RE: [802.3ae_Serial] FW: DJ in fiber
Attached is a freq response based on 45 psec delay gives -6 dB electrical at
about 7.3 GHz (appears higher than the 6.06 GHz you mention below) . 3 dB
electrical at about 5.5 Ghz.
Interesting to me that the eye closure penalty for this filter is less than
1 dB which seems low when compared to the
> 3dB stressed eye closure that is used to test the receiver.
So with this in mind , does this transversal filter adequately generate the
amount of vertical eye closure desired (ignoring jitter aspects).
Thanks
-----Original Message-----
From: Mike Dudek [mailto:mdudek@xxxxxxxxxxxx]
Sent: Monday, March 04, 2002 7:07 PM
To: Russ Patterson
Cc: 'Adam Healey'; Lindsay, Tom; David Kabal; Serial PMD Ad Hoc
Reflector (E-mail) (E-mail); Peter Öhlén
Subject: Re: [802.3ae_Serial] FW: DJ in fiber
This is in response to a private question from Piers on this topic and
another
related one.
1. The original intent of the transversal filter was to simulate the
worst
case DMD of a fiber. In my opinion the fiber does not create Dj it has DMD
which with non-ideal transmitter and receivers creates additional Dj. (I
suggest you try simulating with a source waveform from a screen grab of a
real
waveform that has ringing, overshoot, etc. that just passes the eye mask as
input to the transversal filter. I'd be very surprised if the Dj isn't
different (and probably significantly worse) with the transversal filter.
If
it is the same can someone explain to me how the fiber creates Dj).
Incidentally the delay in the filter was chosen so that it had the worst
case
fiber bandwidth of 6.06GHz given by 66 meters of 400MHz.Km fiber. (All other
fiber/ distance combinations have greater or equal bandwidth. I'm running a
check that this is still the correct delay as it may have been calculated
before we reduced the distances.)
The transversal filter can be replaced by a lumped element filter provided
this
filter matches both amplitude and group delay characteristics of the
transversal filter. (filter theory says that these are totally equivalent.)
However a Bessel Thompson filter with the amplitude characteristic is too
kind
in it's group delay characteristic. (I think you will find that to match
the
transversal filter properly a non-minimum phase filter network is needed
which
will complicate the lumped element filter implementation.).
2. The only reason why we would want the eye diagram for 850nm to be at
"virtual TP3" that I can think of would have been with an eye mask
appropriate
for the stressed Rx test combined with a removal of the requirement for rise
and fall time in the spec. Otherwise I agree with Piers that it should be
at
TP2.
Russ Patterson wrote:
> When I generated this eye, I was using a microwave simulator employing
a
> "real life" non ideal RF splitter and combiner in the model. The
inadequate
> isolation of these units resulted in a multiple reflection generating the
> non ideal response shown.
>
> When I increase the isolation of the two paths, I get the ideal response
> that Tom shows.
>
> -----Original Message-----
> From: Adam Healey [mailto:ahealey@xxxxxxxxx]
> Sent: Friday, March 01, 2002 7:52 AM
> To: Lindsay, Tom; Russ Patterson; David Kabal; Serial PMD Ad Hoc
> Reflector (E-mail) (E-mail)
> Cc: Peter Öhlén
> Subject: RE: [802.3ae_Serial] FW: DJ in fiber
>
> Tom,
>
> I believe that if you set your delay paths to have unequal gains (45/55 or
> 40/60), you will see "two steps" emerge.
>
> Best Regards,
> -Adam
>
> > -----Original Message-----
> > From: owner-stds-802-3-hssg-serialpmd@xxxxxxxxxxxxxxxxxx
> > [mailto:owner-stds-802-3-hssg-serialpmd@xxxxxxxxxxxxxxxxxx]On Behalf Of
> > Lindsay, Tom
> > Sent: Thursday, February 28, 2002 5:23 PM
> > To: Russ Patterson; David Kabal; Serial PMD Ad Hoc Reflector (E-mail)
> > (E-mail)
> > Cc: Peter Öhlén
> > Subject: RE: [802.3ae_Serial] FW: DJ in fiber
> >
> >
> > Russ - this picture is not the same as I am seeing. Your's has two steps
> > on the way up (or down), whereas I believe there should only be one. To
> > get two steps, you would need at least 2 relative delay paths (at least
> > 3 total).
> >
> > My sims on this are attached.
> >
> > Tom
> >
> > -----Original Message-----
> > From: Russ Patterson [mailto:Russ.Patterson@xxxxxxxxxxxxx]
> > Sent: Thursday, February 28, 2002 1:14 PM
> > To: Lindsay, Tom; David Kabal; Serial PMD Ad Hoc Reflector (E-mail)
> > (E-mail)
> > Cc: Russ Patterson; Peter Öhlén
> > Subject: RE: [802.3ae_Serial] FW: DJ in fiber
> >
> >
> > The attached PDF file illustrates what Tom L is saying. The unfiltered
> > response results in a nasty pedastal close to the zero crossing.
