[802.3_100GCU] D1.0

["Charles E. Moore" <charles.moore@xxxxxxxxxxxxx>]

rich,

    I have reviewed your presentation.  On the whole i like it.  I do 
have some questions and some places where i think improvement is possible:

in slide 3, Table 93A-2:

    1.   what is the difference between IL fit at fs and IL at fs?
    2.   Do we need both?
    3.   Do we need Amax/IL ?  How is is used in the spec?
    4.   Do we need ICR?  Is it used in computing MOM?
    5.   Do we need ILD?
    6.   Do we need ILDrms
    7.   How are we going to use PISI?  will we sum it with PMXI?
         Won't this be over conservative?
    8.   Similar questions about PMFEXTI and PMNEXTI.

in Slide 5,
    9.   If s-j*omega

             Gamma= Gamma0 + s/(2(pi*1.25*fb)/(1 + s/(2(pi*1.25*fb)
Slides 4 and 6
   10.   I think we need a package model to deal with the phase of 
Gamma_r and
         Gamma_t.  I suggest:

          A model consisting of a capacitor and a resistor to ground at the
          chip end, transfer channel with phase and log gain both varying
          linearly with frequency (like an a2 only fitting model) and 
another
          small capacitor to ground at the PC end.

          Everyone should know how to concatenate S parameters.  This 
gives us a
          just inside the chip to just inside the chip model with ideal 
50 Ohm
          source and load so H21=Sdd21.
Slide 6.

   11.    Once we start bringing in Tx filter, Rx filter etc, we are 
talking
          about H21 not Sdd21.
   12.    Need a recommendation on how to extrapolate to DC.  I 
recommend linear
          phase and amplitude extrapolation with phase forcing to 0 or 
pi at DC,
          but allow other methods if they are known by the user to be 
better.
   13.    Use dibit amplitude instead of 1UI after Tz.  Tz could be 
subject to
          artifacts due to negligible amounts of non-causality.  You 
get dibit by
          subtracting an SBR which has been delayed by 1UI.
   14.    Find noise at allowed BER.  Noise is found by convolution of:

          A.  SBR at UI intervals outside Ndfe*UI range.
          B.  For each FEXT channel the crosstalk weighted by CTLE 
gain, sinc
              function, Tx lowpass, Rx lowpass, and Tx amplitude, sampled
              at 1 UI intervals with worst case phase (within 0.1UI)
C.  For each FEXT channel the crosstalk weighted by CTLE gain, sinc
              function, Tx lowpass, Rx lowpass, and Tx amplitude, sampled
              at 1 UI intervals with worst case phase (within 0.1UI)

              SNR is dibit SBR divided by noise

                                         charles