Discrete Opamp Open Design

Disabled Account
Joined 2012
So, I go buy this MicroCap sim software for thousands and cant trust results for anything except designs which do not use transistors! Geeez. [That sorry state can only help speed up the demise of analog designing.] I can use a few accurate models and use them as 'place holders' I suppose. Fortunately, thats often good enough for low freq designs. Thx. The food is surprisingly good.... the sea food is kept alive until cooked so it cant get any better and is safe to eat (!). There's nothing like seeing fish lying in the sun waiting to be bought to give you the shivers. You want a leg of lamb to take home? -- they kill the lamb or goat and hack off the leg and hand it to you -- hoof, hair, flees and all. But its fresh. Time to go exploring island caves.
 
Last edited:
There are bits of software around that help the modelling process; the LTspice users' group is very active in exploring the finer points of parts behaviour, they have a number of chunks of software that allow one to try and improve the models -- especially in the area of MOSFETs.

Also, intusoft has a package called SpiceMod which apparently is reasonable enough at creating models.

It just basically needs someone to sit and play around with a few tools long enough to get the quality in the models ...

Frank
 
I have used MicroCap SPICE for years, and I can tell you that it can be used with trannies, too!

Using any SPICE you need to have a gut feeling what seems right and what is not. Making PCBs with your SPICEd design, eventually gives you a feeling of what can be trusted and what not.
THD sims can be taken with a grain of salt and a factor 10X worse can be seen sometimes. But sometimes, THD is scaringly accurate.

Regarding models, you will get a feeling for what can be trusted or not. I have found that noise sims are very accurate once you get the models right (very easy).



/S


So, I go buy this MicroCap sim software for thousands and cant trust results for anything except designs which do not use transistors! Geeez. [That sorry state can only help speed up the demise of analog designing.] I can use a few accurate models and use them as 'place holders' I suppose. Fortunately, thats often good enough for low freq designs. Thx. The food is surprisingly good.... the sea food is kept alive until cooked so it cant get any better and is safe to eat (!). There's nothing like seeing fish lying in the sun waiting to be bought to give you the shivers. You want a leg of lamb to take home? -- they kill the lamb or goat and hack off the leg and hand it to you -- hoof, hair, flees and all. But its fresh. Time to go exploring island caves.
 
Disabled Account
Joined 2012
With simple circuits like audio amps, it is just as fast to measure a breadboard. I use a handheld calculator, curve tracer etc and knowledge of what I want to accomplish and build and test. Just as you get a 'feel' for what is needed in sim software, I get it from years of test and measurements and breadboarding. The lack of good trannie data and all makes it faster to just breadboard it. I can put together a circuit about as fast as I can draw the schematic in MicroCap. I mean, couldnt anyone make an IC opamp audio circuit work well without doing a sim first? Ditto for discrete audio circuits.
I live in hope -so I keep upgrading/updating MicroCap.
 
Last edited:
Who, for examples? Whose could I trust?

I don't do enough with external models to tell you, for instance from the data sheet curves some of those high voltage small signal devices show quite a bit of quasi-saturation and the supplied models work in SPICE but choke in my simulator (in a very non-physical way). We invest millions in staff and equipment (software too) to support our models.

As for JFET's forget it if your simulator does not support Level 3 models, and even if it does you will have to extract the parameters yourself.
 
I mean, couldnt anyone make an IC opamp audio circuit work well without doing a sim first? Ditto for discrete audio circuits.
I live in hope -so I keep upgrading/updating MicroCap.

Essentially all the industry standard circuits were done that way until the 80's. We built discrete breadboards, and frequently had to revise the chip. You can't afford to work that way now, frequently parts from the very first wafer go right to the customer.
 
Disabled Account
Joined 2012
Right or lucky?

I don't do enough with external models to tell you, for instance from the data sheet curves some of those high voltage small signal devices show quite a bit of quasi-saturation and the supplied models work in SPICE but choke in my simulator (in a very non-physical way). We invest millions in staff and equipment (software too) to support our models.

As for JFET's forget it if your simulator does not support Level 3 models, and even if it does you will have to extract the parameters yourself.

So, that's why I spend most of my time with test equipment measuring samples from everywhere until I find something I can use for my appl. I must have spent $800 in samples to make what others will build for a 20 dollar headphone amp cost. But, hey! It works fine and didnt need a sim. Actually, I tried the sim on it... it gave such ridiculously low harmonics that I didnt trust it. But it was right or lucky. -RNM
 
I've found that at least one key thing is include all relevant parasitics: if you assume that a drawn ground is a real ground, and that voltage supplies are perfect, then you're likely to have problems, for starters. The point is to model what will actually be there in the real world, not skate on top of a set of assumptions ...

