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

[BP] Partial analysis of simulation results




-- 
|--------------------------------------------------------------------|
|       Charles Moore 
|       Avago Technologies
|       Image Solutions Division
|       charles.moore@avagotech.com
|       (970) 288-4561
|--------------------------------------------------------------------|

guys,

    I suggested 3 crosstalk models for simulation.  I selected them to 
demonstrate differing spectral content of crosstalk.  Joe Abler has 
simulated them.  Here is an excerpt of his result plus my calculations 
of couple of measures of crosstalk:
     
     crosstalk model                            EYE with  ICR     Power
                                               31mUI DCD  margin  ratio

Molex/1m_OUTBOUND_FEXT/sj5k5g4h4_SPARS.s4p	 34mUI	 4.57dB   27.94dB
Tyco/Case5/Case5DS1310N2D13L6.s4p         	168mUI	 1.79dB   33.94dB
Intel/peters_01_0605_T1_next5.s4p         	183mUI	15.26dB   33.69dB
No crosstalk                                    185mUI  infinite  infinite

ICR margin is the margin by which the channel would pass ICR spec if this 
crosstalk model were combined with the Thru channel joe used (Agilent ITTC23).

Power ratio is the signal power divided by crosstalk power, done by in
a way similar to the one suggested by Rich Mellitz.  I describe the method
a bit more in the appendix.

The sj5kg4h4 model nearly fails with 31mUI DCD and little or no self 
interference.  Joe says that he considers it to have failed.

The Case5_N2 model and the T1_next5 models both pass with some margin
for additional self interference and possibly DCD.

ICR margin would not very useful in determining whether there is too much
crosstalk. The Case5D_N2 model works approximately as well as the 
T1_next5 model but has more than 13dB worse ICR margin, and it is almost
3dB worse in ICR margin than the sj5kg4h4 model but has significantly 
better performance.

On the other hand, the 2 passing cases have nearly equal power ratios and
the failing case is 6dB worse.

So i suggest that we use power ratio to specify crosstalk and use a value
of 33dB as the minimum.

                            charles

Appendix:

if 
xtalk = crosstalk attenuation in dB  = IL
atten = thru attenuation in dB       = PSXT
sinc  = sin(pi*fTt0)/(pi*f*T0)
T0    = 1/(signaling speed)

define:
  
channelGain=integ(2*t0*sinc**2*(10**(-atten/10)))
XtalkGain=integ(2*t0*sinc**2*(10**(-xtalk/10)))
Power ratio=10*log10(channelGain/XtalkGain)