XAUI "de-convolution" of DJ and RJ
Folks - during the XAUI call 2 weeks ago (4/4/01), I took a couple of action
items. Here is the first one.
Annex A.3 in FC-PI, revision 11, document 01-025v2, provides a simple
2-point bathtub curve method for estimation of (RJ and) DJ and TJ. I believe
it is very useful, and it captures the intent of effective jitter. However,
A.3 could benefit from some changes that would make it more accurate.
1. The equation that currently reads Q_n = -qnorm(BER_n ,0,1) needs a
correction factor. It should read: Q_n = -qnorm(2/TD*BER_n ,0,1).
2. Insert the following after the sentence with the equation. "TD is the
transition density of the data pattern being used for the measurement. A
typical value for 8B10B of TD=0.6 may be used."
3. Modify the values in the last sentence of b) so that it reads "For 1E-5,
Q = 3.99 and for 1E-9, Q = 5.8. (Q = 6.87 for 1E-12)."
4. Equation 3 for TJ should use a multiplier of 13.73, not 14.
Two other documents regarding de-convolution have been posted in Fibre
Channel. These are 99-489v0 (Excel) and 99-488v0 (Mathcad), and they offer
better accuracy due to using a higher number of points (more than 2) and a
least-squares curve fit. However, they are more complex and may impose
implementation challenges.
All methods mentioned so far in this email use (inverse) error-function or
equivalent routines. As an alternative, the serial PMD group has proposed
an exponential function that shows excellent agreement with the error
function over the BER range of interest. See PDF-pages 425 and 426 of
revision P802_3ae_D3_0.pdf. I have not attempted to solve the equations for
DJ and RJ, but it should be straightfoward and generally more available than
using the qnorm function given in A.3, for example. Note - I get better
match results to the error function form if the value for A=-1, not
A=-1.152.
Thanks, Tom Lindsay
Stratos