Frank
 
So, I go buy this MicroCap sim software for thousands and cant trust results for anything except designs which do not use transistors!

I used to use Ansoft's Serenade for RF design, including that of SAW and other oscillators. IIRC the transistors were characterized in terms of their S parameters using a network analyzer. This isn't necessarily helpful other than that it tends to indicate that useful simulations including active devices are possible. I haven't had any call to do any of that stuff for years tho'.
 
diyAudio Member RIP
Joined 2005
Scott's remarks about FET models are most apt. My simulator accepts the same parameters as others, but what it winds up doing with them isn't pretty. It makes an abrupt transition from the so-called triode region to an output conductance region, the latter which is modeled as a constant slope!

At some point I may concoct a model using the available math functions which will be cumbersome, but work adequately at low frequencies.
 
Disabled Account
Joined 2012
Essentially all the industry standard circuits were done that way until the 80's. We built discrete breadboards, and frequently had to revise the chip. You can't afford to work that way now, frequently parts from the very first wafer go right to the customer.

For those not in the business, breadboarding is still practical for once in awhile projects. How about we identify the parameters that are most important to audio - those which affect noise, distortion, gain linearity etc and describe how to extract what is needed for the sim model that is typically missing? A short tutorial for the most important - missing-parameters. For bipolar and for jFET. Low freq audio isnt all that complex that DIY'er need extreame modelling accuracy or precision. -Thx
 
Disabled Account
Joined 2012
BTW -- modelling has replaced many cut-n-try approaches to designing; One is nuclear weapons design. Each time a live 'event' (as it was called) was done to test the design, the data gathered via piggy-backed detectors onto the 'device'. would help refine the model. Since, the under ground nuclear testing was stopped, design can continue due to the models. Its is Still a dangerous world.... maybe more so due to the speed of highly developed modelling.
 
Scott's remarks about FET models are most apt. My simulator accepts the same parameters as others, but what it winds up doing with them isn't pretty. It makes an abrupt transition from the so-called triode region to an output conductance region, the latter which is modeled as a constant slope!

At some point I may concoct a model using the available math functions which will be cumbersome, but work adequately at low frequencies.

That's the Level 1 model basicly circa 1952, the discontinuous derivative sucks. You want to search for Curtice and his refined JFET models.


EDIT - W. R. Curtice, "A MESFET Model for Use in Design of GaAs Integrated Circuits," IEEE Trans. Microwave Theory Tech., vol. MTT-28, pp. 448-456, May 1980.
C. T. M. Chang, et al., "A Subthreshold Current Model for GaAs MESFET's," IEEE Electron Device Letters, vol. EDL-8, No. 2, Feb. 1987.
 
Last edited:
Scott's remarks about FET models are most apt. My simulator accepts the same parameters as others, but what it winds up doing with them isn't pretty. It makes an abrupt transition from the so-called triode region to an output conductance region, the latter which is modeled as a constant slope!

At some point I may concoct a model using the available math functions which will be cumbersome, but work adequately at low frequencies.
My understanding is that the transconductance is proportional to the gate voltage in the saturation region, or am I misunderstanding you, or are you talking about more subtle aspects of behaviour?

Thanks,
Frank
 
diyAudio Member RIP
Joined 2005
My understanding is that the transconductance is proportional to the gate voltage in the saturation region, or am I misunderstanding you, or are you talking about more subtle aspects of behaviour?

Thanks,
Frank
Problem is the output conductance (partial derivative of drain current with respect to drain voltage). It's a continuous function with continuous derivatives (or equal-to-zero derivatives), and the dumb model says you have a triode region with a certain curvature, then suddenly you have a linear function beyond a certain drain voltage (the so-called saturation region). Not the way it works, and an especially lousy approximation to short-channel devices like the lately-favored BF862, and the old mainstay the 2SK170.
 
or am I misunderstanding you, or are you talking about more subtle aspects of behaviour?

Thanks,
Frank

Yes, these low noise JFET's are short channel devices and the behavior with Vds is not linear but fits better to a hyperbolic tangent function. This is a fine point that lies on top of the basic behavior that you mention. The distortion characteristics due to Vds are all wrong.