Hi Bob,
I'm using this model, but I haven't gone through to verify the parameters. Removing level=2 is fine for LTspice. LTspice only understands level=4 and level=9, which tell it to use VBIC instead of Gummel-Poon. QCO, RCO and VO are quasi-sat parameters and should be valid for LTspice. All I did was comment out IBC and delete level=2.
One reason for DC convergence problems can be a very high DC output impedance of the VAS. The Early voltage of this device model is very large, so this could be a contributor. Try increasing the iteration limit with:
.options itl1=1000
.options itl6=1000
Sometimes decreasing gmin can help. I am using a VAS with MOSFETs, and I can't get convergence with normal means and have to resort to extreme measures. Hopefully you won't have to do that. Another thing to look at is trying the alternate solver, and making sure you have the latest LTspice. A few months back they added a new source-stepping algorithm to improve convergence.
I'm using this model, but I haven't gone through to verify the parameters. Removing level=2 is fine for LTspice. LTspice only understands level=4 and level=9, which tell it to use VBIC instead of Gummel-Poon. QCO, RCO and VO are quasi-sat parameters and should be valid for LTspice. All I did was comment out IBC and delete level=2.
One reason for DC convergence problems can be a very high DC output impedance of the VAS. The Early voltage of this device model is very large, so this could be a contributor. Try increasing the iteration limit with:
.options itl1=1000
.options itl6=1000
Sometimes decreasing gmin can help. I am using a VAS with MOSFETs, and I can't get convergence with normal means and have to resort to extreme measures. Hopefully you won't have to do that. Another thing to look at is trying the alternate solver, and making sure you have the latest LTspice. A few months back they added a new source-stepping algorithm to improve convergence.
Hi Bob
One trick which is used in at least one professional software simulator is to add a 1 Tohm resistor to all nodes. Especially for MOSFETS this can help to avoid "open capacitor" problems.
In simpler simulators we may have to add a suitable high but acceptable resistor to do this manually...
cheers
John
One trick which is used in at least one professional software simulator is to add a 1 Tohm resistor to all nodes. Especially for MOSFETS this can help to avoid "open capacitor" problems.
In simpler simulators we may have to add a suitable high but acceptable resistor to do this manually...
cheers
John
Re: Fairchild KSC3503/2SC3503 Model
Bob,
There is a Yahoo group dedicated to LTSpice and they have been very helpful with modeling problems.
Look for them here:
LTSpice User's Group
Regards,
Jeff
Bob Cordell said:
Bob,
There is a Yahoo group dedicated to LTSpice and they have been very helpful with modeling problems.
Look for them here:
LTSpice User's Group
Regards,
Jeff
OPA2134 and OPA134 spice models
I have added one spice model to my library today.
See below.
OPA134 .. and of course can be used for OPA2134 as well
1. Would you say this is a good model ?
2. If not, what model should I use for OPA134 ?
Thanks
Lineup
================================================
I have added one spice model to my library today.
See below.
OPA134 .. and of course can be used for OPA2134 as well
1. Would you say this is a good model ?
2. If not, what model should I use for OPA134 ?
Thanks
Lineup
================================================
Lineup, I think it is probably very difficult to judge an op amp model by looking at it (except if your name is Scott Wurcer perhaps? 
). It also depends on what parameters and things you are interested in. A model may be very good for predicting some behaviours and not model others at all. Try simulations to see how it behaves compared to the datasheet for the things you are interested in. Things like frequency response and phase are usually well modelled, other things may vary a lot. Some models take supply voltage inte account, others assume a certain voltage.
I did a really interesting test a couple of years ago, after reading an AD app note about various decoupling schemes depending on internal topology. The AD825 model did have different behaviour depending on how I decoupled it and corresponded to the theory (although I don't know how correct it was in quantitative terms) while some other AD op amp (don't remember which one) obviously didn't model such effects at all.
). It also depends on what parameters and things you are interested in. A model may be very good for predicting some behaviours and not model others at all. Try simulations to see how it behaves compared to the datasheet for the things you are interested in. Things like frequency response and phase are usually well modelled, other things may vary a lot. Some models take supply voltage inte account, others assume a certain voltage.I did a really interesting test a couple of years ago, after reading an AD app note about various decoupling schemes depending on internal topology. The AD825 model did have different behaviour depending on how I decoupled it and corresponded to the theory (although I don't know how correct it was in quantitative terms) while some other AD op amp (don't remember which one) obviously didn't model such effects at all.
From what I have looked at opamp models, I'd say the majority of models is pretty flawed in most aspects important for audio, above all distortion (which is *never* modelled) but also power supply effects (close to never modelled in a realistic way) etc. The twelve(!) year old model for the 134 is pretty coarse, probably not any good for detailed analysis of all but the most basic behaviour (AC stuff).
There are better models for newer opamps, anf for how close they are to the real thing they usually tell us precisly what is modelled and how deep.
The recent OPA211 is a good example, look how big it is and what they state in the header.
- Klaus
There are better models for newer opamps, anf for how close they are to the real thing they usually tell us precisly what is modelled and how deep.
The recent OPA211 is a good example, look how big it is and what they state in the header.
- Klaus
KSTR said:
I know what you mean: 'pretty flawed'
And now we are speaking OP-amp models. And not transistors.
For example the default model of AD797 in my library is totally useless.
It shows nothing but ideal performance
I did run some tests of Open Loop with this OPA134 model.
It does not seems to be too bad:
- Open Loop gain: ~10.000
- Open Loop Unity gain BWidth: Like 8-10 MHz
- Current output where start to distort: ~35 mA peak
I would say it is more of an honest model.
At least many times better than my discusting AD797 spice model
Regars, Lineup
john curl said:I thought that today's SPICE models were 'perfect'.
Are all real components perfect, just because some of them are very good?
It is difficult to make a really good model, and for complex things like op amps there is also the trade off between goodness and simulation speed. And let's not forget that the manufacturers also don't want to give us too many details about what's actually inside them. It would be interesting to know how good the in house simulators of AD, TI and others are. I understand they usually use better simulators than Spice also.
john curl said:I thought that today's SPICE models were 'perfect'.
John.
Some models are excellent.
It has been verified that they can predict real circuits very well.
But some models submitted, probably by manufacturers of products,
are far from reality.
To use them and get some valid results, would be a big mistake.
Just like any other model of reality - there are good and bad.
Take maps.
There are many different maps:
- road maps
- nature topology maps
- different resolutions
And maps can also be more or less precise.
With todays satellite based maps, we can get a very precise model / picture
of the real location.
Sometimes downto to less than 10 meter resolution.
talking of using virtual tools:
John, I am not actually conversating you.
This post is just a virtual conversation/communication I have with you and others.
Using a model of real speech.
Should not be confused with the real thing
... when you can do a listening test of my voice
... and see my facial expressions, while speaking with you.
Sure
there is a difference.Isn't it?
Gee, fellas, I was just joking! I KNOW that most models are not perfect, and that is why when Bob Cordell demanded a SPICE model of my JC-1 power amp, (long ago) I just about fell over laughing.
I know a little something about models, as I used to have to make the transistors models myself with a curve tracer back in 1966, a long time ago. We had problems then, Dr Pederson (the father of SPICE) had problems in 1973, when I was taking his class, and we still have problems (some) now. I don't expect perfection, but I do not rely on imperfection to optimize my circuits in virtual space, either. It is just easier for ME just to try the circuit in real space.
I know a little something about models, as I used to have to make the transistors models myself with a curve tracer back in 1966, a long time ago. We had problems then, Dr Pederson (the father of SPICE) had problems in 1973, when I was taking his class, and we still have problems (some) now. I don't expect perfection, but I do not rely on imperfection to optimize my circuits in virtual space, either. It is just easier for ME just to try the circuit in real space.
john curl said:
Thanks for clearing that up, John. For a minute there, I thought Grey might have broken into your account
.But seriously, most SPICE op-amp macro models are really awful. They do not model distortion at all, though they do exhibit some distortion because the input stage is usually modeled with discrete devices. We discussed that in the Blowtorch thread.
The discrete device models, at least for BJTs, are not so bad, and as Scott has mentioned, can be quite good when professional care is taken to extract the parameters. JFET models are pretty primitive, as there is currently no economic incentive for the development of better ones as there is with MOS devices. If you're interested in learning more about the equations that modern SPICE simulators use for BJTs, I highly recommend Massobrio and Antognetti. It's a great book. In fact, given that you're a "bookworm" kind of guy, I'm surprised that you haven't dug in and pointed out the errors that are often made. It's another world with many new horizons to discover.
We, old geezers, don't need to know about spice models to that extent, but thanks for the tip. I'm sure PMA will be interested, and several others. Since I design almost exclusively with Japanese Jfets, mosfets, and only sometimes use transistors, modeling transistors with Spice is not my first interest. Getting microcap to work easily, for me, is my best priority, and I will leave the models up to you engineers.
Convergence
Hello andy_c
Just out of curiosity have you ever tried different simulators which have claims of better convergence and compared them too Ltspice using same circuit and models , other packages which I refer to are Spectre (cadence) and Simetrix. Dont get me wrong as freeware LTspice is very hard to beat its serious software.
Regards
Arthur
Hello andy_c
Just out of curiosity have you ever tried different simulators which have claims of better convergence and compared them too Ltspice using same circuit and models , other packages which I refer to are Spectre (cadence) and Simetrix. Dont get me wrong as freeware LTspice is very hard to beat its serious software.
Regards
Arthur