Couple of thoughts on the notes sent
earlier on this topic:
1. I thought that the 1 percentile case
applies to one single parameter at its worst case. However, when you have
multiple connectors, since they are independent, you have to treat them
as multiple parameters, and therefore have to use smaller percentiles.
Let say you have transmitter lens (with one loss) and receiver lens with
another loss. If you relax the parameters on these, their worst case values
have to be at 0.5 percentile each. We can apply the same to the connectors,
since each has its own impact. So, you need to tighten the connector
specification to achieve the overall 1 perceintile loss. Since we can't
do this - the connectors are already installed and individually specified
at 0.75dB worst case (3sigma), the rest of the link parameters need to
be tightened to compensate.
2. I believe that in the link connector
loss calculations for Ethernet we used an average value and standard deviation.
So, the typical link had 4 connectors, for which we added the mean values
linearly and the std deviations as square root of sum of the squares.
3. Connector loss degrades with number
of matings. Since many of the legacy installations are old, it is reasonable
to expect the mean and std. dev of the connector loss to have shifted upwards.
4. I believe that the OFL connector
loss measurements do not take into account the mismatch in size of the
P.O.Box 218 (mail)
1101 Kitchawan Road, Rte. 134 (shipping)
Yorktown Heights, NY 10598
David Cunningham <david_cunningham@AGILENT.COM> Sent by: email@example.com
08/09/2004 10:25 AM
Please respond to
"IEEE 802.3aq 10GBASE-LRM"
[10GMMF] 62MMF Worst Case
Connector Offset Calculations
I have attached a presentation which analyzes the connector offset based
on the legacy Ethernet and building cabling standards maximum connector
loss specifications and worst-case link configurations.
Both Ethernet and the building cabling standards limit the maximum per
connector loss with overfilled launch (OFL) to 0.75 dB. Furthermore, the
worst-case cable model for Gigabit Ethernet is one with two 0.75 dB connectors
and, for LX, offset launch jumper cables at each end (in place of normal
I have therefore calculated the loss versus offset for 62 MMF with OFL.
The allowable worst-case offset is then the offset which produces
0.75 dB of loss - this occurs at approximately 6 um.
The main conclusions are:
1) For 62 MMF, if all the loss is attributed to lateral offset, the worst
case connector offset is approximately 6 um for a loss of 0.75 dB(OFL).
2) For 62 MMF, the worst case link for modal noise or channel variance
measurements is one with two connectors each having 6 um offset equivalent
to 0.75 dB (OFL) loss. This is because modal variation effects are
larger when there is a small number of higher loss components in the optical
3) If three connectors are used the maximum loss (OFL) per connector must
be reduced to 0.5 dB which is equivalent to a 4 um lateral offset per connection.
Interestingly, if we use the Rayleigh distribution with a mean of 3.58
um, as proposed by Joerg, the probability of a 6 um offset is about 0.1
and the probability of two 6 um offset connectors is about 0.01. If the
Rayleigh distribution is correct this means that we are investigating the
1 percentile case which is consistent with the Ethernet methodology. This
does not mean that I have confirmed that the proposed distributions of
connector offsets is correct - rather I'm just noting that for the 62 MMF
case there does seem to be agreement between the two methods on this point.
I also have placed the presentation on the FTP site per Jonathan King's
[attachment "Worst Connector Offset Calculations.pdf" deleted
by Petar Pepeljugoski/Watson/IBM]