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

[802.3ae_Serial] Jitter measurements and calibration




Dear all,

I have been thinking about the jitter measurements we have in the draft
and I have some concerns about the possibility to calibrate measurement
equipment to the accuracy needed to actually be able to do meaningful
measurements. The concern is basically with TX characterization which
also has impact on the RX side measurements through calibration of the
stressed eye.

For the TX side measurement, the receiver (O/E all the way to the
decision) shall:

  (1) have a 7.5GHz Bessel filter characteristics
  (2) have negligable noise
  (3) have a BERT with have zero sample and hold time
  (4) have a clock recovery with negligable jitter generation

Maybe you could see item (3) as part of the filter response, but I do
not think that you can.

I don't know of any components that achieve these requirements at the
same time, so my best way forward is to characterize the error sources
that you have in your measurement system and calibrate these out.

(1) I guess you could measure the frequency response of your receiver
circuits (not only O/E, but also error detector response) and check
whether they follow the required response curve. If it does not not you
would need to compensate for the extra jitter introduced, which will be
pattern dependent. This could be possible although not simple, and it is
hard to really know you did it right.

(2) No-noise receivers are hard to get. To make this measurement, you
would need to get close to the noiseless case and/or extrapolate the
measurement to estimate where the no-noise curve would be. Probably
doable, but this extrapolation does depend on the eye shape and overload
behaviour of the receiver, and probably has to make assumptions on the
jitter distribution.

(3) BERTs do not have 0 sample-and-hold time. Again I think it is
possible to characterize fairly good for simple patterns. However, it
could be hard to separate from (1) and might not add up in a simple way
together with (1).

(4) This could be measured and calibrated out. Gets complicated if clock
recovery output depends on the input pattern and signal quality.

If we had one single jitter measurement reference (like the kilo
prototype which used to be in Paris) we would not have a problem because
everybody sees the same "errors". That is not our reality however, and
my fear is that these things make the jitter measurement results
somewhat hard to interpret and reproduce to the level we require (at
least <5ps is my guess). I do not see anything about this issue in the
Fibre Channel documents (MJS 1 & 2), and do not know if they ever tried
this.

However, I would be quite happy if you or someone else explained to me
that I am wrong and should not worry that much.

Best regards,
			Peter

================================================================
We have moved to a new location: see http://www.optillion.com/moving_on/

Optillion - High speed optical tranceivers
Peter Öhlen		   Phone: +46 8 5272 5205
Box 47646		   MPh:   +46 70 181 5205
S-117 94 Stockholm	   Fax:   +46 8 5272 5400
http://www.optillion.com   mailto:peter.ohlen@xxxxxxxxxxxxx