[SI-LIST] Re: Waveform Correlation

Greg:

Thanks for your reply.  Some of your comments need to be addressed by
TDR and VNA/freqeuncy domain measurement gurus.

 I agree that TDR and VNA analyses can prove critical for
reliable waveform correlation.  For instance, when correlating a scope
measured clok signal with corresponding simulated waveform it
 is important to ensure that values assigned to  impedance and velocity
 for simulation transmission line models match the trace characteristic on
actual PCB. This often  necessitates TDR testing.

Similarly, a series of TDR or VNA measurements are usually required to
ascertain high quality wide bandwidth models for vias, connectors and
package parasitics, etc.

Best Regards,

Abe

----- Original Message -----
From: "greg kimball" <gkimball@xxxxxxxxx>
To: <ARIAZI@xxxxxxxxxxx>
Cc: <si-list@xxxxxxxxxxxxx>
Sent: Sunday, May 26, 2002 9:35 AM
Subject: [SI-LIST] Re: Waveform Correlation


>
> Abe et al-
> There is another dimension to the model vs. measurement exercise - one
which i have been grapeling
> with for awhile. This is the  time AND frequency domain correlation. Just
because the time domain
> response matches measurement,  does NOT necessarily produce frequency
domain correlation -
> and visa versa. (i.e. transfer response, or TDT in time domain). And it IS
important to match things
> up - mutually.
> In my work, i initiate my modeling exercise from TDR measurements because
of its greater 'bandwidth'
> compared to my order of magnitude lower BW VNA, and because i need to see
the spacial
> response. However, when i look at transfer response in the freq domain - i
do not see as much modeled  loss
> as i measure. This is partly to do with my  model construct, for sure.
>
> Would appreciate any commentary regarding this disparity from ALL the
community:
> - mutual-domain correlation validation practice
> -mutual-domain non-correlation reasons
>
> Regards,
> Greg Kimball
>
> PS Yes, i also use Excel to marry my PSPICE and measured waveforms
>
> Abe Riazi wrote:
>
> > Dear all,
> >
> > Comparing simulation generated waveforms with the corresponding lab
measured data can yield valuable information towards identifying/eliminating
signal integrity problems, enhancing buffer and
> > interconnect models, and optimizing simulation/measurement techniques,
etc.
> >
> > One effective correlation approach involves superimposing the measured
and simulated
> > waveforms utilizing Microsoft Excel. This is feasible because many
simulator and measurement
> >  tools can output data in a format compatible with Excel. For instance,
the Multi-column (.dat) files exported by XNS and  .csv (Comma Space
Delimited) files stored by digital oscilloscopes can be readily imported
into an Excel spreadsheet.  It is important for the simulated and measured
data
> > to have similar bit patterns and to be captured at the same location
(topology node). Let us
> > consider a sample case.
> >
> > Example.
> >  A signal waveform, belonging to memory subsystem of a high-speed
motherboard, was captured (in Read cycle) using a HP infinium 1.5GHz  8GSa/S
oscilloscope, in conjunction with Tektronix P6248 (1.7 GHz) differential
probes. The result was recorded in  .bmp ( for quick viewing) and .csv ( for
importing to Excel) formats. Waveform examination revealed that the measured
signal initiates
> > in High Z state, makes a transition to Low (remains Low for 5 nS), then
switches to High (stays
> > High for 10 nS), followed by  another transitions to Low (for 5 nS) and
finally to High Z.
> > The corresponding simulations were performed utilizing XTK. A .mdc (
Multi-Driver
> > Contention) file was constructed in a manner that the driver output
exhibited same bit pattern as measured signal.  Furthermore, the simulation
waveform was probed at the topology node
> > consistent with measured data. The simulation result was then exported
and saved in Multi-column (.dat) format.   The measured " .csv " and the
simulated " .dat " files were entered into Excel for correlation analyses.
It was noted that the time axis of simulated result required a conversion
(from nano seconds to seconds) and the measured data needed  a time shift of
~ 800pS, in order to optimally overlay the two signals..
> >
> > Above example has described a way for superimposing measured and
simulated waveforms.
> >
> > With the simulated/measured results properly overlaid, it is possible to
evaluate their sameness
> > and/or differences. The signal features frequently examined include the
low and high DC quiescent voltage levels, the rise/fall times,
overshoot/undershoot,  ringing, ringback, glitches, etc.
> >
> > It is also important that the selected simulation corner (i.e. Min, Typ,
Max) and
> > the voltage/temperature conditions of measured data to be consistent
with each other.
> > Although, the nominal corner correlations are very common; correlating
under Fast (best case)
> > and Slow (worst case) requirements can be also beneficial. Furthermore,
it may be necessary to evaluate the signal waveforms at both the driver
output and the receiver input pins (or pads).
> >
> > Comparison of simulated and measured data usually reveals certain
differences some
> > of which can be due to model deficiencies, simulator limitations, or
certain measurement
> > inaccuracies.  Correctly interpreting the measured/simulated signal
features, their similarities
> > and differences constitute a crucial element of accurate correlation.
> >
> > In summary, comparing key features of simulated and measured waveforms
can be helpful
> > towards identification/ removal of SI degradation, model optimization,
and improvements of simulation/measurement methods. Correlation can be
carried out by overlaying the measured
> > and simulated waveforms using the Excel program. This approach often
necessitates applying
> > certain adjustments/shifts to time axis data of one or both waveforms to
achieve optimum
> > superposition of the results.  It is usually desirable that the points
(topology nodes) of capture,
> > driver output bit patterns, voltage/temperature conditions of simulated
and measured waveforms be nearly matched.
> >
> > Your comments are highly appreciated.
> >
> > Respectfully,
> >
> > Abe Riazi
> > ServerWorks





------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field

or to administer your membership from a web page, go to:
http://www.freelists.org/webpage/si-list

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field

List archives are viewable at:     
                http://www.freelists.org/archives/si-list
or at our remote archives:
                http://groups.yahoo.com/group/si-list/messages 
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu

Other related posts: