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

[BP] Signaling Comparison Spreadsheet Input



Please find below some suggestions for the signaling comparison spreadsheet that I am forwarding on behalf of my colleagues at Agere Systems.  This is not to be interpreted as a position of the Task Force Chair.

 

Thank you,

-Adam

 

 

 

Crosstalk Aggressors:

  • Type:  Deterministic | Random
  • Peak Alignment:  Eye Center | Transition | (Other?), Note: does not apply to Type = Random
  • Scale Factor:  (Aggressor Amplitude)/(Launch Amplitude)

 

It is preferred that the group agree to one or another to reduce the number of simulations that we need to do.  For the basis of comparison, either one seems acceptable.  However, if the group cannot come to agreement on this, the underlying assumption should be stated.

 

TX Jitter

  • DJ or PJ (p-p)
  • RJ (rms)
  • DCD or Even/Odd distortion is also important, but not all simulation platforms may support it

 

This is a re-iteration of Joe Abler’s request.

 

RX Input Jitter

 

This should be removed.  The DJ/RJ at the receiver input is a function of the transmitter and channel and will be channel specific (this has been said before).  If this is intended to represent the DJ and RJ added by the timing recovery system, it is contrary to earlier discussion that timing recovery should not be included as part of the simulations.  The sampling point should be placed optimally in some algorithmic way.  The timing margin observed from these results would dictate the requisite performance of the timing recovery algorithm.  A minimum acceptable horizontal eye opening could be defined that would allow sufficient margin for realistic timing recovery schemes.

 

Similarly, earlier discussions also seemed to indicate that we would take an “idealized” view of the equalizer.  For example, the equalizer taps would have floating point precision.  This would yield a vertical margin that would dictate the performance of the actual equalizer and the slicer.  A minimum acceptable vertical margin would be defined that would allow sufficient margin for fixed precision tap weights, arithmetic, etc. and expected slicer sensitivity, offset, etc.

 

The concern is that if we do not abstract the results in this way, the simulation results will be highly implementation specific and comparisons of identical schemes from two different sources will fail to yield good correlation.  Also, people will be hesitant to reveal such implementation details, making it impossible to identify the differences and thereby cloud the results and bog down the process.  In summary. there are two parts to this problem:  (1) comparison of signaling schemes to select a baseline and (2) determining the timing margin and voltage margin that are required for a practical implementation to be feasible (which feeds back into the channel specification).  The group is working on part (1) at this time.

 

Total System Gain

 

Vertical eye opening must be normalized to launch amplitude and total system gain for objective comparison.  This would be the gain the system introduced by the analog front-end and AGC, and any gain realized in the FFE (computed as the sum of tap weight magnitudes).

 

Complete reporting of voltage and timing margin

 

  • Voltage at which timing margin is computed must be reported (important for multi-level schemes).
  • Time at which voltage margin is computed must be reports (tricky, since time frame of reference may vary among simulations)

 

Also, to assist in the correct interpretation of timing margin, it may be interesting to report the voltage margin from fixed timing offsets relative to the optimum.  The suggestion would be report the timing margin at the eye center, and again at a defined offset.  The proposal would be to report the worst voltage margin from an static offset of -0.10 or +0.10 UI.

 

This is a simplification of the request from Brian Brunn.

 

“Stressed” BER

 

The “deep” BER contours (1E-12/15/18) typically require some analytic method to derive.  The choice of method may influence the actual margin observed.  Also, there had been earlier discussions regarding scaling crosstalk.  This is probe to determine the interest, if any, in introducing new performance metric that might yield better correlation than the more analytical techniques.  This involves scaling the crosstalk to a level that yields a BER high enough to actually estimate via error counting.  This would be several points on a curve with a bottom at 1E-6 to 1E-7 and various scale factors of k. This is similar to the Moore proposal, but rather than using a sinusoidal interferer, it uses scaled actual crosstalk data.  At the very least, it would be sanity check of the various analytical methods and would build confidence in the reported margins.