开云体育

--- In LTspice@..., Message 141, Panama Mike
<panamatex@y...> wrote:
< snip >

BTW, I'm thinking of introducing opamp models that
use a different modeling methodology, similar to
that used for LTspice's SMPS products. The result
would be computationally extremely lightweight and
robust models that model noise too(these PSpice-
style opamp models almost never get the noise
modeled). However, the opamps models would not
run in other SPICE simulators and non-LT opamp
models wouldn't be available. Would you folks be
interested in something like that?

--Mike

Mike, this sounds like something I'd like to dissuade you from.

Part of the strength of the SPICE methodology is that the models are
transmitted as "open source", simple text files. The most significant
advantage is that we Mere Mortals can easily extend, improve, correct,
or modify models as needed.

The parent thread for this posting is a good example. Because almost
everything about the model was in plain view, several minds were
independently analyzing the problem and solving it. I cannot imagine
the problem being resolved nearly as quickly if the model's topology
and parameter values had been locked-up in a proprietary format
readable only by a few people.

The SPICE methodology permits individuals to customize models as
needed. If, for instance, noise is a critical performance
characteristic the necessary elements can be readily included to model
it. Otherwise they may be omitted. Similarly, a small-signal stage
where output limiting is not a concern can get by with a simplified
output circuit in the model.

Along the same line it is relatively easy to adjust model parameters
to fit particular situations. The model can be customized to reflect
the device's behavior at, say, a temperature extreme. Or an engineer
can investigate the implications of using a device whose performance
parameters (like offset voltage or slew rate) are near the data sheet
limits. Likewise the need for parts specially selected for certain
characteristics (such as low offset current) can be evaluated.

Finally the current SPICE modeling methodology allows engineers to
quickly create workable models for new or alternative components.

I hope that whatever modeling methodology you choose will retain these
features.

Dale

--- In LTspice@..., "Dale" <dchishol@c...> wrote:

--- In LTspice@..., Message 141, Panama Mike
<panamatex@y...> wrote:
< snip >

BTW, I'm thinking of introducing opamp models that
use a different modeling methodology, similar to
that used for LTspice's SMPS products. The result
would be computationally extremely lightweight and
robust models that model noise too(these PSpice-
style opamp models almost never get the noise
modeled). However, the opamps models would not
run in other SPICE simulators and non-LT opamp
models wouldn't be available. Would you folks be
interested in something like that?

--Mike

Mike, this sounds like something I'd like to dissuade you from.

Part of the strength of the SPICE methodology is that the models are
transmitted as "open source", simple text files. The most

significant

advantage is that we Mere Mortals can easily extend, improve,

correct,

or modify models as needed.

Hello Dale,
I fully agree with you. The biggest advantage of all the opamp models
from different vendors is that they follow the general accepted SPICE
syntax. This standard has been the base for the success of SPICE over
the last thirty years. This is at least true for most of the analog
parts like diodes, transistors, passive components and the opamps.

It may be different for SMPS, because they are much more mixed signal
devices. Here we have a lack of standard for digital parts and also a
missing standard for behavioral language syntax. One more reason is
the needed compuational speed of SMPS models for effective usage. I
believe it is ok to have special models for the SMPS, because they
are developed independently of the other anaolg/digital circuits of a
design.

Hello Mike,
I recommend to keep the "easier" parts like opamps compatible with
standard (P)SPICE, because many of LT customers use other SPICE
simulators for different reasons.
The provided SPICE models should be also optimized for good
convergence in the simulation. If a model doesn't provide some
features like noise modeling (.AC), it should behave more like an
ideal component in such a type of simulation.

Best Rgeards
Helmut

Dale wrote:

Mike, this sounds like something I'd like to
dissuade you from.

Helmult wrote:
I fully agree with you. The biggest advantage
of all the opamp models
[...]
The provided SPICE models should be also optimized
for good convergence in the simulation. If a model
doesn't provide some features like noise modeling
(.AC), it should behave more like an ideal
component in such a type of simulation.

The problem is that the PSpice models often don't
converge very well and don't model noise correctly.
That has nothing to do with Linear's opamp models,
it's common to most Boyle style models. They don't
converge well in PSpice either. It's just a really
lame modeling methodology. Sometimes I honestly
get the impression that SPICE macromodeling
engineers tweak a model until it doesn't run anymore
and pronounce it done at that point, blaming any
convergence problems are due to the simulator.

Essentially all customer-reported SwCADIII
convergence problems reported deal with using
these opamp models. But inside the mixed-mode
simulator in LTspice is the ability to model
an opamp model like I described. Few convergence
problems, one or two internal nodes, good noise
modeling and almost no load on the simulation
run time. The technology already exists in
LTspice and is used in the SMPS products' error
amps.

Another problem I have is there's some newer
Linear opamps that don't have any SPICE models.
If I go to this new method, then I can make a
model in less than an hour that will be more
accurate than the former PSpice models. It's
much cheaper and it's hard to justify these
expensive PSpice models that don't work well.
Of course the advantage of being able to run
them in PSpice is important.

--Mike

__________________________________________________
Do you Yahoo!?
Yahoo! Platinum - Watch CBS' NCAA March Madness, live on your desktop!

--- In LTspice@..., Panama Mike <panamatex@y...> wrote:

Dale wrote:

Mike, this sounds like something I'd like to
dissuade you from.

Helmult wrote:
I fully agree with you. The biggest advantage
of all the opamp models
[...]
The provided SPICE models should be also optimized
for good convergence in the simulation. If a model
doesn't provide some features like noise modeling
(.AC), it should behave more like an ideal
component in such a type of simulation.

The problem is that the PSpice models often don't
converge very well and don't model noise correctly.
That has nothing to do with Linear's opamp models,
it's common to most Boyle style models.

Hello Mike,
I have never said here that it is specific for LT models. My
statement has been a general one for all vendor's models.
If it is true that the Boyle model is so weak, why not starting with
another SPICE model? I am shure that LT has the right people(you for
example) to make excellent models.

They don't
converge well in PSpice either. It's just a really
lame modeling methodology. Sometimes I honestly
get the impression that SPICE macromodeling
engineers tweak a model until it doesn't run anymore
and pronounce it done at that point, blaming any
convergence problems are due to the simulator.

I have experimented with my own generic opamp model and indeed it
converges very different depending on the choosen circuit.

LTSPICE has been greatly improved over the last year regarding
convergence problems. Most of the problems seem to be history.

Of course the advantage of being able to run
them in PSpice is important.

There are even more SPICE simulators around. Some of them are
specific SPICE simulators like ICAP and others are part of PCB-CAD
packages. All these users need/want SPICE models of LT opamps.
Finally I hope that LT always provide opamp models for the whole
SPICE "family" too.

Best Regards
Helmut