Spice simulators and accuracy results

[rtarbell]

I'm simulating a fully differential circuit with TopSpice to test the common mode feedback loop and dc bias points, and then I tried to simulate the same circuit with the regular Pspice student edition. What a difference! Pspice is showing me the results that I would expect (DC output offset at each output is on the order of microvolts), while TopSpice is showing that both outputs rail in simulation.

I have tried simulating the circuit with the TopSpice transient analysis as well to see if the circuit "adjusts" itself after a few seconds - still no luck, both outputs rail.

My question is: has anyone else experienced this result when using different simulators? I know that each company uses its own algorithm and whatnot, but I wouldn't expect this much difference...

[gootee]

rtarbell,

I only use LTspice (aka SwitcherCad), from http://www.linear.com , and have never used either of the packages you mentioned. There is a support/discussion group for Ltspice, called LT-SPICE, at http://www.yahoogroups.com . Is there something similar where you could get help for TopSpice?

You didn't say anything about having checked for differences in their setups. There are many "internal" variables that are common to most Spice implementations, that are user controllable, e.g. TRTOL, just as an example. In LTspice, many of the common ones are easily accessible from the menu.

And, actually, many Spice implementations do use the same core algorithms, from the original Berkely spice engine.

Also, how did you get the circuit from one platform to another? Is there a possible source of error or differences, in that process? e.g. Did you re-enter everything by hand, or what? Have you compared the two netlist listings, line by line?

Are you much more familiar with one of those packages than you are with the other one?

Have you ever used TopSpice successfully? Are you an experienced TopSpice user?

Obviously, until you figure out exactly what's happening, you probably shouldn't use TopSpice for anything important.

Hopefully, someone with TopSpice experience will chime in, here.

Good luck.

- Tom Gootee

http://www.fullnet.com/~tomg/index.html

-

[I_Forgot]

Are you using the same netlist and models in both cases?

Are you sure?

I_F

[rtarbell]

==> Are you using the same netlist and models in both cases?

Indeed, I have used the same transistors (3904 and 3906), and I have the exact same two power supply voltages (+/- 75V), and resistor values are all the same.

I have used TopSpice for quite awhile (it's the only graphical SPICE simulator that I can afford the full version for). I always trusted the results it gave me - maybe I'll play with the tolerances (like reltol and whatnot).

[Christer]

Originally posted by rtarbell
==> Are you using the same netlist and models in both cases?

Indeed, I have used the same transistors (3904 and 3906), and I have the exact same two power supply voltages (+/- 75V), and resistor values are all the same.

Are the transistor models the same? Don't trust a model to be correct or even reasonable just because it shipped with the simulator.

I have used TopSpice for quite awhile (it's the only graphical SPICE simulator that I can afford the full version for).

I suppose you could afford LTSpice too. It't free.

[gootee]

Originally posted by rtarbell
<snipped>- maybe I'll play with the tolerances (like reltol and whatnot).

That's one thing I was trying to get at, before. Have you made sure that all of those types of settings, or, rather, ALL settings, are exactly the same, for both packages?

- Tom

[soongsc]

Originally posted by rtarbell
I'm simulating a fully differential circuit with TopSpice to test the common mode feedback loop and dc bias points, and then I tried to simulate the same circuit with the regular Pspice student edition. What a difference! Pspice is showing me the results that I would expect (DC output offset at each output is on the order of microvolts), while TopSpice is showing that both outputs rail in simulation.

I have tried simulating the circuit with the TopSpice transient analysis as well to see if the circuit "adjusts" itself after a few seconds - still no luck, both outputs rail.

My question is: has anyone else experienced this result when using different simulators? I know that each company uses its own algorithm and whatnot, but I wouldn't expect this much difference...

Model implementation differences are probably more significant unless the spice engine has a complete different implementation.

[rtarbell]

THank you all for your replies -

Regarding LTSpice, are there any limitations on it (like only 10 transistors ina circuit, only 60 nodes, etc.)? Can I import my own models if their stock catalog does not have a part I need?

[gootee]

Originally posted by rtarbell
THank you all for your replies -

Regarding LTSpice, are there any limitations on it (like only 10 transistors ina circuit, only 60 nodes, etc.)? Can I import my own models if their stock catalog does not have a part I need?

The free LTspice is not a demo version. There are NO limitations. And I have read that it is one of the best Spice packages available, at any price.

The manufacturer, Linear Technology Corp, is a major IC manufacturer. They give away LTspice to help sell their ICs.

Yes, you can use virtually any standard P-Spice model, with LTspice. You will likely need to do so, since their standard libraries are fairly sparse, except for LTC parts. But there are many models in the Files section of the LT-SPICE users' group. And there are many more at, among other places, http://homepages.which.net/~paul.hil...odelIndex.html .

I have a page with a link directly to the LTspice download image, as well as a direct link to the LT-SPICE discussion group (plus a download link for my "transformer modeling directly from measurements" LTspice circuit), at:

http://www.fullnet.com/~tomg/gooteesp.htm

- Tom Gootee

[I_Forgot]

Originally posted by soongsc

Model implementation differences are probably more significant unless the spice engine has a complete different implementation.

There is only one "spice engine"- that produced by UC Berkeley. All the commercial products that are "spice" simulators are shells that provide spice with the netlist and process the output in particular ways.

If the two results are drastically different, it's a safe bet that there are different net lists and/or models being used.

I_F