> > With
> > the added filtering of the 7.46 GHz BT filter, the pedastal becomes less
> > pronounced but the distortion/skew is still evident.
> >
> > There also does appear to be a small amount of jitter caused by the skew
> > and
> > the filtering (about 0.03 UI) in this example
> >
> >
> > Please Note; I am not yet subscribed to the ad hoc reflector
> >
> >
> > Thanks
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > -----Original Message-----
> > From: Lindsay, Tom [mailto:tlindsay@xxxxxxxxxxxxxxxxxxxx]
> > Sent: Thursday, February 28, 2002 12:15 PM
> > To: David Kabal; Serial PMD Ad Hoc Reflector (E-mail) (E-mail)
> > Cc: Russ Patterson; Peter Öhlén
> > Subject: RE: [802.3ae_Serial] FW: DJ in fiber
> >
> >
> > I'll throw something out here.
> >
> > I was not involved in the DMD work for 802.3z and missed much of the
> > discussions. My understanding, however, is that the transversal filter
> > intends to emulate a power split and delay (DMD) between two dominant
> > propagation groups in the fiber.
> >
> > It does not directly cause jitter, but it can certainly cause problems
> > in a receiver with too much bandwidth because the signal "pauses" around
> > the receiver threshold. I believe this is why the max Rx bandwidth spec
> > was put in place.
> >
> > Now, given that our measurement systems should all have Bessel filters
> > at 7.5 GHz, and given other edge rate limitations, the transversal
> > filter does not have too much impact at threshold, but it still does
> > cause general distortion and some vertical eye closure, and so provides
> > some representative stress.
> >
> > Should we do something in addition or instead to add jitter? Good
> > question. It seems that jitter would depend on the asymmetry of the
> > impulse response due again to differing propagation delays through the
> > fiber. The impulse response would be the summation of all the powers and
> > (relative) delays of the mode groups. The delays are determined
> > primarily by the fiber grading; how the powers are divided is a function
> > of launch conditions. In general, if the impulse response is
> > symmetrical, the jitter will be minimal unless ISI becomes fairly
> > severe. If asymmtrical, then jitter will be worse. I have no idea how to
> > predict that in the general case. Anyone else?
> >
> > Tom
> >
> >
> > -----Original Message-----
> > From: David Kabal [mailto:David.Kabal@xxxxxxxxxxxxx]
> > Sent: Thursday, February 28, 2002 8:54 AM
> > To: Serial PMD Ad Hoc Reflector (E-mail) (E-mail)
> > Cc: Russ Patterson; Peter Öhlén
> > Subject: [802.3ae_Serial] FW: DJ in fiber
> >
> >
> >
> > I am forwarding this discussion to the reflector because this seems to
> > be
> > relevant to all.
> >
> > The discussion was between Russ Patterson of Picolight and Peter Öhlén
> > of
> > Optillion, concerning the 10GBASE-S transversal filter used for TDP
> > measurement, but has general applicability to jitter measurements in the
> > entire standard.
> >
> > Please review and comment to the reflector.
> >
> > Cheers,
> > Dave
> >
> > ------
> > David Kabal
> > Sr. Technology Planner
> > Picolight
> >
> > e-mail: david.kabal@xxxxxxxxxxxxx
> > Main: 303-530-3189 x7498
> > Direct: 303-527-7498
> > Fax: 303-527-4961
> >
> >
> > -----Original Message-----
> > From: Peter Öhlén [mailto:Peter.Ohlen@xxxxxxxxxxxxx]
> > Sent: Thursday, February 28, 2002 1:17 AM
> > To: Russ Patterson
> > Cc: David Kabal
> > Subject: RE: DJ in fiber
> >
> > Russ,
> >
> > I agree with Dave, that this is a great discussion for the reflector,
> > but I will just give a quick answer:
> >
> > It is correct that there is no explicit allocation for jitter from the
> > fiber. Now, the rationale for the transversal filter may not be quite
> > clear from the text (which has not really changed in intention). Still,
> > the intention was that that filter would simulate a worst case channel.
> >
> > Regards,
> > Peter
> >
> > > -----Original Message-----
> > > From: Russ Patterson [mailto:Russ.Patterson@xxxxxxxxxxxxx]
> > > Sent: den 28 februari 2002 00:40
> > > To: Peter Öhlén
> > > Cc: David Kabal
> > > Subject: RE: DJ in fiber
> > >
> > >
> > > I thought some more about what you said about jitter. I looked in some
> > > earlier versions of clause 52 (D3.1 with the bathtubs) and
> > > it looks like
> > > there is no Deterministic jitter what so ever allocated to
> > > the fiber for
> > > base SR.
> > >
> > >
> > > In D3.1, the receiver needed to meet the same waterfall
> > > mask (figure 52-6)
> > > as the transmitter (outside for TX, middle for RX). The transmitter
> > > waterfall was to be tested with < 5 meters of fiber.
> > > Therefore no allocation
> > > for fiber jitter.
> > >
> > >
> > > So it looks to me as it there was no deterministic jitter originally
> > > intended for the fiber , only ISI was allocated.
