RE: Estimating the magnitude of PMD
Gair, Vipul, all,
This do be the spooky stuff. I am certainly speculating on the effects of
PMD and using that speculation to draft my opinions. Frankly, I don't know
how to model this. Historically, when we talk about jitter and 10e-12 BER
objectives, I BELEIVE that everyone believed as I do that there were few
burst errors. Or, that N bit errors in a row represented about 10e-12N
probability.
In the days of DMD, we saw a number of examples where the rising edge at the
optical Rx approached a "ledge." In the end, we decided that while it
approached a ledge, it wasn't and that we had few to no examples of the
near-ledge being exactly at the logic threshold.
>From what I understand about PMD, the probability is excellent that the
ledge would be just that, a flat shelf in the rising edge. The probability
that said ledge would be in the center of the rising edge, or, at the logic
threshold would be just that, a probability. For PMD, the question stands as
to whether when this occurs, if it occurs, it would linger there for a while
or not (N bits in a row). Or, just how would this be integrated into the
overall BER calculations.
So, while the overall BER calculations and operation might be correct, the
error rate for the link might, over a specific slot in time, in fact become
10e-12/N. Yes, I understand that this is an extreme representation of the
problem. I would love to have someone prove this wrong.
Oh, and by the way, if any of the above is correct, someone should be ready
to explain where the data dependency factor is in the equation (the reason
why the 10e-12/N is wrong!). To have a ledge on every bit, there has to be a
transition on every bit. If you start in a specific state, say a logic 0,
then for this to occur the pattern is known. But, the probability of that
pattern is quite small.... (e-N?) And, if the probability of making an error
on every ledge is 50%, then there is another factor of 2 somewhere in the
equation.... (0.5e-N?) But, of course it is not this simple.
jonathan
> -----Original Message-----
> From: Gair Brown [mailto:gdbrown@xxxxxxxxxxxxx]
> Sent: Friday, September 29, 2000 5:01 AM
> To: vipul.bhatt@xxxxxxxxxxx
> Cc: stds-802-3-hssg-equal@xxxxxxxx
> Subject: Re: Estimating the magnitude of PMD
>
>
>
> Vipul,
>
> My experience with PMD is limited and I rarely work with it. Your
> question on the magnitude would be best answered by some of the folks
> who do PMD for a living.
>
> A note on Johathon's comment. I don't believe that PMD should be
> accounted for like DCD. The PMD will be convolved with the
> rest of the
> link distortion and will simply (if anything about PMD is
> simple!) lead
> to additional broadening. It will add into the ISI rather
> than decrease
> the bit interval.
>
> Gair
>
>
>
> Vipul Bhatt wrote:
> >
> > Gair,
> >
> > And if there is no multiwavelength averaging, would you consider the
> > pulse spreading values (7 ps and 17.5 ps) described in my
> earlier email a
> > reasonable approximation?
> >
> > Thanks,
> > Vipul
> >
> > ===============
> >
> > > -----Original Message-----
> > > From: owner-stds-802-3-hssg-equal@xxxxxxxx
> > > [mailto:owner-stds-802-3-hssg-equal@xxxxxxxx]On Behalf Of
> Gair Brown
> > > Sent: Thursday, September 28, 2000 12:01 PM
> > > To: vipul.bhatt@xxxxxxxxxxx
> > > Cc: stds-802-3-hssg-equal@xxxxxxxx; HACKERTMJ@xxxxxxxxxxx;
> > > SwansonSE@xxxxxxxxxxx; piers_dawe@xxxxxxxxxxx
> > > Subject: Re: Estimating the magnitude of PMD
> > >
> > >
> > >
> > > Vipul,
> > >
> > > I'm not sure, but I believe you may still have an error in your
> > > argument. The spectral width of typical long haul
> sources at 1550 is
> > > extremely small. I don't believe the PMD response of this
> > > small band of
> > > wavelengths is really independent. So there will be no
> multiwavelength
> > > averaging.
> > >
> > > Gair
> > >
> > >
> > >
>
> --
> Naval Surface Warfare Center
> browngd@xxxxxxxxxxxxx
> Code B35 PH:
> 540-653-1579
> 17320 Dahlgren Road FAX:
> 540-653-8673
> Building 1500 Room 110A
> Dahlgren, VA 22448-5100
>