent treatment of sensitivities and ma
rgins
Sorry, I am a little confused here.
Informative sensitivity is the sensitivity I get when I take my Tx output,
put a VOA and put it into Rx. I then vary the VOA till I start seeing 1e-10
errors in the Rx. The optical power going into the Rx at that point is the
informative sentivity. Is this correct??
Stressed Rx sensitivity is what happens in real life when the Tx actually
goes through fiber so you have ISI etc. and so the sensitivity is lessend
by
ISI power penalty etc. Is this correct??
Supposing, both of my statements are correct, I would still think there is
value in informative sensitivity since it is a very simple test for an
equipment vendor to do when they qualify a vendor. I know that is not a
great thing for real life situations but practically speaking, has anyone
seen a system fail the stressed sensitivity while passing the "informative"
one
thanks
Rohit
-----Original Message-----
From: DAWE,PIERS (A-England,ex1) [ mailto:piers_dawe@xxxxxxxxxxx
<mailto:piers_dawe@xxxxxxxxxxx> ]
Sent: Thursday, October 11, 2001 1:44 PM
To: 'Rahn, Juergen (Juergen)'; '802.3ae Serial'
Cc: 'Tom Lindsay'
Subject: RE: [802.3ae_Serial] Consistent treatment of sensitivities and
ma rgins
Juergen,
As far as I can see, TX power, Stressed sensitivity, and Loss will be
normative.
Tables 52-10, 52-15 and 52-19 are informative - as shown by the subclause
headings. In these tables:
I think Budget is not normative, as it's not a measurable.
Operating distance is not normative, but operating range (different tables)
is, I think.
Channel insertion loss is probably not normative in these tables but is
normative in Table 52-26, I think.
Allocation for penalties is not normative. There is no separate margin any
more.
I don't know if this clarifies, I thought I would write it out in case
anyone else saw things more clearly.
As to whether the informative sensitivity is useful enough to be included,
it may be worth carrying through the draft revision process to remind us to
actually build this stressed eye and see if the stressed and unstressed
sensitivities do in practice relate as we expect. We ought to know this
before we freeze the standard.
Piers
> -----Original Message-----
> From: Rahn, Juergen (Juergen) [ mailto:krahn@xxxxxxxxxx
<mailto:krahn@xxxxxxxxxx> ]
> Sent: 11 October 2001 12:40
> To: '802.3ae Serial'; 'DAWE,PIERS (A-England,ex1)'
> Cc: 'Tom Lindsay'
> Subject: AW: [802.3ae_Serial] Consistent treatment of
> sensitivities and ma rgins
>
> Hi,
> Just for confirmation what will be specified now as normative
> visible values
> in the tables. Are we now at:
> TX power, Stressed sensitivity, Loss, Total sum out of
> remaining impairments
> (penalty) and margin.
> I read: Stressed Rx sensitivity = Tx power - Losses -
> impairments
> not included by
> > the shape of the stressful test eye - Margin
> in your description. Is this now what will be indicated in
> the tables as the
> requirements?
> I would like to make the point ,that the informative and normative
> parameters should be clearly separated. As the informative
> sensitivity is a
> theoretical value that is not possible to be verified , I am severely
> doubting of this is useful to be included. I at least have
> seen no receiver
> model allowing such kind of calculation showing the required
> accuracy in
> agreement with measurements . However there should be
> different receivers
> possible providing the target sensitivity in real life but
> would all require
> different modeling. So what does this bring for the specification.
> Regards Juergen
>
>
>
> > ----------
> > Von: DAWE,PIERS (A-England,ex1)[SMTP:piers_dawe@xxxxxxxxxxx]
> > Gesendet: Mittwoch, 10. Oktober 2001 20:10
> > An: '802.3ae Serial'
> > Cc: 'Tom Lindsay'
> > Betreff: [802.3ae_Serial] Consistent treatment of
> sensitivities and
> > margins
> >
> >
> > I had an action with Tom Lindsay to document the proposed consistent
> > treatment of sensitivities and margins which has gained
> consensus on the
> > serial PMD conference calls. Here it is:
> >
> > First, a description of where we are
> > ------------------------------------
> > Apart from the errors, the position in draft 3.2 and link
> model 2.4.1 is:
> >
> > Budget = Tx power - Informative Rx sensitivity
> >
> > Budget = Impairments + Losses + Margin
> >
> > Where informative Rx sensitivity is also known as "nominal" or
> > "unstressed" sensitivity, and here "Impairments" are
> penalties apart from
> > (broadband) optical attenuation or loss.
> >
> > Also draft 3.2 and link model 2.4.1 have (slightly simplified):
> >
> > Stressed Rx sensitivity = Tx power - Losses
> >
> > The principle is that the impairments are recreated by creating the
> > stressful test eye. However, we noticed that margin was treated
> > differently
> > in the stressed and nominal sensitivities. Also,
> impairments that are not
> > recreated by the test eye are effectively being ignored in
> the stressed Rx
> > sensitivity calculation.
> >
> > Second, a description of where we think we should be
> > ----------------------------------------------------
> >
> > Stressed Rx sensitivity = Informative Rx sensitivity + impairments
> > included
> > by the shape of the stressful test eye
> >
> > and also
> >
> > Stressed Rx sensitivity = Tx power - Losses - impairments
> not included by
> > the shape of the stressful test eye - Margin
> >
> > Subtracting the first equation from the second we get
> >
> > Informative Rx sensitivity = Tx power - Losses - all
> impairments - Margin
> >
> > as before.
> >
> > To simplify this description, questions of measurement at
> TP3 versus TP4,
> > mean power versus OMA, and triple trade offs, are not
> mentioned. They are
> > orthogonal questions to this one, which is what do we want
> stressed Rx
> > sensitivity to mean?
> >
> > Now we had to split the impairments into two categories.
> >
> > Impairments which are included by the shape of the
> stressful test eye:
> > Inter symbol interference penalty P_ISI
> > Deterministic jitter penalty P_DJ
> > The part of Rx baseline wander which is exacerbated by
> the shape of
> > the stressful test eye
> >
> > This class are all pattern dependent penalties. The first two items
> > combine
> > to the "vertical eye closure penalty" in D3.2 52.9.13
> Conformance test
> > signal at TP3 for receiver testing.
> >
> > Impairments which are not included by the shape of the
> stressful test eye:
> > Modal noise
> > Reflection noise
> > Mode partition noise
> > Relative intensity noise
> > Anticipated Tx baseline wander
> >
> > The second class are noise-like, mainly non-pattern
> dependent penalties.
> >
> > The proposed link model 10GEPBud3_1_14.xls implements
> nearly all of this.
> > The fine detail of the baseline wander (in Pcross) is
> simplified and we
> > have
> > found an error in the Rx stressed OMA column: it should not
> contain Pmn.
> > This error does not affect anything but the stressed
> sensitivities in
> > cases
> > with multimode fiber.
> >
> > In the draft standard there is no line item called "Margin"
> any more. In
> > general we have spent most of it on penalties. But it was
> worth going
> > through this exercise to account properly for all the
> noise-like terms in
> > the stressed sensitivity test which is now our only
> normative receiver
> > sensitivity criterion.
> >
> > Piers
> >
>