> > >
> > >
> > > Has this requirement changed in D4.1? If so, sounds like it
> > > clearly needs
> > > to be quantized somewhere in the clause. ie "the worse case
> > > fiber will be
> > > assumed to add TBD UI of deterministic jitter". And then
> > > we would need
> > > to construct a filter to generate the desired amount of
> > > closure and jitter.
> > >
> > >
> > > Russ P
> > >
> > >
> > >
> > > -----Original Message-----
> > > From: Peter Öhlén [mailto:Peter.Ohlen@xxxxxxxxxxxxx]
> > > Sent: Wednesday, February 27, 2002 4:38 AM
> > > To: Russ Patterson
> > > Cc: David Kabal
> > > Subject: RE: Need for Transversal filter
> > >
> > >
> > > Russ,
> > >
> > > I see your point, and this is important input. To me it seems like the
> > > transversal filter does not do the job it is supposed to, which is bad
> > > news. It is clear that something else is needed.
> > >
> > > Many think that jitter is quite important to control, and to use a
> > > filter optimized for flat group delay (low jitter) as an
> > > artefact for a
> > > fiber which would introduce jitter would probably require _very_ solid
> > > arguments. As far as I understand, there is no such thing as a
> > > worst-case multi-mode fiber as the actual bandwidth depends on the
> > > coupling between the TX and the fiber, which is the reason for the
> > > electrical filter.
> > >
> > > /Peter
> > >
> > > > -----Original Message-----
> > > > From: Russ Patterson [mailto:Russ.Patterson@xxxxxxxxxxxxx]
> > > > Sent: den 26 februari 2002 16:52
> > > > To: Peter Öhlén
> > > > Cc: David Kabal
> > > > Subject: RE: Need for Transversal filter
> > > >
> > > >
> > > > Here is a result of a sim I did on the filter. Although the
> > > > eye does not
> > > > appear symmetrical in the time domain I do not see any
> > > > significant amount
> > > > of added deterministic jitter at the average level . So
> > > > where is the added
> > > > jitter that you speak of?
> > > >
> > > >
> > > >
> > > > Thanks
> > > >
> > > >
> > > >
> > > >
> > > > -----Original Message-----
> > > > From: Peter Öhlén [mailto:Peter.Ohlen@xxxxxxxxxxxxx]
> > > > Sent: Tuesday, February 26, 2002 7:49 AM
> > > > To: Russ Patterson
> > > > Cc: David Kabal
> > > > Subject: RE: Need for Transversal filter
> > > >
> > > >
> > > > Russ,
> > > >
> > > > The Bessel-type filters or similar (like the Picosecond ones)
> > > > are really
> > > > designed to have very low jitter, and the purpose of the test is to
> > > > simulate worst-case fiber, including addition of jitter. Therefore I
> > > > don't think it is a good idea to replace the transversal
> > > filter with a
> > > > Bessel filter.
> > > >
> > > > Best regards,
> > > > Peter
> > > >
> > > > > -----Original Message-----
> > > > > From: Russ Patterson [mailto:Russ.Patterson@xxxxxxxxxxxxx]
> > > > > Sent: den 22 februari 2002 01:17
> > > > > To: Peter Öhlén
> > > > > Cc: David Kabal
> > > > > Subject: Need for Transversal filter
> > > > >
> > > > >
> > > > >
> > > > > Peter:
> > > > >
> > > > > re 52.9.12.3
> > > > >
> > > > >
> > > > > is it really necessary that we construct a transversal type
> > > > filter to
> > > > > simulate eye closure? A simulation of the filter shows I can
> > > > > accomplish
> > > > > much the same ISI effect using a linear phase nth order roll
> > > > > off filter.
> > > > >
> > > > > The Transversal delay line filter repeats in the frequency
> > > > > domain with a
> > > > > period close to 10 GHz , however the 7.45 GHz cutoff of
> > > > the receiver
> > > > > essentially removes all but the baseband response.
> > > > >
> > > > >
> > > > > For production test purposes I feel it is simpler to have a
> > > > > picosecond
> > > > > pulse labs type filter rather than try and construct a
> > > > > transversal filter
> > > > > from splitters and combiners, line stretchers etc.
> > > > >
> > > > >
> > > > > Could we modify the text in 52.9.12.3 to say
> > > > >
> > > > > "transversal filter or equivalent lumped element filter to
> > > > > simulate the ISI
> > > > > created by two equal amplitude paths with a differential
> > > > > delay of xx psec
> > > > > where xx is adjusted to obtain the desired amount of ISI"
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > thanks
> > > > >
> > > > >
> > > > >
> > > > > Russ Patterson
> > > > > Opto Electronic Design Engineer
> > > > > Phone: 303-530-3189 ext 7447
> > > > > FAX 303-527-4968
> > > > >
> > > > > Picolight Inc
> > > > > 4665 Nautilus Court
> > > > > Boulder, Colorado
> > > > > 80301
> > > > > russ.patterson@xxxxxxxxxxxxx
> > > > >
> > > > >
> > > >
> > > >
> > >
> >
> >
TFILTER.pdf