Marek,
I agree with Arpad and Todd, and would like to add a few details. First, you
mentioned IBIS AMI simulation, which assumes that you have IBIS models with
reference to the corresponding IBIS AMI DLLs. In general, IBIS models require
Spice-like simulator, because they may describe non-linearities, e.g., by IV
tables and clamping devices. Some have 4-port S-parameters describing Tx and Rx
buffers as a linear time-invariant models. In either case, Spice-type simulator
will let you find the waveform at the Rx pin or die, when the input is a
digital transition from low to high. This way, you will receive the rising edge
waveform that reflects the properties of the channel, and parasitic effects in
the buffers, such as finite Tx waveform rise time, loading effects because of
impedance mismatch, etc. In many case you may have a differential channel and
only need to measure the signal between positive and negative outputs.
Then, you can find the "impulse response", required by IBIS AMI models by
numerical differentiation of the found edge. The result will be Dirac pulse
with high magnitude because you divide the differences on the time step, which
should be small enough, e.g., providing 32 sample points per unit interval.
When the channel or the Tx/Rx buffers are defined by S-parameters, the issues
mentioned by Aldo related to accurate fitting of the S-parameters are real, and
should be taken seriously. However, depending on your task, you may have a
channel that consists of standard circuit elements (R/L/C/lossy Tlines, etc.)
and may not vector fit to convert your S-parameters into the time response.
Spice-type simulators can handle S-parameters as well, including the vector fit
followed by fast recursive convolution in time domain.
A separate case is when you don't have such simulator. Then, if the channel and
the buffers are defined by S-parameters, you may take several steps, such as
cascading you Tx, channel and Rx defined by s4p files, then finding the
differential 2x2 submatrix only, applying termination on both sides to convert
it into a scalar transfer function of the channel. Then, convert the frequency
response into a Dirac impulse by IFFT. By integrating Diract response you will
find the step transition, then the pulse (response to an isolated symbol) can
be found as the difference between the step transition and its copy delayed by
on symbol interval. This method is practically identical to what they define in
IEEE COM (channel operating margin) specification. A convenient way is to
perform such transformations with Matlab, and there also exists IEEE Matlab
script that computes COM for a number of popular operating modes.
Vladimir
Msg: #1 in digest
From: "arpad.muranyi @ siemens . com" <dmarc-noreply@xxxxxxxxxxxxx>
Subject: [SI-LIST] Re: IBIS-AMI pulse shapes for channel characterization
Date: Thu, 4 Apr 2024 01:49:06 +0000
Marek,
I can't recommend any books or papers on this subject, which doesn't mean that
there aren't any...
However, I would like to offer a simple answer to your question. While it is
true that SPICE-like simulators can't simulate
an infinitely vertical pulse or impulse, I don't think that it is really
necessary in practice. Consider a simple illustration
that you could even try easily in your favorite simulator, an RC circuit with
an ideal pulse source with a finite edge (slope)
as the driver. The resistor is between the source and the capacitor, and you
are interested in the waveform at the capacitor.
This represents a "channel" in which the voltage source and resistor is the
transmitter (Tx) and the capacitor represents the
losses in the traces (channel) and the receiver's input capacitance.
How steep does the edge of the pulse of this ideal source need to be to obtain
the correct RC response waveform at the
capacitor?
Mathematically it should be a vertical edge. But is it going to make a
difference if it is perfectly vertical, say if it has a 1 ps slope? Well, if
the RC constant is significantly larger than that 1 ps edge, say 1 ns, you will
probably not even notice that
the driving edge is not vertical, or whether it has a 1 ps or 2 ps slope. On
the contrary, if the RC circuit have a 1 ps time
constant and your voltage source has a 1 ns slope, you would probably only see
the 1 ns slope at the capacitor, instead of
the usual exponential RC waveform.
This handwaving argument applies to IBIS-AMI and similar channel
characterization simulations too. The conclusion is
that you only need a faster edge than the time constant of the channel you are
measuring, but it doesn't have to be a
perfectly vertical edge as mathematics would like to have it.
Does this help?
Thanks,
Arpad
------------------------------------------------------------------
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:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu