Circuit Simulator faults

[LvW]

At present, I am investigating and comparing some of the commercially available circuit simulation programs (analog only).
In particular, I am interested to find out if they can fail in certain cases.
To avoid misunderstandings - I am NOT speaking about some simplifications implemented into the models .
(see for example:https://www.edaboard.com/threads/273084/#post1171145).

I am working with circuit simulators (in most cases PSpice, LTSpice, Micro-Cap, TopSpice) since more than 20 years - and I must confess
that I didn´t detect any systematic program error up to now.

If I got weird results - the reason was always
* an error during describing the circuit (net list, schematic entry),
* a misinterpretation of the results,
* false expectations (caused by a lack of knowledge)
* model simplification (ideal resistors/capacitors without parasitics, etc.),
* or a conflict between real and idealized world (circuit oscillates with ideal VCVS, but not with real opamp model).

But in no cases, the program itself introduced some errors. That`s the background of my question.

Thus, another form of my question could be:
Are there circuit simulation programs which produce different results - based on the same models describing the parts properties?

It would be interesting to share the experience of other forum members regarding this topic.


Thank you
LvW

 

[keith1200rs]

I am not sure if this example is what you are after but... I use SIMetrix and was working on a design for someone who was using LTSpice. The noise results with the same circuit were different. It was ultimately tracked down, with the help of SIMetrix and Linear Technology, to be the way noise was modelled in the ADA4898 (and some other Analog Devices models). This was the reply from Linear Technology:

The problem is that the model uses diodes to add
noise without any current in the diodes. Most
SPICE programs make a grave error in the diode
noise in the unbiased, linear region, where it
acts like a resistor and exhibits Johnson noise.
At zero current, the diode has the I-V curve of a
resistor and does, in fact, have Johnson noise.

LTspice does not do this.

The response from SIMetrix was:

Yes the original SPICE diode model is wrong at very low currents. There are related problems with all models that use PN junctions (BJTs etc).

In reality there are very few applications where this is significant - but if someone is exploiting this deficiency then we will have a problem. My intention is to provide an option setting that will switch the correction on. In other words it's wrong by default.

I am not sure which one is actually "correct". As the reply from SIMetrix says, it depends on what behaviour Analog Devices were expecting and which simulator they were using when making the model in the first place.

Keith.

 

[LvW]

Keith, thanks for your reply.
But - if I understood correctly: The mentioned discrepancies were model related. Right?
On the other hand, I must confess that - up to now - the question of noise (noise sources, different noise forms) didn`t play a major role in my simulations.
Thus, I have not much experience in this particular field.
LvW

 

[keith1200rs]

Yes, it is model related but it is based on original flaws in the diode model. Some simulators ignore it, some fix it. It is not a simulator bug in the sense of the simulator doing something it shouldn't so is maybe not the example you are looking for (although there is no way of knowing how specific simulators are going to behave with such a diode noise model as it is not well publicised).

I have had the odd problem over the years with simulators which needed to be fixed - a problem with binned models in Monte Carlo analysis and some issues with SOI (but the SOI was not officially supported so was supplied "as is") but no current bugs that I know of.

Keith

 

[BradtheRad]

A simulator has to do a certain number of iterations between each time step if it is to have any semblance of accuracy.

Ideally it should converge on a consistent set of values at every node, capacitor charge, inductor charge, etc., at that instant in time.

The programmer realizes it will not look good if his simulator takes a long time to calculate frames. End users will turn to some other program.

However sometimes the simulator, for whatever reason, cannot reach a convergence.

The programmer also knows it will not look good for his simulator to display a lot of 'Convergence failed' errors. End users will turn to some other program.

So the programmer feels a strong incentive to say 'Okay, we did 20 iterations for this frame. That should be enough. Let's display the results as they are, and get on to the next frame.'

The task is compounded as simulators have become increasingly sophisticated over the decades. They have added more and more parameters for every component behavior.

It's a good thing computer speeds have increased as well, to handle all those calculations within a reasonable time period.

 

[FvM]

The diode noise problem sounds like a very special case. As it refers to built-in models, it seems to belong to the category of erroneous simulator behaviour addressed in the original post. Needless to say that I didn't experience it yet, or at least wasn't aware of.

Another category of simulation problems that might be considered here is the effect of various simulations parameters like node gmin, or accepted absolute and relative error in iteration. Different simulators possibly have different default settings which can cause considerably different results with "extreme" circuits. Although it's up to the user to adjust parameters according to simuation requirements, he'll possibly never does due to lack of understanding.

 

[LvW]

Another category of simulation problems that might be considered here is the effect of various simulations parameters like node gmin, or accepted absolute and relative error in iteration. Different simulators possibly have different default settings which can cause considerably different results with "extreme" circuits. .

Oh yes - that´s certainly true. One has to be aware of it. Thanks.

 

[phongphanp]

Hard problem ! It's not disclose for data of simulator ex. renew some part : initial files , all detail data of equipment . It's just choise to specific aim , not all.

 

[godfreyl]

....and I must confess
that I didn't detect any systematic program error up to now.

I did a few years ago, when I was using the free "student" version of Circuitmaker.

The problem started when I was trying to do AC analyses of a tuned RF amp. This resulted in endless hours of frustration, trying to understand results that just didn't seem to make sense, before I realized some of the simulation results were just plain wrong.

The penny dropped when I plotted amplitude vs frequency for both the voltage across a resistor and the current through it, and got completely different curves!

At that point I abandoned Circuitmaker and went looking for other software. No problems since.

My impressions so far:

1. Circuitmaker: free, easy to use, has serious bugs.
2. LTspice: free, difficult to use, works properly.
3. SIMetrix: free, easy to use, works properly.

 

[phongphanp]

May be about the easy circuit project , when Prof. Dr. who produce simulator looks your circuit then he may confuse some choice on simple part.
I'm think of "Student Version simulator" must use basic calculations before the circuit evaluated/generated by computer because it's not fully function version.


You should inform term of error to simulator company, directly.