I made some Vbe plots using an EG of 1.15 for both devices. These look correct in shape, but are off a bit still. It could be a function of device variations? These should be comparable in format to the data sheet. The current polarity is upside down for the PNP transistor. I couldn't figure out how to get the curves to go left to right and bottom to top with negative values.
Last edited:
You're right, EG should be 1.15 for the KSC3503 as well. Don't worry about the simulated curves not being exactly the same as the ones in the datasheet: on the one hand, curves shown on a datasheet are either an average of a number of samples or correspond to a single sample that the manufacturer took as representative, so there will always be variation between an actual sample and the datasheet, and on the other hand, Spice models, complex as they are, are only a mathematical approximation of a much more complex physical device, so getting as close as possible to the measured and/or datasheet curves, even if you can't get a perfect match, is the best you can do.
Here's YAIQ (Yet Another Ignorant Question).
I see that Nexperia, for one, has some more recent SPICE models available for some devices, like the MPSA42/92 series in various packages. The issue date is just last year, although they reference an extraction date of 2019. They even have some fix, of sorts, for quasi saturation characteristics.
Are any of these models to be trusted? I don't have a lot of faith in some faceless corporation's data, which is odd, since I do have faith in some faceless diyAudio member's hand tested measurements (yours 🙂). Can I at least trust the DC parameters of these from Nexperia, such as EG? Many current data sheets don't even show performance curves - the manufacturer provides a SPICE model instead.
I know the ideal solution is to hand measure the devices myself (and share the results). But, by that point it's sort of pointless to even be using a SPICE like simulator if I am going to have to essentially build the device into a fixture to extract the information. May as well just take a guess and build the amplifier circuit and measure the performance of that. Simulators are supposed to make life a bit easier by at least giving the user a general idea of how a device will in a circuit. They won't give exact results due to the reasons we all know, but it'd be nice to have some faith in the rough results.
I see that Nexperia, for one, has some more recent SPICE models available for some devices, like the MPSA42/92 series in various packages. The issue date is just last year, although they reference an extraction date of 2019. They even have some fix, of sorts, for quasi saturation characteristics.
Are any of these models to be trusted? I don't have a lot of faith in some faceless corporation's data, which is odd, since I do have faith in some faceless diyAudio member's hand tested measurements (yours 🙂). Can I at least trust the DC parameters of these from Nexperia, such as EG? Many current data sheets don't even show performance curves - the manufacturer provides a SPICE model instead.
I know the ideal solution is to hand measure the devices myself (and share the results). But, by that point it's sort of pointless to even be using a SPICE like simulator if I am going to have to essentially build the device into a fixture to extract the information. May as well just take a guess and build the amplifier circuit and measure the performance of that. Simulators are supposed to make life a bit easier by at least giving the user a general idea of how a device will in a circuit. They won't give exact results due to the reasons we all know, but it'd be nice to have some faith in the rough results.
Could you please share those models? I can't find them in Nexperia's page and I'd be interested to see a manufacturer's model that includes quasi-saturation, I don't think I've ever seen one, maybe they've actually taken the time to generate a good model, which is rarely the case.
A while back I was in the same situation as you are now. Having concluded that, generally speaking, no, you can't trust manufacturers models (there are some outrageous ones, like the Toshibas mentioned above), I decided to get the Peak Atlas DCA75 to make my own models, in the process learning a lot about how these models work. If only for that reason, it was worth the investment. As to why manufacturer's models are so bad, I cannot fathom: you would think a major semiconductor manufacturer should have the resources to hire or train someone to produce decent models, but for some reason they don't. Go figure...
A while back I was in the same situation as you are now. Having concluded that, generally speaking, no, you can't trust manufacturers models (there are some outrageous ones, like the Toshibas mentioned above), I decided to get the Peak Atlas DCA75 to make my own models, in the process learning a lot about how these models work. If only for that reason, it was worth the investment. As to why manufacturer's models are so bad, I cannot fathom: you would think a major semiconductor manufacturer should have the resources to hire or train someone to produce decent models, but for some reason they don't. Go figure...
I also bought a DCA75, just to verify the authenticity of transistors and to help match devices. Perhaps I should do more. I haven't yet dived to the depths that you have, though - I just want to build a preamp or two and maybe an amplifier. Maybe I need to go further. 😕
Here's an example or two of models that claim to model quasi-saturation. They're surface mount equivalents of MPSA42 and MPSa92.
Nexperia PZTA42
Nexperia PZTA92
Here's an example or two of models that claim to model quasi-saturation. They're surface mount equivalents of MPSA42 and MPSa92.
Nexperia PZTA42
Nexperia PZTA92
I've just looked at the PZTA42, it doesn't look good. These are the forward and reverse curves with my model, simulated vs. measured:
And this is what I get with Nexperia's model, same set of forward and reverse Ib values, same X and Y axis scale:
Actually the reverse one doesn't look so bad, but the forward doesn't even show quasi-saturation and the Early effect is way too much. I don't know why they would try to model these effects by using a sub-circuit with additional components (the standard Spice NPN model already has parameters to model all that), but it looks like they didn't quite succeed...
And this is what I get with Nexperia's model, same set of forward and reverse Ib values, same X and Y axis scale:
Actually the reverse one doesn't look so bad, but the forward doesn't even show quasi-saturation and the Early effect is way too much. I don't know why they would try to model these effects by using a sub-circuit with additional components (the standard Spice NPN model already has parameters to model all that), but it looks like they didn't quite succeed...
I kind of thought the same thing myself. But, I wanted to be sure since I am NO expert.
I wonder how and why they get this so wrong. You'd think that they could get better results with any test gear they might buy - the DCA75 is great for hobbyists, but I am sure something better is available for professionals. Maybe they just don't want to make the effort. Or, their products really aren't that good.
Thanks very much!
I wonder how and why they get this so wrong. You'd think that they could get better results with any test gear they might buy - the DCA75 is great for hobbyists, but I am sure something better is available for professionals. Maybe they just don't want to make the effort. Or, their products really aren't that good.
Thanks very much!
It is not far from the datasheet. I feel the ksc3503/ksa1381 are pretty much useless. The gain drops off the cliff after 50ma, which is too low current for a driver. On the other hand, the early effect is too strong to be a good VAS. Thus, I say it is pretty useless.
CG,
RBM = 460 is bad! RBM = 460m.I will express my opinion about the manufacturers' models. They don't take model building seriously. They assign models to unqualified people who do not know what quasi-saturation is! As a result, many models have greatly underestimated VAF values. In particular, there are many such models in Multisim. The funny thing is that this model has a 2N2222 transistor (VAF=10), which is often used by students. And as it turned out, their teachers are the same "literates".