Re: [802.3_100GCU] measuring odd-even jitter

[Charles Moore <charles.moore@xxxxxxxxxxxxx>]

ed,

  No i have no reference to "phase noise amplification", it is a term i 
heard used back in 2004 to refer to the fact that odd-even jitter seemed 
to grow with channel loss.  I never liked the term and if it is no 
longer used that is good.  I prefer to look at the special problem with 
odd-even jitter as being due to the fact that odd-even jitter was is 
modulation at (baud rate)/2.  This put sidebands on all the data 
spectral components, one of which got pushed up to higher frequency 
where it got attenuated away, and one got pushed to low frequency where 
it was not attenuated and cause baseline wander.  The baseline wander 
caused shifts in zero crossings, called jitter.  If you look at a simple 
alternating 10 pattern, dc coupled, you will see how a small amount of 
odd-even jitter will cause a small baseline shift.  After a long 
channel, the baseline shift will still be there but the 10 pattern will 
be attenuated into a small sine wave which might not even cross zero due 
the baseline shift.

                                                      charles

|--------------------------------------------------------------------|
|       Charles Moore 
|       Avago Technologies
|       APD
|       charles.moore@xxxxxxxxxxxxx
|       (970) 288-4561
|--------------------------------------------------------------------|



Ed Sayre wrote:
> Charles:
>
> Do you have a reference to "phase noise amplification"?
>
> Ed
>
> Dr. Edward P. Sayre, P. E.
> North East Systems Asscoiates, Inc.
> 9 Maple Lane
> PO Box 807
> Marshfield, MA  02050  USA
> [T]: +1-781-837-9088
> [C]: +1-978-314-4940
> [E]: esayre@xxxxxxxx
>
> -----Original Message-----
> From: Charles Moore [mailto:charles.moore@xxxxxxxxxxxxx] 
> Sent: Wednesday, August 08, 2012 4:18 PM
> To: STDS-802-3-100GCU@xxxxxxxxxxxxxxxxx
> Subject: [802.3_100GCU] measuring odd-even jitter
>
> guys,
>
>    All of the standard PRBS patterns have odd lengths.  If the pattern 
> twice, every possible transition will occur twice, once as an even 
> numbered transition once as an odd numbered transition.  If you measure 
> the average jitter (deviation from some ideal, equally spaced transition
>
> time) on all even numbered transitions and the average jitter on all odd
>
> numbered transitions the difference, divided by 2, will be the odd-even 
> jitter.  On average the odd and even transitions will suffer the same 
> ISI shift and error due to improperly chosen threshold so the difference
>
> will be free from these effects.  This gives a very clean way of 
> measuring odd-even jitter.
>
>    One error will occur if there is substantial loss at Nyquist due to 
> the mystically named "phase noise amplification", which will expand 
> (amplify) odd-even jitter but it will not create it.
>
>                                charles
>
>