[ibis-macro] Re: What happens when Init_Returns_Impulse = False?
- From: "Muranyi, Arpad" <Arpad_Muranyi@xxxxxxxxxx>
- To: <ibis-macro@xxxxxxxxxxxxx>
- Date: Tue, 27 Apr 2010 07:54:52 -0700
Todd,
I will put my comments/questions between your lines.
Thanks,
Arpad
====================================================
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Todd Westerhoff
Sent: Tuesday, April 27, 2010 9:12 AM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] Re: What happens when Init_Returns_Impulse =
False?
Arpad,
This is a really good question.
I view "Use_Init_Output" as controlling whether or not the impulse
response returned by Init should be factored into the Time-Domain
result. In truth, it only really has meaning when both
Init_Returns_Impulse and Getwave_Exists are True. I maintain that the
case:
Init_Returns_Impulse = True
Getwave_Exists = False
Use_Init_Output = False
Makes no sense, which is why I marked it as Invalid. The first two
parameters are declaring that the model's equalization is adequately
characterized by an impulse response (i.e., is LTI). If that is the
case, then Time-Domain simulation can derive any waveform behavior it
needs through convolution. The fine point here is how you define
"Use_Init_Output" .... if it is N/A when either of the two other
parameters are False, then you don't need to mark the above case as
Invalid. If you define "Use_Init_Output" as controlling whether the
model's impulse response gets factored into the Time-Domain result
(which, in my opinion, is a better definition from a flow standpoint),
then Use_Init_Output = False in the above case would tell the simulator
to omit the model's behavior from the Time-Domain result when it doesn't
need to, and therefore makes no practical sense.
Thus, I marked it as Invalid.
AM: This is all true for TD simulations. But I wanted to point out
that this combination is perfectly valid for a statistical simulation,
where Use_Init_Output is ignored (according to your last sentence
below),
GetWave is not needed (Exists=False), and Init returns a perfectly
usable
modified impulse response... So in your truth table I would have put
this as a valid combination with a note saying "Model supports
statistical
simulation only".
Your second case:
Init_Returns_Impulse = False
Getwave_Exists = True
Use_Init_Output = False
Declares that the model supports Time-Domain simulation but not
Statistical simulation. Remember that you can still run a Statistical
simulation, but the model's equalization will effectively be omitted.
Do you allow the Statistical simulation to be run or not? We do, but
the user needs to understand what the results represent.
AM: I agree with this, but this was not my point. I was referring to:
Init_Returns_Impulse = False
GetWave_Exists = True
Use_Init_Output = True
You marked this as Invalid, but I think this combination is equivalent
to the combination you marked with Note 2:
Init_Returns_Impulse = False
GetWave_Exists = True
Use_Init_Output = False
because the only difference between these two is the Use_Init_Output,
but since Init_Returns_Impulse = False, the output of Init is the same
as its input (Init doesn't modify the input), therefore whether we
pick its input or output by Use_Init_Output makes no difference. So
if one of these two cases is valid with Note two, the other should also
be just as valid with Note 2, yet you marked one of these as invalid.
Consider the case where the TX declares:
Init_Returns_Impulse = True
Getwave_Exists = False
Use_Init_Output = True
And the RX declares:
Init_Returns_Impulse = False
Getwave_Exists = True
Use_Init_Output = False
We note that the output of Statistical simulation represents the
waveform at the RX pad, instead of the decision point.
AM: In the light of your last sentence, this combination makes me wonder
again what really is the role of Use_Init_Output. If it should be
ignored
in statistical simulations, I would have just used an "X" after the
equal
sign that follows it. This way I feel that it does have a certain
meaning
and therefore I need to show it on my flow diagrams even for the
statistical
flow. This gives me the impression that we say that it should be
ignored,
but at the same time we are using it meaningfully in the examples...
Make sense?
