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

RE: jitter methodology, 1550 nm TX chirp / fiber impairments on RX




Hi all

I add some clarification points just to shine more light on the topic.


Peter Olhen wrote: 

>For the 1550 nm case, TP3 would be the compliance point....
>....any two 40 km G.652 single-mode fibers are much more
>equal than any two transmitters.

I agree. This means that the fiber (a worst case regarding accumulated
dispersion on a 40 km SMF) is considered part of the TX (transmitter) in
order to cope with the chirp / dispersion interaction converting to jitter
(DCD, RJ, DJ) in the optical eye diagram and it's corresponding RX
sensitivity penalty.

Actually, as part of the test (one should also make a back-to-back
measurement), in order to do a reference measurement.  Also, this give a way
to seperate the impacts (penalties) when adding the fiber. 



>The TX is
>tested with a golden RX where sinusoidal jitter can be added to the
>recovered clock. Then the BER penalty with jitter is then measured.

The situations here is, that this is a TX module compliance test and one
does/may not have access to the TX PLL (where sinusoidal jitter could be
added.)

I am not quite sure how this 'golden' RX works or can be implemented for all
conditions. This may work only for analog PLL based clock extractors (ie.
non-linear signal procesing, mixer based PLLs). 

For BERT receivers, the clock phase can be adjusted manually (in order to
find the eye phase opening), but how can this be done with addition of
sinusoidal jitter?

 If one uses a CDR as the golden receiver, (in GIGA's CDR: a Bang-Bang phase
detector), the PLL may become very non-linear in it response to sinusoidal
(deterministic) jitter added to the RX, so the amount of jitter added may be
limited to some bounded value (< 0.4 UIpp, I guess).

I will try to do the jitter tolerance measurement on our 10 Gb/s CDR using
jitter injected at the RX VCO, and compare it to the convential test method
(adding sinusoidal jitter to the TX signal).



>For the RX test a similar approach is taken. A stressed eye is still used
to
>create DCD, but in addition sinusoidal jitter is added to the stressed
>signal. Here, the frequency of the sinusoidal jitter has to be higher than
>the maximum corner frequency of the PLL used in the RX (D.U.T.). The
>stressed sensitivity of the RX (D.U.T.) is then measured and compared to
the
>spec.

Is the 'stressed eye' generated from a 'golden' TX?. I suppose that this
measurement is made without fiber (< 20 m OK). So the RX is tested to the
specification (and against the golden RX, regarding non-stressed
sensitivity.)




>I deliberately left out any numbers. Instead I am hoping to get some
>comments on the methodology.

I think the methodology principles are OK, especially taking the fiber
effects to be part of the TX and its chirp characteristics. 

The balancing is to find the less restrictive specifications (for cost and
manufacture aspetcs) and still ensuring the interoperability for all
compliant TXs and RXs.

As I think it over, the problem is still to determine how the link power
margin should be divided and allocated to the TX+fiber vs. a realistic RX /
golden receiver, especially taking the added sinusoidal jitter (DJ) and it
coversion to penalties. 

If one allocates xx dB to the TX+fiber (for chirp / dispersion and RJ), then
stressing with yy UIpp sinusoidal jitter (DJ) at the golden RX adds
additional some zz(yy) dB penalty. 

So, the total of xx+zz(yy) dB should be less that the total margin set aside
for the RX stressed eye sensitivity specifications.


Next, what kind of degradation is expected from a realistic (cheap?) RX
compared to the golden RX? 

1) The absolute sensitivity (b-b), but that is not so important as long as
the sensitivity is in accordance with specs (<-18 dBm). If one has a better
(or worse) receiver, but it still performs well with the stressed eye
sensitivity (in terms of more tolerant to zz(yy) degradation, due to worse
(better) jitter tolerance - ie. the CDR setup/hold time) how will that trade
off? 

Or in other words: The absolute (unstressed) sensitivity may not be as
important as the stressed sensitivity.

2) the relative (DJ) penalties (ie. zz(yy)) depends on the CDR
characteristics (technology), as I presented on the Austin meeting
(comparing the tolerance of  total jitter for two CDRs: a 2.5 Gb/s (about
0.8 UIpp) and 10 Gb/s (<0.5 UIpp)). For a 1 dB optical penalty and suitable
SDH filters, the sinusoidal jitter (DJ) tolerance is drops to something like
<0.3 UIpp. 

This shows that the zz() functions also depends on the AM-PM noise
conversion due to finite system / signal bandwidth (limitations). As the
golden receiver already is specified? with som known characteristics (ie 4.
order BesselThomson), so it may already have been included in the reference
measurement.


So, the questions left are:

1) what realistic value for yy?

2) what zz() dB is realistic??


3) Does this 1+2) ensure interoperability?



Peter,

This reply actually became longer that I expected, because I reflected over
the different thoughts crossing my mind.

Is this kind of discussion over the reflector beneficial (burdening all
others with replies on replies etc.) or should take a private E-mail
correspondance to discuss details in the proposal? 

I think as long as the Subject is clearly specified its ok to go through the
reflector (so I added the '1550 nm TX ......').


Regards

Benny Christensen
--------------------------------------------
  llllllll   ii     llllllll     llllll
ll                ll           ll      ll
ll    llll   ll   ll    llll   llllllllll
ll      ll   ll   ll      ll   ll      ll
  llllllll   ll     llllllll   ll      ll

GIGA, an Intel company
Benny Christensen, M.Sc.E.E, Ph.D.
Mileparken 22, DK-2740 Skovlunde, Denmark
Tel: +45 7010 1062, Fax: +45 7010 1063
e-mail: benny.christensen@xxxxxxxxx, http://www.giga.dk