|
March 14, 1997
Using multiline
models in dual-
and single-point grounding configurations
Timothy
R Minnick and Hank Herrmann, Amp Incorporated
When using
connector multiline models, choose your simulation's ground connections
with care. Single-point grounding onfigurations inaccurately predict
connector propagation delay, impedance, and crosstalk noise.
As data-transfer speeds
increase and digital-switching systems proliferate, the interconnection
system's impact grows. At high speeds, board-level connectors and device
sockets become electrically significant components that can
cause noise, reflections, crosstalk, and delays if improperly designed.
Circuit modeling using Spice allows engineers to circumvent the
design/build/test trap in designing connectors into systems, but you
must consider the model's design when simulating a new connector or
circuit. Accurately modeling a connector is a nontrivial task because
of the complex geometry involved.
The popularity of circuit simulators
that use a single-point ground raises questions about the proper use of
multiline models (MLMs), such as Amp's models. The MLM is a precise
Spice connector model, which represents several lines of a
connector, including all significant line-to-line couplings. The model
has no dedicated ground lines and allows independent ground connections
on each side. This configuration produces an accurate simulation of
near- and far-end crosstalk and also of ground-bounce effects
generated when signals pass through the connector. Not only do MLMs
provide a more precise portrayal of multiline switching, but they also
predict the effects of ground bounce.
 Figure 1 shows a simple MLM application. Note that the
grounds of the two pc boards connect through the ground lines of the
connector model, which accurately represents the application
environment; connectors connect boards. The grounds are free to float
with the common-mode voltages that exist across the interconnection.
Artificially tying the near and far grounds to one point is an incorrect
use of the model. A single-point ground shorts the model's
ground interconnection path and causes gross inaccuracy in the
simulation.
Specifically, single-point grounding
configurations inaccurately predict connector propagation delay,
impedance, and crosstalk noise of multiline connector models. Although
some characteristics correlate with correct data that the dual-point
grounding system provides, valid numbers are unpredictable and
inconsistent. Therefore, it is imperative that you use no single-point
grounding schemes when running simulations with MLMs designed for dual-
or multiple-point grounds, or the model provides misleading results. The
way to use these models is with a dual-point grounding system, as
the model's developers intended.
Previous tests have established the
validity and accuracy of simulations using MLMs in dual-point grounding
configurations. New tests compare the performance of MLMs in dual- and
single-point grounding configurations and show that simulations
using a single-point ground produce different results than do the
correct simulations with dual-point grounds.
Researchers at Amp investigated the
magnitude and nature of the errors that a single-point ground produces
for five connectors. Table
1 lists these
MLMs and the edge rate and system impedance that each test uses. Four of
the models are existing connectors for high-speed digital applications.
The last one, IMAGCON.SUB, is an imaginary, open-pin-field
connector model created specifically for this test. This model provides
an example of the simplest connection possible.
 The models vary in physical characteristics and
electrical-modeling complexity, providing a range of possibilities for
signal behavior. The signal-to-ground ratio of contacts is
approximately 3 to 1 for all connectors. Figure 2 shows the connectors and wiring patterns. The
legend details which lines connect to ground, which lines are active or
quiet, and which lines each test monitors.
Researchers tested and compared all
connector models in dual- and single-point grounding schemes. Also, the
tests excited the connectors with single and multiple signal contacts.
The drive signals consisted of various piecewise linear models
representing voltage levels and rise times that represent typical
application waveforms.
The test results characterized model
performance in the areas of propagation delay, impedance, near-end noise
(NEN), and far-end noise (FEN). Researchers evaluated propagation
delay between the input and output nodes of the connector model,
measured as the delay of the test signal's rising-edge midpoint. They
reported impedance as the reflected impedance calculated by dividing the
input current into the input voltage of a driven contact, and
they recorded the peak of the resulting impedance waveform.
Researchers measured NEN and FEN as
voltage levels on quiet (nondriven) contacts at the input (NEN) and
output (FEN) of the connector model. They always measured the FEN, in
the dual-point grounding configuration, with respect to the far ground.
Because the edge rates and the driven voltage swings vary for
different connector models, the testers measured data in percentage
noise of the driven voltage swing. Considering the different edge rates
and wiring patterns, it is unreasonable to compare the noise levels of
these connectors. The testing's intent is to compare grounding
schemes within each connector individually.
Analyze the results
 Figure 3 presents the propagation-delay results. For all
of the connectors, the figure provides data for one driven line (Figures 2a, 2c, 2e,
2g, and 2i) and
multiple driven lines (Figures 2b, 2d, 2f, 2h, and 2j). In most cases, the projected delay for a
simulation using a single-point grounding configuration is less than the
projected delay when using the MLM as designed. In all
multidriven-signal cases, the propagation delay is much less with one
ground. The significant reduction in propagation delay yields incorrect
data.
 Figure 4 summarizes the impedance-testing results. The
maximum-impedance figures are maximum-reflected impedances measured
at the model's near-end input. In all cases, the impedance measurements
of the single-point grounding configurations are less than those of the
dual-point grounding assignments. When driving one line, the
impedances are relatively similar between different grounding schemes.
However, when you excite multiple contacts with the same signals, the
differences in grounding schemes become much more significant. The
difference can be as great as one-third of the dual-point grounding
system's impedance.
 Figure 5 presents the crosstalk results of NEN and FEN for
single- and multiple-driven contacts. Figures 5a and 5b depict the NEN and FEN, respectively, of nearby
"quiet" contacts (those labeled "Q" in Figures 2a,
2c, 2e, 2g, and 2i) when driving one signal contact. Figure 5c shows both NEN and FEN when driving multiple
signal contacts and monitoring the only "quiet" contact (those
labeled "Q" in Figures 2b, 2d, 2f, 2h, and 2j).
Single-point-ground errors
In all cases of NEN and most of FEN,
the noise results of the dual-point grounding situation are
significantly larger than those of the single-point grounding. In almost
all FEN simulations, the single-point ground not only reduces the
noise magnitude significantly, but also changes the polarity. Also, the
data shows no consistent change in magnitude or percentage that you can
use to predict the error that the single-point ground introduces.
Noise results for one driven contact
vary from 90% more noise to 102.9% less noise when using one ground. A
noise reduction of greater than 100% results when the polarity also
changes. When you drive multiple contacts at once, the noise difference
becomes even more significant. The single-point grounding system
inaccurately depicts the actual noise of the interconnection.
| 