FWIW, my understanding (and strong opinion) is that Use_Init_Output
never, ever, ever, ever affects Statistical simulation results. That's
tantamount to saying that we're just going to ignore the model's
behavior - and there's no reason to do that.
Todd.
________________________
Todd Westerhoff
VP, Software Products
SiSoft
6 Clock Tower Place, Suite 250
Maynard, MA 01754
(978) 461-0449 x24
twesterh@xxxxxxxxxx
www.sisoft.com
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Tuesday, April 27, 2010 2:00 AM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] Re: What happens when Init_Returns_Impulse =
False?
Forgot to explain myself on the first question:
If we use the "Use_Init_Output" in the sense that it won't
have an effect on statistical simulations, then the case
where GetWave Exists = False and Init_Returns_Impulse = True
and Use_Init_Output = False should also be valid for
statistical simulations, since Use_Init_Output is ignored
in statistical simulations, and the box should say
"Model supports statistical simulation ONLY".
On the other hand, if you made that case invalid because you
used "Use_Init_Output" in a sense that it also affects the
statistical path, then the case where GetWave_Exists = True
and Use_Init_Output = False and Init_Returns_Impulse = True
would not be able to support statistical simulations...
Arpad
=============================================================
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Tuesday, April 27, 2010 12:42 AM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] Re: What happens when Init_Returns_Impulse =
False?
Todd,
I am searching for truth in your truth table... :-) and have
two questions/comments:
1) In what sense did you use the "Use_Init_Output" in your
table? In the sense that it controls either the Statistical
or the TD flow, or in a sense that it only effects the TD
flow?
2) Shouldn't the case below "Note 2" also be valid and be
labeled with Note 2? Since Init_Returns_Impulse is False,
the output of Init is the same as its input, therefore the
value of Use_Init_Output shouldn't matter, we still get the
input of Init...
Thanks,
Arpad
=============================================================
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Todd Westerhoff
Sent: Monday, April 26, 2010 4:18 PM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] Re: What happens when Init_Returns_Impulse =
False?
Resend ... looks like the graphic got mangled.
--------------------
Arpad,
Here's a cut at what I think a full truth table could look like:
"Invalid" means the combination is not allowed. Use_Init_Output really
only has meaning when Init_Returns_Impulse and Getwave_Exists are both
true ... you could argue that when either Init_Returns_Impulse or
Getwave_Exists is False, Use_Init_Output is N/A. I chose not to
represent it that way because I found this representation to be simpler.
While I'm not fond of the names of these reserved parameters (I don't
like the terms "Init" and "Getwave" much either), there are many models
out there already that use them. I'm (personally) not strictly against
deprecating these parameter names and replacing them, but there'd better
be a darn good argument for it. Since these switches are basically used
only by model makers and EDA companies, I'd say we live with the current
names and do a better job of documenting how they are to be used.
Strict minimalists will point out that since we have three Boolean
variables and only four legal states, it ought to be possible to use
only two Boolean variables to encode what we need. While I agree that
should be possible, I'll also state that I don't want to go there, nor
do I think there's any tangible benefit in doing so.
Todd.
________________________
Todd Westerhoff
VP, Software Products
SiSoft
6 Clock Tower Place, Suite 250
Maynard, MA 01754
(978) 461-0449 x24
twesterh@xxxxxxxxxx
www.sisoft.com
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Monday, April 26, 2010 12:43 PM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] Re: What happens when Init_Returns_Impulse =
False?
Thanks Todd for the "truth table". I think it makes sense, but I
wonder what approach we should take to fix the spec. Here is the
problem.
As the spec is written now, there is no mention that Use_Init_Output
applies to time domain simulations only. As a consequence, the first
two cases you describe come across as having two controls which have
basically the same effect, i.e. duplicate controls. After having
gone through the email thread last week I understand now that this
was not the intent, but the spec can be interpreted this way without
fixing it.
So the question is this: We can tailor the spec to the controls or
the controls to the spec. Which way should we try to fix this?
Should we just change the explanation text in the spec to reflect
these intentions and then draw the flow diagrams accordingly, or
should try to find a flow diagram that may work in both time domain
and statistical modes with perhaps eliminating one or the other of
these Booleans? To be honest, I really don't like these two controls
the way they do almost the same, and that their name doesn't reflect
what the real difference is between them, i.e. that one is only
effecting time domain simulations but the other affects both...
Thanks,
Arpad
========================================================================
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Todd Westerhoff
Sent: Friday, April 23, 2010 3:13 PM
To: ibis-macro@xxxxxxxxxxxxx
Subject: [ibis-macro] Re: What happens when Init_Returns_Impulse =
False?
Arpad,
Sorry for the delay in reply.
Given: Use_Init_Output only applies to Time Domain simulation,
Then:
Init_Returns_Impulse = False, Use_Init_Output = False, Getwave_Exists =
True
and
Init_Returns_Impulse = True, Use_Init_Output = False, Getwave_Exists =
True
are functionally equivalent from a Time Domain simulation point of view.
Extra credit: the two cases are not the same from a Statistical
simulation point of view. The eye produced by the first case will not
include the model's equalization, while the eye produced by the second
case will.
Therefore, models that conform to:
Init_Returns_Impulse = True, Use_Init_Output =
False, Getwave_Exists = True
Fully support both Statistical and Time Domain simulation. Is that the
answer to the question you were asking?
Todd.
________________________
Todd Westerhoff
VP, Software Products
SiSoft
6 Clock Tower Place, Suite 250
Maynard, MA 01754
(978) 461-0449 x24
twesterh@xxxxxxxxxx
www.sisoft.com
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Thursday, April 22, 2010 7:31 PM
To: IBIS-ATM
Subject: [ibis-macro] Re: What happens when Init_Returns_Impulse =
False?
Todd,
In that case, how is Init_Returns_Impulse = False
different from Use_Init_Output = False?
I admit, I may have to brainwash myself now with
the recent revelations about section 2.3, but I
would appreciate your (or any one else's) insight
on this one...
Thanks,
Arpad
===================================================
________________________________
From: Todd Westerhoff [mailto:twesterh@xxxxxxxxxx]
Sent: Thursday, April 22, 2010 12:13 PM
To: Muranyi, Arpad; 'IBIS-ATM'
Subject: RE: [ibis-macro] What happens when Init_Returns_Impulse =
False?
All,
As I understand it, Init is expected to modify the impulse response in
place in memory. The simulator passes the model a pointer to the
impulse response, and uses that same pointer to access the result when
the Init call returns control to the simulator.
If the model doesn't touch the impulse response at all, it's there in
memory just as the simulator supplied it.
Todd.
________________________
Todd Westerhoff
VP, Software Products
SiSoft
6 Clock Tower Place, Suite 250
Maynard, MA 01754
(978) 461-0449 x24
twesterh@xxxxxxxxxx
www.sisoft.com
________________________________
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Thursday, April 22, 2010 12:00 PM
To: IBIS-ATM
Subject: [ibis-macro] What happens when Init_Returns_Impulse = False?
I would like to get some ideas from the authors
of the AMI specification on what the intent was
when Init_Returns_Impulse = False.
I would also like to get some suggestions on what
the Init function should return when
Init_Returns_Impulse = False.
Options:
- return the same thing that was sent into Init
- return impulse response of the filter
- return uninitialized memory (garbage or noise)
It seems that either the first or second answer
may be the correct one if I consider that the
calculations are supposed to be done "in place",
i.e. in the same memory space where the input
was placed. If nothing is done, the memory will
still contain the input data in that place. If
something is done, like returning the filter only
impulse response, we will get that in that memory
space. It is hard to turn that memory space into
uninitialized memory, unless we use a random number
generator to do something that can mimic that effect.
Thanks,
Arpad
======================================================
Other related posts:
